SlideShare a Scribd company logo
1 of 20
Download to read offline
89 Fifth Avenue, 7th Floor
  New York, NY 10003
  www.TheEdison.com
       212.367.7400




                             White Paper



                             Virtualization Performance on the IBM
                             PureFlex System
Printed in the United States of America
Copyright  2012 Edison Group, Inc. New York. Edison Group offers no warranty either
expressed or implied on the information contained herein and shall be held harmless for errors
resulting from its use.

All products are trademarks of their respective owners.
First Publication: April 2012
Produced by: Craig Norris, Sr Analyst; Barry Cohen, Editor-in-Chief; Manny Frishberg, Editor

This document was developed with IBM funding. Although the document may utilize publicly
available material from various vendors, including IBM, it does not necessarily reflect the
positions of such vendors on the issues addressed in this document.
Table of Contents

Executive Summary ..................................................................................................................... 1
Introduction .................................................................................................................................. 3
      Objective .................................................................................................................................. 3
      Audience .................................................................................................................................. 3
      Contents of this Report .......................................................................................................... 3
The Business Value of Virtualization ...................................................................................... 4
IBM's Unified Virtualization Infrastructure: PureFlex ........................................................ 5
TPoX: Benchmark Workload for Modern Virtualized DBMS-Based Environments...... 7
Benchmark Comparison Study and Results ........................................................................... 8
      Test System .............................................................................................................................. 8
      Test #1: Virtualization Tests and Results............................................................................. 9
           Methodology .................................................................................................................... 9
           Results.............................................................................................................................. 11
      Test #2: Storage ..................................................................................................................... 13
Conclusions ................................................................................................................................. 15
Executive Summary

Using virtualization to consolidate data center servers has become an integral
component of the way successful companies design their IT infrastructures. Most,
however, use consolidation ratios of only around six virtual machines (VMs) per
physical server. Even world-class organizations typically are only consolidating at a
ratio of about 18 to 1 at best.1 Yet many commonly published virtualization benchmarks
test with hundreds of VMs per physical server, which the average data center
administrator has difficulty relating to their actual environment. This white paper is
meant to appeal to anyone seeking benchmark evaluations of consolidation solutions
that look more like what they themselves are doing or considering for their
organizations.

The tests described in this paper applied Transaction Processing over XML (TPoX)
benchmark workloads to the IBM PureFlex system: a synergistic solution capable of
simultaneously running and managing the best of heterogeneous IT environments —
Linux, AIX, Windows, and IBM i operating systems, POWER7- and x86-based integrated
technology elements (ITEs), PowerVM and VMware vSphere virtualization, HDD and
SSD storage — all in a general-purpose system having all the simplicity of an appliance.

TPoX is a much more suitable benchmark for assessing transaction processing
throughput for today’s complex, virtualized IT environments than traditional industry
standard benchmarks. In this paper, benchmarks are used to examine the throughput
performance for various system configurations the PureFlex system can accommodate,
as well as the ability of the solution to scale without degrading that performance. All
tests were performed on the same self-contained PureFlex system unit.

Key findings include the following:
   IBM PowerVM on the IBM POWER7 ITE can deliver over 78 percent more
    throughput at 80 virtual machines (VMs) than vSphere 5 on an Intel Xeon E5-2690
    (Sandy Bridge2)-based ITE.
   AIX and Linux deliver near equivalent performance in leveraging this PowerVM
    capacity.




1Aberdeen Group report, Best-in-Class Practices for Virtualizing Microsoft Applications, August 2010
2Sandy Bridge is the codename for both the Intel microarchitecture innovation following Nehalem, and
generally for the associated family of 32nm processors based upon that microarchitecture.


Edison: IBM – TPoX Benchmark Results on IBM PureFlex System White Paper                          Page 1
   On either Intel-based or POWER7-based platforms, the PureFlex system exploits
    IBM Storwize V7000 to deliver up to 10 times the number of VMS per SSD than is
    possible using HDDs.

This paper makes clear that the IBM PureFlex system is a superior solution for running
and managing mixed IT environments in a unified solution. It also show that, for those
interested in getting serious about virtualized consolidation, PowerVM virtualization
technology on POWER7 processor-based platforms offers greater performance than that
offered by VMware vSphere 5 on Intel x86 platforms. It enables high consolidation ratios
and increased flexibility for a far superior virtualization solution.




Edison: IBM – TPoX Benchmark Results on IBM PureFlex System White Paper           Page 2
Introduction


Objective

The objective of this white paper is to demonstrate the performance and capabilities of
IBM’s PureFlex system, presenting results that showcase throughput and virtualization
capacity for each of the multiple operating systems and ITEs comprised by this
advanced modular server system, as well as its storage subsystem and its next-
generation networking system. It describes tests using the industry-standard TPoX
benchmark to compare virtualization technologies. The results were reviewed, analyzed,
and presented by Edison Group.


Audience

This paper is intended for anyone interested in the advantages of server consolidation
through virtualization. The testing described herein employed a means of assessing
virtualization efficiency at a 5 VM-per-core ratio, which more closely reflects typical data
center environments than the hundreds of VMs per server tested in other commonly
published virtualization benchmarks.

CTOs, CEOs, IT managers, and others will find valuable information here that could
help them further enhance and/or adopt virtualization technology within their IT
environments.


Contents of this Report

This white paper contains the following major sections:
   The Business Value of Virtualization — This section discusses the business value
    propositions underlying the benchmark evaluations presented in this paper.
   IBM’s Unified Virtualization Infrastructure: PureFlex — This section describes
    IBM’s PureFlex system offering on which the benchmarks were conducted.
   TPoX — This section describes the TPoX benchmark and the reason for using it to
    evaluate transaction throughput in modern virtualized environments.
   Benchmark Comparison Study & Results — This section presents the comparative
    testing, describing the test equipment setup, the benchmarks, the actual tests, and
    the results of the tests.




Edison: IBM – TPoX Benchmark Results on IBM PureFlex System White Paper              Page 3
The Business Value of Virtualization

As the foundation platform for today’s data center, server virtualization is quickly
reaching maturity. Virtual machines now host more than half of business server
workloads.3 Virtualization has become the default build for new server installations,
reducing costs and establishing the foundation for more efficient and flexible
configurations and technology platforms. 4 The performance of virtualization is critical
to realizing success of server pools and cloud computing (and is also a key component in
IBM’s roadmap in its Smarter Computing initiative).

Well-implemented virtualization solutions may be employed to:
   Reduce hardware expenditures by consolidating multiple environments, including
    underutilized servers, and systems with varied and dynamic resource requirements.
   Reduce costs for power and cooling, floor space, hardware maintenance, and
    software licensing.
   Grow and shrink resources dynamically according to business needs.
   Deploy new workloads through provisioning VMs on new systems rapidly to meet
    changing business demands.
   Develop and test applications in secure, independent domains while allocating
    production to its own domain on the same system.
   Transfer live workloads to support server migrations, balance system load, or avoid
    planned downtime that can otherwise adversely impact productivity.
   Control server sprawl, reducing system management costs.

Despite these benefits, the majority of businesses fall far short of seizing upon the full
potential of server consolidation. Their average consolidation ratios hover around six
VMs per server,5 yet economic advantages from data center consolidation increase
significantly at much higher VM densities. By increasing the consolidation ratio per
system, businesses can reduce capital expenditures and operational costs by reducing
the number of systems in their data center or IT organization.
IBM’s Smarter Computing systems, which allow for greater VM density without
degrading system performance, can deliver considerable economic advantages to
organizations using them. This study examines the performance and scaling aspects of
the IBM PureFlex system across the commonly employed industry benchmark TPoX.

3 The Value of Memory-Dense Servers: IBM’s System x MAX5 for its eX5 Server Family, March 2010, IDC
4 Ibid.
5 According to an Aberdeen Group report, Best-in-Class Practices for Virtualizing Microsoft Applications,

August 2010, even the best-in-class organizations in the study consolidate at only an 18:1 ratio.


Edison: IBM – TPoX Benchmark Results on IBM PureFlex System White Paper                                     Page 4
IBM’S Unified Virtualization Infrastructure:
PureFlex

With the PureFlex system, IBM
brings together advanced
modular server and storage
technology in a virtualization
infrastructure that leverages
superior unified management
and integrated Ethernet and
fibre channel networking
technology.

The PureFlex system brings
IBM’s leadership in computing
systems and extensive expertise
in systems integration to a
synergistic solution capable of
simultaneously running and managing the best of heterogeneous IT environments —
Linux, AIX, Windows, and IBM i operating systems, Power Systems and x86-based ITEs,
PowerVM and various x86 virtualization platforms, HDD and SSD storage — in a
general-purpose system having all the simplicity of an appliance.

Based on solidly established patterns of expertise, the PureFlex system is designed to get
data center operations up and running in as little as four hours, cutting months off the
deployment time for new application projects. The major subsystems integrated into the
PureFlex system are:

   POWER7 Processor-Based Power Systems ITEs — Automatically optimizing
    performance and capacity at either a system or VM level, these ITEs benefit from the
    POWER7 processor, which contains innovative technologies that help maximize
    performance and optimize energy efficiency. They represent the most flexible and
    cost-efficient solutions for UNIX, IBM i, and Linux deployments available in the
    market.
   IBM x86-Based ITEs — These ITEs support a wide selection of processor
    technologies and operating systems. They are designed to reduce complexity,
    improve systems management, and increase energy efficiency while driving down
    total cost of ownership.




Edison: IBM – TPoX Benchmark Results on IBM PureFlex System White Paper             Page 5
   IBM Storwize V7000 — This is a virtualized storage controller designed to
    consolidate block and file workloads into a single storage system for simplicity of
    management, reduced cost, highly scalable capacity, performance, and high
    availability. It offers improved efficiency and flexibility through built-in solid state
    drive (SSD) optimization, thin provisioning, and non-disruptive migration of data
    from existing storage.
   Next-Generation Networking — The PureFlex system incorporates a platform
    networking and storage fabric technology, utilizing integrated 16 Gb fibre channel
    switches and 10 Gb Ethernet switches to enable heterogeneous ITEs to run
    concurrently with little or no impact on each other, tying together these various
    subsystems with Ethernet and/or FC into a single manageable system.




Edison: IBM – TPoX Benchmark Results on IBM PureFlex System White Paper                 Page 6
TPoX: A Benchmark Workload for Modern
Virtualized DBMS-Based Environments

One of the goals of this study was to more closely approximate typical virtualized data
center usage by organizations. This includes the type of application workloads to be
used in the testing. Cloud computing, database-as-a-service, and virtualization are
changing the way customers deploy databases, and a benchmark workload used to test
the transaction processing performance of DBMS-based applications should be able to
address these architectures, and illustrate to data center operators what they can expect
as they increasingly leverage virtualization.

TPoX (Transaction Processing over XML) is an application-level “XML database”
benchmark. It is an XML OLTP benchmark using data-oriented XML structures, very
large numbers of relatively small XML documents (1 kb to 20 kb), short read/write
transactions, and a high degree of concurrency. It essentially models a security-trading
scenario that uses a real-world XML Schema (FIXML).

TPoX can be used to evaluate any database that offers XML support. The TPoX
workload driver is architected such that only a thin layer (a single Java class) deals with
the specific interaction to the database system under test. The combination of these XML
and Java components makes the workload relevant for a wide range of web and
collaboration applications.

Applications in the real world typically have many concurrent users, a mix of read and
write operations, and millions or even billions of XML documents. This is exactly what
the TPoX benchmark is designed to capture. It simulates an actual application that
performs queries, inserts, updates, and deletes in a concurrent multi-user workload
using a real-world database. In the study to follow, the enterprise-class DB2 database
was used on all three platforms tested.

As a database-centric application, TPoX stresses CPU, memory, and storage I/O. It is
capable of driving high CPU utilization to study true technology capability. In addition,
in a multi-VM environment such as that tested for this study, it stresses the
virtualization infrastructure supporting these resources on the tested platforms.

TPoX is an open-source benchmark jointly developed by IBM, Intel, and others. The
benchmark, along with documentation and published results, is publicly available at:
http://tpox.sourceforge.net 6

6   Reference: http://nativexmldatabase.com/2011/03/04/new-tpox-benchmark-results-available/


Edison: IBM – TPoX Benchmark Results on IBM PureFlex System White Paper                        Page 7
Benchmark Comparison Study and Results

This section of the paper describes the system configuration used in the tests,
methodology employed, and results for each.


Test System

The test system consists of a single PureFlex system provisioned with the following
major subsystems:
   One IBM Power p260 node with two 3.56 GHz Power 7 processors
   One 2-socket ITE with 2.9 GHz Intel Xeon E5-2690 (Sandy Bridge) processors
   Two Storwize V7000 storage controllers, each provisioned with 24 300 GB SSDs
   Networking fabric comprising two 16 Gb fiber channel switches integrated into the
    PureFlex chassis

Figure 1 (below) shows the positioning of the subsystems within the overall PureFlex
system unit.




                             Figure 1. Diagram of Test System


Edison: IBM – TPoX Benchmark Results on IBM PureFlex System White Paper           Page 8
Table 1 (below) presents how the PureFlex system was configured for the TPoX
performance benchmarks.

              NOTE: Not every component was necessarily utilized for every test.



                                 IBM Flex System P260           Sandy Bridge 2-Socket
                                    Compute Node                   Compute Node
         Processors                2 x IBM POWER7                2 x Intel Xeon E5-2690
     Guest Operating                     AIX 7.1
       System(s)                                                SUSE Linux Enterprise
                                 SUSE Linux Enterprise
                                                                   Server 11 SP2
                                    Server 11 SP2
       Virtualization          IBM Power VM Enterprise             VMware vSphere 5
                                     Edition v2.2
    Database Backend                                 IBM DB2 v9.7
         Workload                                      TPoX v2.1
          Storage                              IBM Storwize V7000 / SSD

         Table 1: Configuration of the PureFlex System for TPoX Benchmarks




Test # 1: Virtualization Tests and Results

The goal of these tests was to determine aggregate transaction throughput performance
and VM scalability.


Methodology
These tests employed a VM-to-core ratio of 5:1. The system was scaled up from a single
core to two full sockets. In this way, tests were run on sets of 5, 10, 20, 40, and 80 VMs,
each running TPoX, as shown in Figure 2, below.




Edison: IBM – TPoX Benchmark Results on IBM PureFlex System White Paper               Page 9
TPoX Virtualization Tests
                                                                          2 socket ITE




                                                            1 socket




         1 core




       5 VMs            10 VMs          20 VMs           40 VMs           80 VMs

                     Figure 2: TPoX Virtualization Tests Methodology


The TPoX benchmark is I/O-intensive and its performance is partially dependent on
storage performance. Both ITE platforms used the same PureFlex system storage
subsystem although each used a different Storwize V7000 Controller to store its data and
logs.

Each VM had its own XML database servicing clients; a 1 GB database was used in order
to match up with each VM’s CPU (0.2 core) and memory capacity (3 GB). A single-tier
TPoX configuration was chosen for each VM where the client and the database reside in
the same VM. The goal was to run enough clients to ensure that each VM reached high
utilization. The database for each VM on each of the platforms was populated with the
same configuration set. The transaction rate for populating the database is shown in
Table 2.

                                      Flex System            Sandy Bridge / VMware
                                  p260/PowerVM/AIX                 vSphere 5
  Order (inserts per second)              1915                         1028
 Custacc (inserts per second)              769                         400
Table 2. TPoX Database Populated Rate for First Configuration Set



Edison: IBM – TPoX Benchmark Results on IBM PureFlex System White Paper                  Page 10
The VM configuration has multiple options on both PowerVM and VMware vSphere 5
technologies . These tests strove to utilize the full capacity of each core deployed. In the
case of PowerVM, five virtual CPUs (vCPUs) per core were sufficient to accomplish this.
In the case of vSphere 5, it was established in an earlier study 7 that two vCPUs per VM
was more efficient than one vCPU per VM. Thus, each core was configured with 10
vCPUs for the vSphere 5 subsystem.

For each test, the per VM transaction rate was measured, and the aggregate transaction
rate at the system level was reported.


Results
The results presented in Figure 3, below, indicate that the throughput performance rate
for populating the database are impressive for the Intel Xeon E5-2690 Sandy Bridge
compute node on the IBM PureFlex system. However, as may be expected, throughput
performance for Linux on POWER7/PowerVM technology averaged over 54 percent
better. A slightly greater advantage proved to be the case when the workload was run
on the AIX operating system. The POWER7/PowerVM advantage over Sandy
Bridge/VMware also tended to grow with the number of VMs, with POWER7/AIX
showing a nearly 76 percent advantage at 80 VMs, and POWER7/Linux showing a
similar advantage at 72 percent.

                                      PowerVM / Flex p260 vs. vSphere / Sandy Bridge

                              10000
                               9000
                               8000
    Transactions per Second




                               7000
                               6000
                               5000
                               4000
                               3000
                               2000
                               1000
                                  0
                                        5VMs          10VMs         20VMs         40VMs         80VMs
                                      Figure 3: TPoX Benchmark Results for Throughput Performance


7IBM PowerVM Virtualization Technology on IBM POWER7 Systems - A Comparison of PowerVM and VMware
Virtualization Performance, Edison Group, Inc., 2012.


Edison: IBM – TPoX Benchmark Results on IBM PureFlex System White Paper                                 Page 11
Table 3, below, presents detailed information on the total number of TPoX users used in
each test, throughput, and VM configuration for each tests.

                                                         Total
  System Configuration for TPoX Benchmark               Virtual Total # of Transactions
            (1 to 80 VM Scaling)          # of VMs      CPUs TPoX Users per second
 IBM Flex p260 3.56 GHz DPSM mode, 2
 socket, 256 GB RAM, SMT4 enabled,
 PowerVM, SUSE Linux Enterprise Server 11
 SP2 is the host OS for each VM. Each LPAR is      5      5         50         600
 configured with 1 vCPU/ uncapped/3 GB
 RAM 4 LPARs have 0.2 cores and 1 LPAR
 have 0.1 core. Shared pool has one core
 9 LPARs are configured each with 0.2/1 vCPU/
 uncapped/ 3 GB memory, 1 LPAR is
                                                   10     10        100        1112
 configured with 0.1/1 vCPU/uncapped/3 GB
 memory. Shared pool has two cores
 18 LPARs are configured each with 0.2/1
 vCPU/ uncapped/ 3 GB memory, 2 LPARs are
                                                   20     20        200        2316
 configured with 0.1/1 vCPU/uncapped/3 GB
 memory. Shared pool has four cores
 37 LPARs configured with 0.2core/
 1vCPU/uncapped and 3 LPARs configured
                                                   40     40        400        4710
 with 0.3 cores/1vCPU/ uncapped. Shared pool
 has eight cores
 74 LPARs are configured each with 0.2/1
 vCPU/ uncapped/ 3 GB memory, 4 LPARs are
                                                   80     80        800        9499
 configured with 0.1/1 vCPU/uncapped/ 3 GB
 memory. Shared pool has 16 cores
 2 socket 2.9 GHz, Intel Xeon E5-2690 compute
 node with 256 GB system RAM (HT and
 Turbo enabled in BIOS Intel VTx with EPT
 HW virtualization assist) VMware vSphere 5.
 Each VM has guest OS SUSE Linux Enterprise        5      10        50         428
 Server 11 SP2. Each VM is given 0.2 of a core/1
 2 vCPUs/3 GB memory. DB2 buffer pool for
 data is configured in each VM. Schedule
 affinity is set to cpu0 and cpu1.
 Schedule affinity is set to cpu0 to cpu3          10     20        100        785
 Schedule affinity is set to cpu0 to cpu7          20     40        200        1526
 Schedule affinity is set to cpu0 to cpu15         40     80        400        2839
 Schedule affinity is set to cpu0 to cpu31         80    160        800        5516


Edison: IBM – TPoX Benchmark Results on IBM PureFlex System White Paper          Page 12
Total
  System Configuration for TPoX Benchmark               Virtual Total # of Transactions
            (1 to 80 VM Scaling)          # of VMs      CPUs TPoX Users per second
 IBM Flex p260 3.56 GHz DPSM mode, 2
 socket, 256 GB RAM, SMT4 enabled,
 PowerVM, IBM AIX .1 is the host OS for each
 VM. VIOS is configured with 0.1 core/1 vCPU/
                                                 5         5         50          629
 uncapped mode/ 4 GB RAM. Each LPAR is
 configured with 1 vCPU/ uncapped/3 GB
 RAM 4 LPARs have 0.2 cores and 1 LPARs
 have 0.1 core Shared pool has one core
 9 LPARs are configured each with 0.2/1 vCPU/
 uncapped/ 3 GB memory, 1 LPAR is
                                                10        10        100         1307
 configured with 0.1/1 vCPU/uncapped/3 GB
 memory. Shared pool has two cores
 18 LPARs are configured each with 0.2/1
 vCPU/ uncapped/ 3 GB memory, 2 LPARs are
                                                20        20        200         2642
 configured with 0.1/1 vCPU/uncapped/3 GB
 memory. Shared pool has four cores
 37LPARs configured with 0.2core/
 1vCPU/uncapped and 3 LPARs configured
                                                40        40        400         5063
 with 0.3core/1vCPU/ uncapped. Shared pool
 has eight cores
 LPARs are configured each with 0.2/1 vCPU/
 uncapped/ 3 GB memory, 4 LPARs are
                                                80        80        800         9703
 configured with 0.1/1 vCPU/uncapped/ 3 GB
 memory. Shared pool has 16 cores

Table 3. TPoX Benchmark Detailed Test Information




Test # 2: Storage

The Storwize V7000 storage controller in the PureFlex system provides the flexibility to
deploy any mix of both hard drive disks (HDDs) and Solid State Drives (SSDs). Since
OLTP applications (such as TPoX) particularly benefit from high-throughput storage,
the goal of this setup was to determine the effect that provisioning the Storwize V7000
controller with SSDs would have on the VM density capability of the PureFlex system.

Figure 4, below, presents a visual representation comparing maximum virtualization
possible using the TPoX benchmarks for the Storwize V7000 Controller provisioned with




Edison: IBM – TPoX Benchmark Results on IBM PureFlex System White Paper           Page 13
a full 24 HDDs (for an earlier study 8), and the same controller provisioned with 24 SSDs
for the present study. As the figure indicates, a tenfold increase in the possible number
of VMs per core results in provisioning SSDs rather than HDDs.




                  HDD                                                 SSD
    Storwize V7000 Controller with 24 HDDs            Storwize V7000 Controller with 24 SSDs


               Figure 4: VM Density Comparison between HDDs & SSDs



When the Storwize V7000 was provisioned with 24 HDDs, 12 HD spindles were
required for every four VMs in order to achieve sufficient parallelization to overcome
the latency (seek time) associated with HDDs. Unlike the mechanical HDDs, which are
limited by factors such as spindle speeds and actuator movement, SSDs involve no
moving parts. They consume a fraction of the power required to operate and to cool
HDDs, while taking up a fraction of the space.

SSDs can also yield I/O response times up to 100 times greater than HDDs, making them
particularly good choices for transaction-intensive workloads typical in database
applications. While once too costly for all but the highest-end performance-intensive
uses, SSDs have dropped in price considerably which — along with their savings in
space, power, and maintenance (due to much fewer devices needed and their non-
mechanical nature) — they can in certain scenarios actually be the more cost-effective
choice over HDDS.




8IBM PowerVM Virtualization Technology on IBM POWER7 Systems - A Comparison of PowerVM and VMware
Virtualization Performance, Edison Group, Inc., 2012.


Edison: IBM – TPoX Benchmark Results on IBM PureFlex System White Paper                    Page 14
Conclusions

This paper shows that the IBM PureFlex system can deliver VM consolidation in a
heterogeneous, self-contained environment capable of impressive levels of throughput
performance. It can dramatically reduce time to production for virtualized data center
application operations, providing multiple compute and operating system platforms,
advanced storage, and integrated networking in a single manageable system.

The benchmark results presented in this paper show that the IBM POWER7 processor-
based ITE platform using PowerVM virtualization, in particular, demonstrated
significantly superior throughput performance at all levels of VM population over the
x86 Sandy Bridge ITE platform using vSphere 5. IBM PowerVM technology — the
virtualization software built into the POWER7 processor-based systems — offers an
unprecedented level of platform support, scalability, efficient resource utilization,
flexibility, and heterogeneous server management. IBM PowerVM virtualization offers
autonomic resource affinity, resulting in higher workload performance in a virtualized
environment.

The IBM PureFlex system delivers efficient virtualization of IBM POWER7 Systems and
PowerVM technology, along with integrated IBM Storwize V7000 controllers that can
accommodate SSDs for maximum per-core VM density, making the IBM PureFlex
system an excellent foundation for cloud computing environments.




                                                                      POL03114-USEN-01
Edison: IBM – TPoX Benchmark Results on IBM PureFlex System White Paper          Page 15
Appendix A

This appendix describes the test methodology used to compare PowerVM to VMware
application benchmark performance across three (pLinux, AIX and xLinux) operating
systems on the IBM Unified Virtualization Infrastructure.

This performance evaluation characterizes hypervisor efficiency in scaling virtual
machines on POWER7- and x86-based integrated technology elements (ITEs ). The
experiments conducted assess the effect of adding VMs incrementally as well as
incrementally increasing system resource consumption. Throughput performance was
measured as the number of VMs was scaled from 5 to 80.

To ensure fair comparison across platforms and to remove variability across each set of
tests, the following actions were taken:
1. Deployment of optimized VM configurations in terms of virtual processors on each
   platform and similar memory allocated per VM.
    a. a. On PowerVM, resources can be allocated to an I/O hosting partition, therefore
       resource allocation to I/O hosting partition is optimized for each set of tests on
       Power systems. This is not applicable to VMware.
2. The same set of “benchmark parameters” were used across platforms.
3. Except for the virtualized workloads under investigation, a common software stack
   was maintained across the three operating systems (DB server).
4. Tuning was performed based on best practices for respective platforms.
5. VMware vSphere 5.0, SuSE 11 SP2 (xLinux and pLinux), AIX 7.1, DB2 tuning.


System Tuning

X86 (Sandy Bridge) Integrated Technology Element
   System (uEFI/BIOS) default settings were used (HT enabled, Turbo mode enabled,
    Operating Mode set to MAX performance).


IBM POWER7 Integrated Technology Element
   Dynamic Power Management / Favor Performance was enabled.



                                                                       POL03114-USEN-01
Edison: IBM – TPoX Benchmark Results on IBM PureFlex System White Paper            Page 16
Virtualization Tuning

VMware Virtual Machine Configuration Details
     VMware VMs were created using Virtual Machine version 8.
     Virtual Disk LSI Logic Parallel was used. It was noted that the LSI Logic Parallel
      adapter and the LSI Logic SAS adapter offer equivalent performance. 9
     Updates were made to the latest VMware tools.
     esxtop –ab and vmstat were collected from the VM.


AIX and pLinux (PowerVM Guest OS) and xLinux (VMware Guest OS)
Tunings

AIX Tunings
          *raso -r -o mtrc_enabled=0    (disables lightweight memory trace)
          *ctctrl -P memtraceoff        (disables component trace)
          *errctrl -P errcheckoff       (disables run-time error checking)
          *skeyctl -k off               (disables storage keys)
           dscrctl -n -s 1              (disable prefetch)

     Remove the Java 5 version from AIX 7.1 install (the benchmark used Java 6).
     Disk queue depth increased to 20 on the VM (default 3 is low for this I/O-intensive
      workload).


Disclaimer: These AIX tunings were done for benchmarking reasons; customers do not change
any of these tunables in a production environment unless they are asked to by the AIX support
team in the process of resolving an issue.


pLinux Tunings
     ppc64_cpu –dscrctl=1              (disable prefetch)


DB2 Tuning on all three operating systems (xLinux, pLinux, AIX 7.1)
     Database buffer pool (0.2 GB) is allocated for table space on each VM.




9   vSphere Help
                                                                            POL03114-USEN-01
Edison: IBM – TPoX Benchmark Results on IBM PureFlex System White Paper                 Page 17

More Related Content

What's hot

Vm Ware X Xen Server
Vm Ware X Xen ServerVm Ware X Xen Server
Vm Ware X Xen ServerAndre Flor
 
X-Pod for VDI Reference Architecture Enabled by Cisco UCS, VMware Horizon Vie...
X-Pod for VDI Reference Architecture Enabled by Cisco UCS, VMware Horizon Vie...X-Pod for VDI Reference Architecture Enabled by Cisco UCS, VMware Horizon Vie...
X-Pod for VDI Reference Architecture Enabled by Cisco UCS, VMware Horizon Vie...X-IO Technologies
 
IBM pureflex system and vmware vcloud enterprise suite reference architecture
IBM pureflex system and vmware vcloud enterprise suite reference architectureIBM pureflex system and vmware vcloud enterprise suite reference architecture
IBM pureflex system and vmware vcloud enterprise suite reference architectureIBM India Smarter Computing
 
Ibm pure systems sales bootcamp
Ibm pure systems sales bootcampIbm pure systems sales bootcamp
Ibm pure systems sales bootcampsolarisyougood
 
IBM Flex System Reference Architecture for Microsoft SQL Server 2012 High Ava...
IBM Flex System Reference Architecture for Microsoft SQL Server 2012 High Ava...IBM Flex System Reference Architecture for Microsoft SQL Server 2012 High Ava...
IBM Flex System Reference Architecture for Microsoft SQL Server 2012 High Ava...IBM India Smarter Computing
 
Flex system client_presentation
Flex system client_presentationFlex system client_presentation
Flex system client_presentationNatalija Pavic
 
Grid rac preso 051007
Grid rac preso 051007Grid rac preso 051007
Grid rac preso 051007Sal Marcus
 
VDI performance comparison: Dell PowerEdge FX2 and FC430 servers with VMware ...
VDI performance comparison: Dell PowerEdge FX2 and FC430 servers with VMware ...VDI performance comparison: Dell PowerEdge FX2 and FC430 servers with VMware ...
VDI performance comparison: Dell PowerEdge FX2 and FC430 servers with VMware ...Principled Technologies
 
A Dell Latitude 5420 laptop powered by a four-core Intel Core i5-1145G7 vPro ...
A Dell Latitude 5420 laptop powered by a four-core Intel Core i5-1145G7 vPro ...A Dell Latitude 5420 laptop powered by a four-core Intel Core i5-1145G7 vPro ...
A Dell Latitude 5420 laptop powered by a four-core Intel Core i5-1145G7 vPro ...Principled Technologies
 
Xiv cloud machine_webinar_090414
Xiv cloud machine_webinar_090414Xiv cloud machine_webinar_090414
Xiv cloud machine_webinar_090414Jinesh Shah
 
Server TCO Showdown -- Lenovo x3950 X6 and IBM Storwize V7000 vs HP superdome
Server TCO Showdown -- Lenovo x3950 X6 and IBM Storwize V7000 vs  HP superdomeServer TCO Showdown -- Lenovo x3950 X6 and IBM Storwize V7000 vs  HP superdome
Server TCO Showdown -- Lenovo x3950 X6 and IBM Storwize V7000 vs HP superdomeLenovo Data Center
 
Lenovo System x3950 X6 vs HP Superdome 2
Lenovo System x3950 X6 vs  HP Superdome 2Lenovo System x3950 X6 vs  HP Superdome 2
Lenovo System x3950 X6 vs HP Superdome 2Lenovo Data Center
 
Fusion-io Virtualization Reference Architecture: Deploying Server and Desktop...
Fusion-io Virtualization Reference Architecture: Deploying Server and Desktop...Fusion-io Virtualization Reference Architecture: Deploying Server and Desktop...
Fusion-io Virtualization Reference Architecture: Deploying Server and Desktop...Principled Technologies
 
Comparing performance and cost: Dell PowerEdge VRTX vs. legacy hardware solution
Comparing performance and cost: Dell PowerEdge VRTX vs. legacy hardware solutionComparing performance and cost: Dell PowerEdge VRTX vs. legacy hardware solution
Comparing performance and cost: Dell PowerEdge VRTX vs. legacy hardware solutionPrincipled Technologies
 
Upgrade to Dell EMC PowerEdge R940 servers with VMware vSphere 7.0 and gain g...
Upgrade to Dell EMC PowerEdge R940 servers with VMware vSphere 7.0 and gain g...Upgrade to Dell EMC PowerEdge R940 servers with VMware vSphere 7.0 and gain g...
Upgrade to Dell EMC PowerEdge R940 servers with VMware vSphere 7.0 and gain g...Principled Technologies
 
Dell PowerEdge VRTX and M-series compute nodes configuration study
Dell PowerEdge VRTX and M-series compute nodes configuration studyDell PowerEdge VRTX and M-series compute nodes configuration study
Dell PowerEdge VRTX and M-series compute nodes configuration studyPrincipled Technologies
 
HRG Assessment:Comparing IBM PureSystems and Cisco UCS
HRG Assessment:Comparing IBM PureSystems and Cisco UCSHRG Assessment:Comparing IBM PureSystems and Cisco UCS
HRG Assessment:Comparing IBM PureSystems and Cisco UCSIBM India Smarter Computing
 
Database performance: Dell PowerEdge R730xd vs. Lenovo ThinkServer RD650
Database performance: Dell PowerEdge R730xd vs. Lenovo ThinkServer RD650Database performance: Dell PowerEdge R730xd vs. Lenovo ThinkServer RD650
Database performance: Dell PowerEdge R730xd vs. Lenovo ThinkServer RD650Principled Technologies
 
Citrix XenDesktop 5.5 vs. VMware View 5: User experience and bandwidth consum...
Citrix XenDesktop 5.5 vs. VMware View 5: User experience and bandwidth consum...Citrix XenDesktop 5.5 vs. VMware View 5: User experience and bandwidth consum...
Citrix XenDesktop 5.5 vs. VMware View 5: User experience and bandwidth consum...Principled Technologies
 
IBM PureFlex Solution for Cloud Backup and Recovery: Private Cloud Disaster R...
IBM PureFlex Solution for Cloud Backup and Recovery: Private Cloud Disaster R...IBM PureFlex Solution for Cloud Backup and Recovery: Private Cloud Disaster R...
IBM PureFlex Solution for Cloud Backup and Recovery: Private Cloud Disaster R...IBM India Smarter Computing
 

What's hot (20)

Vm Ware X Xen Server
Vm Ware X Xen ServerVm Ware X Xen Server
Vm Ware X Xen Server
 
X-Pod for VDI Reference Architecture Enabled by Cisco UCS, VMware Horizon Vie...
X-Pod for VDI Reference Architecture Enabled by Cisco UCS, VMware Horizon Vie...X-Pod for VDI Reference Architecture Enabled by Cisco UCS, VMware Horizon Vie...
X-Pod for VDI Reference Architecture Enabled by Cisco UCS, VMware Horizon Vie...
 
IBM pureflex system and vmware vcloud enterprise suite reference architecture
IBM pureflex system and vmware vcloud enterprise suite reference architectureIBM pureflex system and vmware vcloud enterprise suite reference architecture
IBM pureflex system and vmware vcloud enterprise suite reference architecture
 
Ibm pure systems sales bootcamp
Ibm pure systems sales bootcampIbm pure systems sales bootcamp
Ibm pure systems sales bootcamp
 
IBM Flex System Reference Architecture for Microsoft SQL Server 2012 High Ava...
IBM Flex System Reference Architecture for Microsoft SQL Server 2012 High Ava...IBM Flex System Reference Architecture for Microsoft SQL Server 2012 High Ava...
IBM Flex System Reference Architecture for Microsoft SQL Server 2012 High Ava...
 
Flex system client_presentation
Flex system client_presentationFlex system client_presentation
Flex system client_presentation
 
Grid rac preso 051007
Grid rac preso 051007Grid rac preso 051007
Grid rac preso 051007
 
VDI performance comparison: Dell PowerEdge FX2 and FC430 servers with VMware ...
VDI performance comparison: Dell PowerEdge FX2 and FC430 servers with VMware ...VDI performance comparison: Dell PowerEdge FX2 and FC430 servers with VMware ...
VDI performance comparison: Dell PowerEdge FX2 and FC430 servers with VMware ...
 
A Dell Latitude 5420 laptop powered by a four-core Intel Core i5-1145G7 vPro ...
A Dell Latitude 5420 laptop powered by a four-core Intel Core i5-1145G7 vPro ...A Dell Latitude 5420 laptop powered by a four-core Intel Core i5-1145G7 vPro ...
A Dell Latitude 5420 laptop powered by a four-core Intel Core i5-1145G7 vPro ...
 
Xiv cloud machine_webinar_090414
Xiv cloud machine_webinar_090414Xiv cloud machine_webinar_090414
Xiv cloud machine_webinar_090414
 
Server TCO Showdown -- Lenovo x3950 X6 and IBM Storwize V7000 vs HP superdome
Server TCO Showdown -- Lenovo x3950 X6 and IBM Storwize V7000 vs  HP superdomeServer TCO Showdown -- Lenovo x3950 X6 and IBM Storwize V7000 vs  HP superdome
Server TCO Showdown -- Lenovo x3950 X6 and IBM Storwize V7000 vs HP superdome
 
Lenovo System x3950 X6 vs HP Superdome 2
Lenovo System x3950 X6 vs  HP Superdome 2Lenovo System x3950 X6 vs  HP Superdome 2
Lenovo System x3950 X6 vs HP Superdome 2
 
Fusion-io Virtualization Reference Architecture: Deploying Server and Desktop...
Fusion-io Virtualization Reference Architecture: Deploying Server and Desktop...Fusion-io Virtualization Reference Architecture: Deploying Server and Desktop...
Fusion-io Virtualization Reference Architecture: Deploying Server and Desktop...
 
Comparing performance and cost: Dell PowerEdge VRTX vs. legacy hardware solution
Comparing performance and cost: Dell PowerEdge VRTX vs. legacy hardware solutionComparing performance and cost: Dell PowerEdge VRTX vs. legacy hardware solution
Comparing performance and cost: Dell PowerEdge VRTX vs. legacy hardware solution
 
Upgrade to Dell EMC PowerEdge R940 servers with VMware vSphere 7.0 and gain g...
Upgrade to Dell EMC PowerEdge R940 servers with VMware vSphere 7.0 and gain g...Upgrade to Dell EMC PowerEdge R940 servers with VMware vSphere 7.0 and gain g...
Upgrade to Dell EMC PowerEdge R940 servers with VMware vSphere 7.0 and gain g...
 
Dell PowerEdge VRTX and M-series compute nodes configuration study
Dell PowerEdge VRTX and M-series compute nodes configuration studyDell PowerEdge VRTX and M-series compute nodes configuration study
Dell PowerEdge VRTX and M-series compute nodes configuration study
 
HRG Assessment:Comparing IBM PureSystems and Cisco UCS
HRG Assessment:Comparing IBM PureSystems and Cisco UCSHRG Assessment:Comparing IBM PureSystems and Cisco UCS
HRG Assessment:Comparing IBM PureSystems and Cisco UCS
 
Database performance: Dell PowerEdge R730xd vs. Lenovo ThinkServer RD650
Database performance: Dell PowerEdge R730xd vs. Lenovo ThinkServer RD650Database performance: Dell PowerEdge R730xd vs. Lenovo ThinkServer RD650
Database performance: Dell PowerEdge R730xd vs. Lenovo ThinkServer RD650
 
Citrix XenDesktop 5.5 vs. VMware View 5: User experience and bandwidth consum...
Citrix XenDesktop 5.5 vs. VMware View 5: User experience and bandwidth consum...Citrix XenDesktop 5.5 vs. VMware View 5: User experience and bandwidth consum...
Citrix XenDesktop 5.5 vs. VMware View 5: User experience and bandwidth consum...
 
IBM PureFlex Solution for Cloud Backup and Recovery: Private Cloud Disaster R...
IBM PureFlex Solution for Cloud Backup and Recovery: Private Cloud Disaster R...IBM PureFlex Solution for Cloud Backup and Recovery: Private Cloud Disaster R...
IBM PureFlex Solution for Cloud Backup and Recovery: Private Cloud Disaster R...
 

Similar to Virtualization Performance on the IBM PureFlex System

IBM PowerVM Virtualization Technology on IBM POWER7 Systems
IBM PowerVM Virtualization Technology on IBM POWER7 SystemsIBM PowerVM Virtualization Technology on IBM POWER7 Systems
IBM PowerVM Virtualization Technology on IBM POWER7 SystemsIBM India Smarter Computing
 
IBM BCFC White Paper - Why Choose IBM BladeCenter Foundation for Cloud
IBM BCFC White Paper - Why Choose IBM BladeCenter Foundation for CloudIBM BCFC White Paper - Why Choose IBM BladeCenter Foundation for Cloud
IBM BCFC White Paper - Why Choose IBM BladeCenter Foundation for CloudIBM India Smarter Computing
 
Managing Complexity in the x86 Data Center: The User Experience
Managing Complexity in the x86 Data Center: The User ExperienceManaging Complexity in the x86 Data Center: The User Experience
Managing Complexity in the x86 Data Center: The User ExperienceIBM India Smarter Computing
 
Why Choose VMware for Server Virtualization
Why Choose VMware for Server VirtualizationWhy Choose VMware for Server Virtualization
Why Choose VMware for Server VirtualizationVMware
 
IBM i for Midsize Businesses Minimizing Costs and Risks for Midsize Businesses
IBM i for Midsize Businesses Minimizing Costs and Risks for Midsize BusinessesIBM i for Midsize Businesses Minimizing Costs and Risks for Midsize Businesses
IBM i for Midsize Businesses Minimizing Costs and Risks for Midsize BusinessesIBM India Smarter Computing
 
VMware vSphere 5 and IBM XIV Gen3 end-to-end virtualization
VMware vSphere 5 and  IBM XIV Gen3 end-to-end  virtualizationVMware vSphere 5 and  IBM XIV Gen3 end-to-end  virtualization
VMware vSphere 5 and IBM XIV Gen3 end-to-end virtualizationIBM India Smarter Computing
 
Future of Power: IBM PureFlex - Kim Mortensen
Future of Power: IBM PureFlex - Kim MortensenFuture of Power: IBM PureFlex - Kim Mortensen
Future of Power: IBM PureFlex - Kim MortensenIBM Danmark
 
Optimized Systems: Matching technologies for business success.
Optimized Systems: Matching technologies for business success.Optimized Systems: Matching technologies for business success.
Optimized Systems: Matching technologies for business success.Karl Roche
 
IBM PureFlex System: The Future of Data center Management
IBM PureFlex System: The Future of Data center ManagementIBM PureFlex System: The Future of Data center Management
IBM PureFlex System: The Future of Data center ManagementIBM India Smarter Computing
 
IBM PureFlex System: The Future of Datacenter Management
IBM PureFlex System: The Future of Datacenter ManagementIBM PureFlex System: The Future of Datacenter Management
IBM PureFlex System: The Future of Datacenter ManagementIBM India Smarter Computing
 
Intel® Xeon® Scalable Processors Enabled Applications Marketing Guide
Intel® Xeon® Scalable Processors Enabled Applications Marketing GuideIntel® Xeon® Scalable Processors Enabled Applications Marketing Guide
Intel® Xeon® Scalable Processors Enabled Applications Marketing GuideIntel IT Center
 
X-Pod for Citrix VDI on UCS with ISE 700 Hybrid Storage Array
X-Pod for Citrix VDI on UCS with ISE 700 Hybrid Storage ArrayX-Pod for Citrix VDI on UCS with ISE 700 Hybrid Storage Array
X-Pod for Citrix VDI on UCS with ISE 700 Hybrid Storage ArrayX-IO Technologies
 
A foundation for database simplifi cation, consolidation and cost reduction
A foundation for  database simplifi cation,  consolidation and  cost reductionA foundation for  database simplifi cation,  consolidation and  cost reduction
A foundation for database simplifi cation, consolidation and cost reductionIBM India Smarter Computing
 
Presentazione IBM Flex System e System x Evento Venaria 14 ottobre
Presentazione IBM Flex System e System x Evento Venaria 14 ottobrePresentazione IBM Flex System e System x Evento Venaria 14 ottobre
Presentazione IBM Flex System e System x Evento Venaria 14 ottobrePRAGMA PROGETTI
 
Ibpstc30951 wdg ib_mbooklet
Ibpstc30951 wdg ib_mbookletIbpstc30951 wdg ib_mbooklet
Ibpstc30951 wdg ib_mbookletBloombase
 
Backing Up Mountains of Data to Disk
Backing Up Mountains of Data to DiskBacking Up Mountains of Data to Disk
Backing Up Mountains of Data to DiskIT Brand Pulse
 
IBM SONAS and VMware vSphere 5 scale-out cloud foundation: A reference guide ...
IBM SONAS and VMware vSphere 5 scale-out cloud foundation: A reference guide ...IBM SONAS and VMware vSphere 5 scale-out cloud foundation: A reference guide ...
IBM SONAS and VMware vSphere 5 scale-out cloud foundation: A reference guide ...IBM India Smarter Computing
 
Superior Cloud Economics with Power Systems
Superior Cloud Economics with Power Systems Superior Cloud Economics with Power Systems
Superior Cloud Economics with Power Systems IBM Power Systems
 

Similar to Virtualization Performance on the IBM PureFlex System (20)

IBM PowerVM Virtualization Technology on IBM POWER7 Systems
IBM PowerVM Virtualization Technology on IBM POWER7 SystemsIBM PowerVM Virtualization Technology on IBM POWER7 Systems
IBM PowerVM Virtualization Technology on IBM POWER7 Systems
 
IBM BCFC White Paper - Why Choose IBM BladeCenter Foundation for Cloud
IBM BCFC White Paper - Why Choose IBM BladeCenter Foundation for CloudIBM BCFC White Paper - Why Choose IBM BladeCenter Foundation for Cloud
IBM BCFC White Paper - Why Choose IBM BladeCenter Foundation for Cloud
 
Managing Complexity in the x86 Data Center: The User Experience
Managing Complexity in the x86 Data Center: The User ExperienceManaging Complexity in the x86 Data Center: The User Experience
Managing Complexity in the x86 Data Center: The User Experience
 
Why Choose VMware for Server Virtualization
Why Choose VMware for Server VirtualizationWhy Choose VMware for Server Virtualization
Why Choose VMware for Server Virtualization
 
IBM i for Midsize Businesses Minimizing Costs and Risks for Midsize Businesses
IBM i for Midsize Businesses Minimizing Costs and Risks for Midsize BusinessesIBM i for Midsize Businesses Minimizing Costs and Risks for Midsize Businesses
IBM i for Midsize Businesses Minimizing Costs and Risks for Midsize Businesses
 
VMware vSphere 5 and IBM XIV Gen3 end-to-end virtualization
VMware vSphere 5 and  IBM XIV Gen3 end-to-end  virtualizationVMware vSphere 5 and  IBM XIV Gen3 end-to-end  virtualization
VMware vSphere 5 and IBM XIV Gen3 end-to-end virtualization
 
Future of Power: IBM PureFlex - Kim Mortensen
Future of Power: IBM PureFlex - Kim MortensenFuture of Power: IBM PureFlex - Kim Mortensen
Future of Power: IBM PureFlex - Kim Mortensen
 
Optimized Systems: Matching technologies for business success.
Optimized Systems: Matching technologies for business success.Optimized Systems: Matching technologies for business success.
Optimized Systems: Matching technologies for business success.
 
IBM PureFlex System: The Future of Data center Management
IBM PureFlex System: The Future of Data center ManagementIBM PureFlex System: The Future of Data center Management
IBM PureFlex System: The Future of Data center Management
 
IBM PureFlex System: The Future of Datacenter Management
IBM PureFlex System: The Future of Datacenter ManagementIBM PureFlex System: The Future of Datacenter Management
IBM PureFlex System: The Future of Datacenter Management
 
Build your own Cloud
Build your own CloudBuild your own Cloud
Build your own Cloud
 
Intel® Xeon® Scalable Processors Enabled Applications Marketing Guide
Intel® Xeon® Scalable Processors Enabled Applications Marketing GuideIntel® Xeon® Scalable Processors Enabled Applications Marketing Guide
Intel® Xeon® Scalable Processors Enabled Applications Marketing Guide
 
X-Pod for Citrix VDI on UCS with ISE 700 Hybrid Storage Array
X-Pod for Citrix VDI on UCS with ISE 700 Hybrid Storage ArrayX-Pod for Citrix VDI on UCS with ISE 700 Hybrid Storage Array
X-Pod for Citrix VDI on UCS with ISE 700 Hybrid Storage Array
 
IBM zEnterprise Strategy for the Private Cloud
IBM zEnterprise Strategy for the Private CloudIBM zEnterprise Strategy for the Private Cloud
IBM zEnterprise Strategy for the Private Cloud
 
A foundation for database simplifi cation, consolidation and cost reduction
A foundation for  database simplifi cation,  consolidation and  cost reductionA foundation for  database simplifi cation,  consolidation and  cost reduction
A foundation for database simplifi cation, consolidation and cost reduction
 
Presentazione IBM Flex System e System x Evento Venaria 14 ottobre
Presentazione IBM Flex System e System x Evento Venaria 14 ottobrePresentazione IBM Flex System e System x Evento Venaria 14 ottobre
Presentazione IBM Flex System e System x Evento Venaria 14 ottobre
 
Ibpstc30951 wdg ib_mbooklet
Ibpstc30951 wdg ib_mbookletIbpstc30951 wdg ib_mbooklet
Ibpstc30951 wdg ib_mbooklet
 
Backing Up Mountains of Data to Disk
Backing Up Mountains of Data to DiskBacking Up Mountains of Data to Disk
Backing Up Mountains of Data to Disk
 
IBM SONAS and VMware vSphere 5 scale-out cloud foundation: A reference guide ...
IBM SONAS and VMware vSphere 5 scale-out cloud foundation: A reference guide ...IBM SONAS and VMware vSphere 5 scale-out cloud foundation: A reference guide ...
IBM SONAS and VMware vSphere 5 scale-out cloud foundation: A reference guide ...
 
Superior Cloud Economics with Power Systems
Superior Cloud Economics with Power Systems Superior Cloud Economics with Power Systems
Superior Cloud Economics with Power Systems
 

More from IBM India Smarter Computing

Using the IBM XIV Storage System in OpenStack Cloud Environments
Using the IBM XIV Storage System in OpenStack Cloud Environments Using the IBM XIV Storage System in OpenStack Cloud Environments
Using the IBM XIV Storage System in OpenStack Cloud Environments IBM India Smarter Computing
 
TSL03104USEN Exploring VMware vSphere Storage API for Array Integration on th...
TSL03104USEN Exploring VMware vSphere Storage API for Array Integration on th...TSL03104USEN Exploring VMware vSphere Storage API for Array Integration on th...
TSL03104USEN Exploring VMware vSphere Storage API for Array Integration on th...IBM India Smarter Computing
 
A Comparison of PowerVM and Vmware Virtualization Performance
A Comparison of PowerVM and Vmware Virtualization PerformanceA Comparison of PowerVM and Vmware Virtualization Performance
A Comparison of PowerVM and Vmware Virtualization PerformanceIBM India Smarter Computing
 

More from IBM India Smarter Computing (20)

Using the IBM XIV Storage System in OpenStack Cloud Environments
Using the IBM XIV Storage System in OpenStack Cloud Environments Using the IBM XIV Storage System in OpenStack Cloud Environments
Using the IBM XIV Storage System in OpenStack Cloud Environments
 
All-flash Needs End to End Storage Efficiency
All-flash Needs End to End Storage EfficiencyAll-flash Needs End to End Storage Efficiency
All-flash Needs End to End Storage Efficiency
 
TSL03104USEN Exploring VMware vSphere Storage API for Array Integration on th...
TSL03104USEN Exploring VMware vSphere Storage API for Array Integration on th...TSL03104USEN Exploring VMware vSphere Storage API for Array Integration on th...
TSL03104USEN Exploring VMware vSphere Storage API for Array Integration on th...
 
IBM FlashSystem 840 Product Guide
IBM FlashSystem 840 Product GuideIBM FlashSystem 840 Product Guide
IBM FlashSystem 840 Product Guide
 
IBM System x3250 M5
IBM System x3250 M5IBM System x3250 M5
IBM System x3250 M5
 
IBM NeXtScale nx360 M4
IBM NeXtScale nx360 M4IBM NeXtScale nx360 M4
IBM NeXtScale nx360 M4
 
IBM System x3650 M4 HD
IBM System x3650 M4 HDIBM System x3650 M4 HD
IBM System x3650 M4 HD
 
IBM System x3300 M4
IBM System x3300 M4IBM System x3300 M4
IBM System x3300 M4
 
IBM System x iDataPlex dx360 M4
IBM System x iDataPlex dx360 M4IBM System x iDataPlex dx360 M4
IBM System x iDataPlex dx360 M4
 
IBM System x3500 M4
IBM System x3500 M4IBM System x3500 M4
IBM System x3500 M4
 
IBM System x3550 M4
IBM System x3550 M4IBM System x3550 M4
IBM System x3550 M4
 
IBM System x3650 M4
IBM System x3650 M4IBM System x3650 M4
IBM System x3650 M4
 
IBM System x3500 M3
IBM System x3500 M3IBM System x3500 M3
IBM System x3500 M3
 
IBM System x3400 M3
IBM System x3400 M3IBM System x3400 M3
IBM System x3400 M3
 
IBM System x3250 M3
IBM System x3250 M3IBM System x3250 M3
IBM System x3250 M3
 
IBM System x3200 M3
IBM System x3200 M3IBM System x3200 M3
IBM System x3200 M3
 
IBM PowerVC Introduction and Configuration
IBM PowerVC Introduction and ConfigurationIBM PowerVC Introduction and Configuration
IBM PowerVC Introduction and Configuration
 
A Comparison of PowerVM and Vmware Virtualization Performance
A Comparison of PowerVM and Vmware Virtualization PerformanceA Comparison of PowerVM and Vmware Virtualization Performance
A Comparison of PowerVM and Vmware Virtualization Performance
 
X6: The sixth generation of EXA Technology
X6: The sixth generation of EXA TechnologyX6: The sixth generation of EXA Technology
X6: The sixth generation of EXA Technology
 
Stephen Leonard IBM Big Data and cloud
Stephen Leonard IBM Big Data and cloudStephen Leonard IBM Big Data and cloud
Stephen Leonard IBM Big Data and cloud
 

Virtualization Performance on the IBM PureFlex System

  • 1. 89 Fifth Avenue, 7th Floor New York, NY 10003 www.TheEdison.com 212.367.7400 White Paper Virtualization Performance on the IBM PureFlex System
  • 2. Printed in the United States of America Copyright  2012 Edison Group, Inc. New York. Edison Group offers no warranty either expressed or implied on the information contained herein and shall be held harmless for errors resulting from its use. All products are trademarks of their respective owners. First Publication: April 2012 Produced by: Craig Norris, Sr Analyst; Barry Cohen, Editor-in-Chief; Manny Frishberg, Editor This document was developed with IBM funding. Although the document may utilize publicly available material from various vendors, including IBM, it does not necessarily reflect the positions of such vendors on the issues addressed in this document.
  • 3. Table of Contents Executive Summary ..................................................................................................................... 1 Introduction .................................................................................................................................. 3 Objective .................................................................................................................................. 3 Audience .................................................................................................................................. 3 Contents of this Report .......................................................................................................... 3 The Business Value of Virtualization ...................................................................................... 4 IBM's Unified Virtualization Infrastructure: PureFlex ........................................................ 5 TPoX: Benchmark Workload for Modern Virtualized DBMS-Based Environments...... 7 Benchmark Comparison Study and Results ........................................................................... 8 Test System .............................................................................................................................. 8 Test #1: Virtualization Tests and Results............................................................................. 9  Methodology .................................................................................................................... 9  Results.............................................................................................................................. 11 Test #2: Storage ..................................................................................................................... 13 Conclusions ................................................................................................................................. 15
  • 4. Executive Summary Using virtualization to consolidate data center servers has become an integral component of the way successful companies design their IT infrastructures. Most, however, use consolidation ratios of only around six virtual machines (VMs) per physical server. Even world-class organizations typically are only consolidating at a ratio of about 18 to 1 at best.1 Yet many commonly published virtualization benchmarks test with hundreds of VMs per physical server, which the average data center administrator has difficulty relating to their actual environment. This white paper is meant to appeal to anyone seeking benchmark evaluations of consolidation solutions that look more like what they themselves are doing or considering for their organizations. The tests described in this paper applied Transaction Processing over XML (TPoX) benchmark workloads to the IBM PureFlex system: a synergistic solution capable of simultaneously running and managing the best of heterogeneous IT environments — Linux, AIX, Windows, and IBM i operating systems, POWER7- and x86-based integrated technology elements (ITEs), PowerVM and VMware vSphere virtualization, HDD and SSD storage — all in a general-purpose system having all the simplicity of an appliance. TPoX is a much more suitable benchmark for assessing transaction processing throughput for today’s complex, virtualized IT environments than traditional industry standard benchmarks. In this paper, benchmarks are used to examine the throughput performance for various system configurations the PureFlex system can accommodate, as well as the ability of the solution to scale without degrading that performance. All tests were performed on the same self-contained PureFlex system unit. Key findings include the following:  IBM PowerVM on the IBM POWER7 ITE can deliver over 78 percent more throughput at 80 virtual machines (VMs) than vSphere 5 on an Intel Xeon E5-2690 (Sandy Bridge2)-based ITE.  AIX and Linux deliver near equivalent performance in leveraging this PowerVM capacity. 1Aberdeen Group report, Best-in-Class Practices for Virtualizing Microsoft Applications, August 2010 2Sandy Bridge is the codename for both the Intel microarchitecture innovation following Nehalem, and generally for the associated family of 32nm processors based upon that microarchitecture. Edison: IBM – TPoX Benchmark Results on IBM PureFlex System White Paper Page 1
  • 5. On either Intel-based or POWER7-based platforms, the PureFlex system exploits IBM Storwize V7000 to deliver up to 10 times the number of VMS per SSD than is possible using HDDs. This paper makes clear that the IBM PureFlex system is a superior solution for running and managing mixed IT environments in a unified solution. It also show that, for those interested in getting serious about virtualized consolidation, PowerVM virtualization technology on POWER7 processor-based platforms offers greater performance than that offered by VMware vSphere 5 on Intel x86 platforms. It enables high consolidation ratios and increased flexibility for a far superior virtualization solution. Edison: IBM – TPoX Benchmark Results on IBM PureFlex System White Paper Page 2
  • 6. Introduction Objective The objective of this white paper is to demonstrate the performance and capabilities of IBM’s PureFlex system, presenting results that showcase throughput and virtualization capacity for each of the multiple operating systems and ITEs comprised by this advanced modular server system, as well as its storage subsystem and its next- generation networking system. It describes tests using the industry-standard TPoX benchmark to compare virtualization technologies. The results were reviewed, analyzed, and presented by Edison Group. Audience This paper is intended for anyone interested in the advantages of server consolidation through virtualization. The testing described herein employed a means of assessing virtualization efficiency at a 5 VM-per-core ratio, which more closely reflects typical data center environments than the hundreds of VMs per server tested in other commonly published virtualization benchmarks. CTOs, CEOs, IT managers, and others will find valuable information here that could help them further enhance and/or adopt virtualization technology within their IT environments. Contents of this Report This white paper contains the following major sections:  The Business Value of Virtualization — This section discusses the business value propositions underlying the benchmark evaluations presented in this paper.  IBM’s Unified Virtualization Infrastructure: PureFlex — This section describes IBM’s PureFlex system offering on which the benchmarks were conducted.  TPoX — This section describes the TPoX benchmark and the reason for using it to evaluate transaction throughput in modern virtualized environments.  Benchmark Comparison Study & Results — This section presents the comparative testing, describing the test equipment setup, the benchmarks, the actual tests, and the results of the tests. Edison: IBM – TPoX Benchmark Results on IBM PureFlex System White Paper Page 3
  • 7. The Business Value of Virtualization As the foundation platform for today’s data center, server virtualization is quickly reaching maturity. Virtual machines now host more than half of business server workloads.3 Virtualization has become the default build for new server installations, reducing costs and establishing the foundation for more efficient and flexible configurations and technology platforms. 4 The performance of virtualization is critical to realizing success of server pools and cloud computing (and is also a key component in IBM’s roadmap in its Smarter Computing initiative). Well-implemented virtualization solutions may be employed to:  Reduce hardware expenditures by consolidating multiple environments, including underutilized servers, and systems with varied and dynamic resource requirements.  Reduce costs for power and cooling, floor space, hardware maintenance, and software licensing.  Grow and shrink resources dynamically according to business needs.  Deploy new workloads through provisioning VMs on new systems rapidly to meet changing business demands.  Develop and test applications in secure, independent domains while allocating production to its own domain on the same system.  Transfer live workloads to support server migrations, balance system load, or avoid planned downtime that can otherwise adversely impact productivity.  Control server sprawl, reducing system management costs. Despite these benefits, the majority of businesses fall far short of seizing upon the full potential of server consolidation. Their average consolidation ratios hover around six VMs per server,5 yet economic advantages from data center consolidation increase significantly at much higher VM densities. By increasing the consolidation ratio per system, businesses can reduce capital expenditures and operational costs by reducing the number of systems in their data center or IT organization. IBM’s Smarter Computing systems, which allow for greater VM density without degrading system performance, can deliver considerable economic advantages to organizations using them. This study examines the performance and scaling aspects of the IBM PureFlex system across the commonly employed industry benchmark TPoX. 3 The Value of Memory-Dense Servers: IBM’s System x MAX5 for its eX5 Server Family, March 2010, IDC 4 Ibid. 5 According to an Aberdeen Group report, Best-in-Class Practices for Virtualizing Microsoft Applications, August 2010, even the best-in-class organizations in the study consolidate at only an 18:1 ratio. Edison: IBM – TPoX Benchmark Results on IBM PureFlex System White Paper Page 4
  • 8. IBM’S Unified Virtualization Infrastructure: PureFlex With the PureFlex system, IBM brings together advanced modular server and storage technology in a virtualization infrastructure that leverages superior unified management and integrated Ethernet and fibre channel networking technology. The PureFlex system brings IBM’s leadership in computing systems and extensive expertise in systems integration to a synergistic solution capable of simultaneously running and managing the best of heterogeneous IT environments — Linux, AIX, Windows, and IBM i operating systems, Power Systems and x86-based ITEs, PowerVM and various x86 virtualization platforms, HDD and SSD storage — in a general-purpose system having all the simplicity of an appliance. Based on solidly established patterns of expertise, the PureFlex system is designed to get data center operations up and running in as little as four hours, cutting months off the deployment time for new application projects. The major subsystems integrated into the PureFlex system are:  POWER7 Processor-Based Power Systems ITEs — Automatically optimizing performance and capacity at either a system or VM level, these ITEs benefit from the POWER7 processor, which contains innovative technologies that help maximize performance and optimize energy efficiency. They represent the most flexible and cost-efficient solutions for UNIX, IBM i, and Linux deployments available in the market.  IBM x86-Based ITEs — These ITEs support a wide selection of processor technologies and operating systems. They are designed to reduce complexity, improve systems management, and increase energy efficiency while driving down total cost of ownership. Edison: IBM – TPoX Benchmark Results on IBM PureFlex System White Paper Page 5
  • 9. IBM Storwize V7000 — This is a virtualized storage controller designed to consolidate block and file workloads into a single storage system for simplicity of management, reduced cost, highly scalable capacity, performance, and high availability. It offers improved efficiency and flexibility through built-in solid state drive (SSD) optimization, thin provisioning, and non-disruptive migration of data from existing storage.  Next-Generation Networking — The PureFlex system incorporates a platform networking and storage fabric technology, utilizing integrated 16 Gb fibre channel switches and 10 Gb Ethernet switches to enable heterogeneous ITEs to run concurrently with little or no impact on each other, tying together these various subsystems with Ethernet and/or FC into a single manageable system. Edison: IBM – TPoX Benchmark Results on IBM PureFlex System White Paper Page 6
  • 10. TPoX: A Benchmark Workload for Modern Virtualized DBMS-Based Environments One of the goals of this study was to more closely approximate typical virtualized data center usage by organizations. This includes the type of application workloads to be used in the testing. Cloud computing, database-as-a-service, and virtualization are changing the way customers deploy databases, and a benchmark workload used to test the transaction processing performance of DBMS-based applications should be able to address these architectures, and illustrate to data center operators what they can expect as they increasingly leverage virtualization. TPoX (Transaction Processing over XML) is an application-level “XML database” benchmark. It is an XML OLTP benchmark using data-oriented XML structures, very large numbers of relatively small XML documents (1 kb to 20 kb), short read/write transactions, and a high degree of concurrency. It essentially models a security-trading scenario that uses a real-world XML Schema (FIXML). TPoX can be used to evaluate any database that offers XML support. The TPoX workload driver is architected such that only a thin layer (a single Java class) deals with the specific interaction to the database system under test. The combination of these XML and Java components makes the workload relevant for a wide range of web and collaboration applications. Applications in the real world typically have many concurrent users, a mix of read and write operations, and millions or even billions of XML documents. This is exactly what the TPoX benchmark is designed to capture. It simulates an actual application that performs queries, inserts, updates, and deletes in a concurrent multi-user workload using a real-world database. In the study to follow, the enterprise-class DB2 database was used on all three platforms tested. As a database-centric application, TPoX stresses CPU, memory, and storage I/O. It is capable of driving high CPU utilization to study true technology capability. In addition, in a multi-VM environment such as that tested for this study, it stresses the virtualization infrastructure supporting these resources on the tested platforms. TPoX is an open-source benchmark jointly developed by IBM, Intel, and others. The benchmark, along with documentation and published results, is publicly available at: http://tpox.sourceforge.net 6 6 Reference: http://nativexmldatabase.com/2011/03/04/new-tpox-benchmark-results-available/ Edison: IBM – TPoX Benchmark Results on IBM PureFlex System White Paper Page 7
  • 11. Benchmark Comparison Study and Results This section of the paper describes the system configuration used in the tests, methodology employed, and results for each. Test System The test system consists of a single PureFlex system provisioned with the following major subsystems:  One IBM Power p260 node with two 3.56 GHz Power 7 processors  One 2-socket ITE with 2.9 GHz Intel Xeon E5-2690 (Sandy Bridge) processors  Two Storwize V7000 storage controllers, each provisioned with 24 300 GB SSDs  Networking fabric comprising two 16 Gb fiber channel switches integrated into the PureFlex chassis Figure 1 (below) shows the positioning of the subsystems within the overall PureFlex system unit. Figure 1. Diagram of Test System Edison: IBM – TPoX Benchmark Results on IBM PureFlex System White Paper Page 8
  • 12. Table 1 (below) presents how the PureFlex system was configured for the TPoX performance benchmarks. NOTE: Not every component was necessarily utilized for every test. IBM Flex System P260 Sandy Bridge 2-Socket Compute Node Compute Node Processors 2 x IBM POWER7 2 x Intel Xeon E5-2690 Guest Operating AIX 7.1 System(s) SUSE Linux Enterprise SUSE Linux Enterprise Server 11 SP2 Server 11 SP2 Virtualization IBM Power VM Enterprise VMware vSphere 5 Edition v2.2 Database Backend IBM DB2 v9.7 Workload TPoX v2.1 Storage IBM Storwize V7000 / SSD Table 1: Configuration of the PureFlex System for TPoX Benchmarks Test # 1: Virtualization Tests and Results The goal of these tests was to determine aggregate transaction throughput performance and VM scalability. Methodology These tests employed a VM-to-core ratio of 5:1. The system was scaled up from a single core to two full sockets. In this way, tests were run on sets of 5, 10, 20, 40, and 80 VMs, each running TPoX, as shown in Figure 2, below. Edison: IBM – TPoX Benchmark Results on IBM PureFlex System White Paper Page 9
  • 13. TPoX Virtualization Tests 2 socket ITE 1 socket 1 core 5 VMs 10 VMs 20 VMs 40 VMs 80 VMs Figure 2: TPoX Virtualization Tests Methodology The TPoX benchmark is I/O-intensive and its performance is partially dependent on storage performance. Both ITE platforms used the same PureFlex system storage subsystem although each used a different Storwize V7000 Controller to store its data and logs. Each VM had its own XML database servicing clients; a 1 GB database was used in order to match up with each VM’s CPU (0.2 core) and memory capacity (3 GB). A single-tier TPoX configuration was chosen for each VM where the client and the database reside in the same VM. The goal was to run enough clients to ensure that each VM reached high utilization. The database for each VM on each of the platforms was populated with the same configuration set. The transaction rate for populating the database is shown in Table 2. Flex System Sandy Bridge / VMware p260/PowerVM/AIX vSphere 5 Order (inserts per second) 1915 1028 Custacc (inserts per second) 769 400 Table 2. TPoX Database Populated Rate for First Configuration Set Edison: IBM – TPoX Benchmark Results on IBM PureFlex System White Paper Page 10
  • 14. The VM configuration has multiple options on both PowerVM and VMware vSphere 5 technologies . These tests strove to utilize the full capacity of each core deployed. In the case of PowerVM, five virtual CPUs (vCPUs) per core were sufficient to accomplish this. In the case of vSphere 5, it was established in an earlier study 7 that two vCPUs per VM was more efficient than one vCPU per VM. Thus, each core was configured with 10 vCPUs for the vSphere 5 subsystem. For each test, the per VM transaction rate was measured, and the aggregate transaction rate at the system level was reported. Results The results presented in Figure 3, below, indicate that the throughput performance rate for populating the database are impressive for the Intel Xeon E5-2690 Sandy Bridge compute node on the IBM PureFlex system. However, as may be expected, throughput performance for Linux on POWER7/PowerVM technology averaged over 54 percent better. A slightly greater advantage proved to be the case when the workload was run on the AIX operating system. The POWER7/PowerVM advantage over Sandy Bridge/VMware also tended to grow with the number of VMs, with POWER7/AIX showing a nearly 76 percent advantage at 80 VMs, and POWER7/Linux showing a similar advantage at 72 percent. PowerVM / Flex p260 vs. vSphere / Sandy Bridge 10000 9000 8000 Transactions per Second 7000 6000 5000 4000 3000 2000 1000 0 5VMs 10VMs 20VMs 40VMs 80VMs Figure 3: TPoX Benchmark Results for Throughput Performance 7IBM PowerVM Virtualization Technology on IBM POWER7 Systems - A Comparison of PowerVM and VMware Virtualization Performance, Edison Group, Inc., 2012. Edison: IBM – TPoX Benchmark Results on IBM PureFlex System White Paper Page 11
  • 15. Table 3, below, presents detailed information on the total number of TPoX users used in each test, throughput, and VM configuration for each tests. Total System Configuration for TPoX Benchmark Virtual Total # of Transactions (1 to 80 VM Scaling) # of VMs CPUs TPoX Users per second IBM Flex p260 3.56 GHz DPSM mode, 2 socket, 256 GB RAM, SMT4 enabled, PowerVM, SUSE Linux Enterprise Server 11 SP2 is the host OS for each VM. Each LPAR is 5 5 50 600 configured with 1 vCPU/ uncapped/3 GB RAM 4 LPARs have 0.2 cores and 1 LPAR have 0.1 core. Shared pool has one core 9 LPARs are configured each with 0.2/1 vCPU/ uncapped/ 3 GB memory, 1 LPAR is 10 10 100 1112 configured with 0.1/1 vCPU/uncapped/3 GB memory. Shared pool has two cores 18 LPARs are configured each with 0.2/1 vCPU/ uncapped/ 3 GB memory, 2 LPARs are 20 20 200 2316 configured with 0.1/1 vCPU/uncapped/3 GB memory. Shared pool has four cores 37 LPARs configured with 0.2core/ 1vCPU/uncapped and 3 LPARs configured 40 40 400 4710 with 0.3 cores/1vCPU/ uncapped. Shared pool has eight cores 74 LPARs are configured each with 0.2/1 vCPU/ uncapped/ 3 GB memory, 4 LPARs are 80 80 800 9499 configured with 0.1/1 vCPU/uncapped/ 3 GB memory. Shared pool has 16 cores 2 socket 2.9 GHz, Intel Xeon E5-2690 compute node with 256 GB system RAM (HT and Turbo enabled in BIOS Intel VTx with EPT HW virtualization assist) VMware vSphere 5. Each VM has guest OS SUSE Linux Enterprise 5 10 50 428 Server 11 SP2. Each VM is given 0.2 of a core/1 2 vCPUs/3 GB memory. DB2 buffer pool for data is configured in each VM. Schedule affinity is set to cpu0 and cpu1. Schedule affinity is set to cpu0 to cpu3 10 20 100 785 Schedule affinity is set to cpu0 to cpu7 20 40 200 1526 Schedule affinity is set to cpu0 to cpu15 40 80 400 2839 Schedule affinity is set to cpu0 to cpu31 80 160 800 5516 Edison: IBM – TPoX Benchmark Results on IBM PureFlex System White Paper Page 12
  • 16. Total System Configuration for TPoX Benchmark Virtual Total # of Transactions (1 to 80 VM Scaling) # of VMs CPUs TPoX Users per second IBM Flex p260 3.56 GHz DPSM mode, 2 socket, 256 GB RAM, SMT4 enabled, PowerVM, IBM AIX .1 is the host OS for each VM. VIOS is configured with 0.1 core/1 vCPU/ 5 5 50 629 uncapped mode/ 4 GB RAM. Each LPAR is configured with 1 vCPU/ uncapped/3 GB RAM 4 LPARs have 0.2 cores and 1 LPARs have 0.1 core Shared pool has one core 9 LPARs are configured each with 0.2/1 vCPU/ uncapped/ 3 GB memory, 1 LPAR is 10 10 100 1307 configured with 0.1/1 vCPU/uncapped/3 GB memory. Shared pool has two cores 18 LPARs are configured each with 0.2/1 vCPU/ uncapped/ 3 GB memory, 2 LPARs are 20 20 200 2642 configured with 0.1/1 vCPU/uncapped/3 GB memory. Shared pool has four cores 37LPARs configured with 0.2core/ 1vCPU/uncapped and 3 LPARs configured 40 40 400 5063 with 0.3core/1vCPU/ uncapped. Shared pool has eight cores LPARs are configured each with 0.2/1 vCPU/ uncapped/ 3 GB memory, 4 LPARs are 80 80 800 9703 configured with 0.1/1 vCPU/uncapped/ 3 GB memory. Shared pool has 16 cores Table 3. TPoX Benchmark Detailed Test Information Test # 2: Storage The Storwize V7000 storage controller in the PureFlex system provides the flexibility to deploy any mix of both hard drive disks (HDDs) and Solid State Drives (SSDs). Since OLTP applications (such as TPoX) particularly benefit from high-throughput storage, the goal of this setup was to determine the effect that provisioning the Storwize V7000 controller with SSDs would have on the VM density capability of the PureFlex system. Figure 4, below, presents a visual representation comparing maximum virtualization possible using the TPoX benchmarks for the Storwize V7000 Controller provisioned with Edison: IBM – TPoX Benchmark Results on IBM PureFlex System White Paper Page 13
  • 17. a full 24 HDDs (for an earlier study 8), and the same controller provisioned with 24 SSDs for the present study. As the figure indicates, a tenfold increase in the possible number of VMs per core results in provisioning SSDs rather than HDDs. HDD SSD Storwize V7000 Controller with 24 HDDs Storwize V7000 Controller with 24 SSDs Figure 4: VM Density Comparison between HDDs & SSDs When the Storwize V7000 was provisioned with 24 HDDs, 12 HD spindles were required for every four VMs in order to achieve sufficient parallelization to overcome the latency (seek time) associated with HDDs. Unlike the mechanical HDDs, which are limited by factors such as spindle speeds and actuator movement, SSDs involve no moving parts. They consume a fraction of the power required to operate and to cool HDDs, while taking up a fraction of the space. SSDs can also yield I/O response times up to 100 times greater than HDDs, making them particularly good choices for transaction-intensive workloads typical in database applications. While once too costly for all but the highest-end performance-intensive uses, SSDs have dropped in price considerably which — along with their savings in space, power, and maintenance (due to much fewer devices needed and their non- mechanical nature) — they can in certain scenarios actually be the more cost-effective choice over HDDS. 8IBM PowerVM Virtualization Technology on IBM POWER7 Systems - A Comparison of PowerVM and VMware Virtualization Performance, Edison Group, Inc., 2012. Edison: IBM – TPoX Benchmark Results on IBM PureFlex System White Paper Page 14
  • 18. Conclusions This paper shows that the IBM PureFlex system can deliver VM consolidation in a heterogeneous, self-contained environment capable of impressive levels of throughput performance. It can dramatically reduce time to production for virtualized data center application operations, providing multiple compute and operating system platforms, advanced storage, and integrated networking in a single manageable system. The benchmark results presented in this paper show that the IBM POWER7 processor- based ITE platform using PowerVM virtualization, in particular, demonstrated significantly superior throughput performance at all levels of VM population over the x86 Sandy Bridge ITE platform using vSphere 5. IBM PowerVM technology — the virtualization software built into the POWER7 processor-based systems — offers an unprecedented level of platform support, scalability, efficient resource utilization, flexibility, and heterogeneous server management. IBM PowerVM virtualization offers autonomic resource affinity, resulting in higher workload performance in a virtualized environment. The IBM PureFlex system delivers efficient virtualization of IBM POWER7 Systems and PowerVM technology, along with integrated IBM Storwize V7000 controllers that can accommodate SSDs for maximum per-core VM density, making the IBM PureFlex system an excellent foundation for cloud computing environments. POL03114-USEN-01 Edison: IBM – TPoX Benchmark Results on IBM PureFlex System White Paper Page 15
  • 19. Appendix A This appendix describes the test methodology used to compare PowerVM to VMware application benchmark performance across three (pLinux, AIX and xLinux) operating systems on the IBM Unified Virtualization Infrastructure. This performance evaluation characterizes hypervisor efficiency in scaling virtual machines on POWER7- and x86-based integrated technology elements (ITEs ). The experiments conducted assess the effect of adding VMs incrementally as well as incrementally increasing system resource consumption. Throughput performance was measured as the number of VMs was scaled from 5 to 80. To ensure fair comparison across platforms and to remove variability across each set of tests, the following actions were taken: 1. Deployment of optimized VM configurations in terms of virtual processors on each platform and similar memory allocated per VM. a. a. On PowerVM, resources can be allocated to an I/O hosting partition, therefore resource allocation to I/O hosting partition is optimized for each set of tests on Power systems. This is not applicable to VMware. 2. The same set of “benchmark parameters” were used across platforms. 3. Except for the virtualized workloads under investigation, a common software stack was maintained across the three operating systems (DB server). 4. Tuning was performed based on best practices for respective platforms. 5. VMware vSphere 5.0, SuSE 11 SP2 (xLinux and pLinux), AIX 7.1, DB2 tuning. System Tuning X86 (Sandy Bridge) Integrated Technology Element  System (uEFI/BIOS) default settings were used (HT enabled, Turbo mode enabled, Operating Mode set to MAX performance). IBM POWER7 Integrated Technology Element  Dynamic Power Management / Favor Performance was enabled. POL03114-USEN-01 Edison: IBM – TPoX Benchmark Results on IBM PureFlex System White Paper Page 16
  • 20. Virtualization Tuning VMware Virtual Machine Configuration Details  VMware VMs were created using Virtual Machine version 8.  Virtual Disk LSI Logic Parallel was used. It was noted that the LSI Logic Parallel adapter and the LSI Logic SAS adapter offer equivalent performance. 9  Updates were made to the latest VMware tools.  esxtop –ab and vmstat were collected from the VM. AIX and pLinux (PowerVM Guest OS) and xLinux (VMware Guest OS) Tunings AIX Tunings *raso -r -o mtrc_enabled=0 (disables lightweight memory trace) *ctctrl -P memtraceoff (disables component trace) *errctrl -P errcheckoff (disables run-time error checking) *skeyctl -k off (disables storage keys) dscrctl -n -s 1 (disable prefetch)  Remove the Java 5 version from AIX 7.1 install (the benchmark used Java 6).  Disk queue depth increased to 20 on the VM (default 3 is low for this I/O-intensive workload). Disclaimer: These AIX tunings were done for benchmarking reasons; customers do not change any of these tunables in a production environment unless they are asked to by the AIX support team in the process of resolving an issue. pLinux Tunings  ppc64_cpu –dscrctl=1 (disable prefetch) DB2 Tuning on all three operating systems (xLinux, pLinux, AIX 7.1)  Database buffer pool (0.2 GB) is allocated for table space on each VM. 9 vSphere Help POL03114-USEN-01 Edison: IBM – TPoX Benchmark Results on IBM PureFlex System White Paper Page 17