DataCore Case Study on Hyperconverged

Copyright © 2017 DataCore Software Corp. – All Rights Reserved. DataCore SoftwareThe Data Infrastructure Software Company 1
Case Study on
Hyperconverged
Increase Performance,
Availability and Density

Copyright © 2017 DataCore Software Corp. – All Rights Reserved. DataCore Software
• Located in Jackson County on the
southern Oregon border
• Serves an estimated population of
200K over 2.8K square miles
• Receives over 300K calls per year
• 40 Dispatchers with 12
administrative staff
• Works with 30 agencies covering
Fire, Law enforcement and
Emergency Medical agencies
2
Emergency Communications
of Southern Oregon

• Mission-critical application: Call dispatch application built on MS SQL
Server
• Enter 911 call details
• Query databases to access call and case histories
• Find the closest unit to send for help using GPS
• Application was unable to keep up with dispatcher
• The dispatcher would have to wait for queries or commands to finish
• Root cause analysis showed bottleneck was storage
• Multiple times per day, latencies were 200ms+
3
Key Challenge: SQL Server Performance

Copyright © 2017 DataCore Software Corp. – All Rights Reserved. DataCore Software 4
Better Availability
• RPO goal of 0 (no data loss)
• Wanted synchronous replication
to DR site
• RTO goal of 30 mins
• Still long time for emergency
communications to be down
• Wanted vMotion to DR site when
needed
• Able to test failover and failback
Lower Costs
• Continue using existing NetApp
storage (only a year and half old)
• Needed new hosts with
additional compute and memory
• Wanted a single pane of glass for
management
Other Challenges
(aside from storage latency)

Simplified View of the Dispatch Application
5
Call Taker inputs call information to database (create to entry)
Dispatcher receives
priors and premise
information
Call summary sent to
Responders’ phone
Call details sent to
Responders’ laptops

• Law Enforcement – Dispatch Benchmarks
(Priority E & 1)
• Processed within 90 seconds, 90% of the
time (Creation to enter)
time (Creation to enter)
• Emergency Medical – Dispatch
Benchmarks (Priority E & 1)
time (Creation to entry)
• Dispatched to the appropriate agency(s)
within 30 seconds, 90% of the time (Entry
to dispatch)
• Fire – Dispatch Benchmarks (Emergency
fire incidents)
• Processed and dispatched within 90
seconds, 90% of the time (creation to
dispatch)
• Create to Entry = 60 seconds
• Entry to Dispatch = 30 seconds
• Processed and dispatched within 120
seconds, 95% of the time (creation to
dispatch).
• Create to Entry = 90 seconds
• Entry to Dispatch = 30 seconds
ECSO Call & Dispatch Goals

ESCO IT Infrastructure (Previous)
Virtualized Infrastructure
• 1 host each at primary
and secondary
• 25 VMs: Standard
business applications,
Active Directory and
Exchange
10 most critical VMs
Storage
• NetApp FAS2240-4 at
primary and secondary
(24TB each)
• Scheduled synchronization
job periodically depending
upon the importance of the
data
• For critical data, job ran
every 15 minutes
• Most data was 12 hours old
Physical Servers
• SQL Server was NOT
virtualized due to
inadequate storage
performance
• SQL Server hosts were
clustered (DAS)

Software
• ESXI accelerator software
(PernixData, Infinio)
• Not enough performance; didn’t
address availability
• VMware vSAN software
• Expensive; incompatible with
existing hardware
• Zerto
• Didn’t address performance
Hardware
• New storage arrays
• NetApp
• Pure Storage
• Nimble Storage
• Decent performance but not a
huge increase
Alternatives Considered

• Used Iometer
• Block sizes: 2K, 4K, 8K, 32K, 64K, 128K, and 256K
• Options: Random vs sequential / Read vs write
• 28 total runs
• 2K, Random, Read
• 2K, Random, Write (and so on)
• Note: Same VM was used in testing, but hardware was different
• NetApp had separate servers and storage
• DataCore hyperconverged had combined server and storage
• Same servers generating I/O load had to handle it
9
Testing I/O Performance

0
5,000
10,000
15,000
20,000
25,000
30,000
Average Median
IOPS
NetApp DataCore
0
50
100
150
200
250
300
350
400
450
500
Average Median
Throughput (MB/s)
NetApp DataCore
Comparison of I/O Performance
421% increase
1,374% increase
715% increase
429% increase

• Numbers
• In Production: 20x improvement in latency (on average)
• Previously, peak latency was 200ms and hit multiple times per day
• Current peak is less than 5ms (40x improvement)
• In Lab: 4x to 14x performance increase
• In some cases 100X improvement in IOPS and Throughput
• Anecdotes
• Dispatchers no longer complain about the input lag they were experiencing
• Able to vMotion virtual machines to secondary site for maintenance and users
unaware they are no longer running on primary site
11
Summary of Performance Improvement

ESCO IT Infrastructure (Current)
Hosts
• SQL Server
instances were
virtualized
• Went from 5 total
servers (2 virtual +
3 physical) down to
2 virtualized hosts
Hosts
• Went from 3 total
servers (1 virtual +
2 physical) down to
1 virtualized host
Storage
• DataCore
Hyperconverged
Virtual SAN (1 per
host) with direct-
attached storage
• Data is
synchronously
mirrored to DR site
• 911 calls, radio
traffic, and critical
SQL Server data
are using CDP
• NetApp storage
became tier 3
(archive)

Better Availability
• Synchronous replication between hosts so RPO is 0 (no loss of data)
• RTO is time to restart VMs
Increased Flexibility
• Added memory to increase Cache
• Adding NVME to add storage tier
• SQL Servers are now virtualized
13
Availability & Flexibility Benefits

• 60%+ less infrastructure
• Reduced hosts from 8 to 3
• Eliminated external storage for production data
• Save on administration time and PCSE
14
Infrastructure Cost Reduction

• Saved ~$100K on SQL Server licenses
• Change to per core licensing was going to cost ~$120K with physical hosts
• Each physical host was single processor with 12 cores (5 servers x 12 cores)
• By virtualizing SQL Server, able to control number of cores
• 1 VM with 8 vCPUs – Cores and 2 VMs with 4 vCPUs – Cores (total of 16 vCPUs – Cores)
• Better performance enabled consolidation of instances
• Went from 5 physical servers to 3 VMs
• ~$25K in license costs afterwards
15
SQL Server License Cost Reduction

• UPS at primary site was to be replaced
• Would knock out power for 6-10 hours to primary site
• All VMs were “vMotioned” to secondary site in 10 minutes
• Users were able to continue accessing applications
• No performance degradation (completely transparent to them)
• Once UPS was replaced, all VMs were “vMotioned” back to primary site
• Started at midnight, back to primary by 9am
• Storage is no longer a single point of failure, with no data loss due to
outages
16
Real-World Availability Situation

DataCore
Hyperconverged Virtual SAN
17

• Hyperconverged has been used primary for VDI and ROBO
• VDI environments typically require 200K IOPS
• Databases and other Tier 1 workloads require much more I/O
performance and low latency
• 1M IOPS or more are needed
• Key metric is latency
18
I/O Performance determines
Use Cases for Hyperconverged
Sources: IDC

IO Gap
Compute has outpaced Storage Performance
19
Increasingly faster
Uni-processors Serial IO
Work
Potential
2010 20202000
CPU clock rates
slow down
More cores
per socket

SPC Performance Benchmark:
Database I/O Workload
Criteria SPC Benchmark
Industry Standard
Independently Verified & Audited
Peer Reviewed
Covers different types & generations of technology
Maps to “real world” performance (OLTP databases)
Shows cost for achieving performance level
Enterprise Vendors run SPC Benchmarks

DataCore™ Hyperconverged DataCore™ Hyperconverged
DataCore: First Hyperconverged product to run SPC-1
World Record Results!
PERFORMANCE
1.2 million
SPC-1 IOPSTM
Highest
Productivity
PRICE-PERFORMANCE
$0.10
Per SPC-1 IOPSTM
Unmatched
Economics
RESPONSE TIME
0.22
milliseconds
Ultra Fast
Applications
High-Availability / Multi-Node / Fibre Channel
Source: Storage Performance Council SPC-1 Benchmark as of Nov 3, 20161

0.80 0.76
1.00
0.99
0.22
0
0.25
0.5
0.75
1
DataCore
$0.10
HPE 3PAR
8450
$0.23
NetApp
EF560
$0.30
Dell
SC4020
$0.37
EMC
VNX8000
$0.41
Latency (Milliseconds)
Increasing Price per SPC-1 IOPS
4.8
22
DataCore Hyperconverged is Faster than AFAs

DataCore Ran Both SPC-1 Database
Workload + Storage Load at SAME Time!
All-inclusive
Price
(HA 2 Nodes)
Hyperconverged SPC-1
$115K
1.2M IOPS
SAN
(Only Partial
Price)
Servers
Not Included
in Price
Traditional SPC-1
$177K
435K IOPS>>
<<
Source: Storage Performance Council SPC-1 Benchmark as of Nov 3, 20162

Nutanix’s I/O Performance
(from Nutanix Staff Solution Architect)
24
https://twitter.com/josh_odgers/status/728938346106343425

Nutanix hasn’t allowed an Independent, 3rd
Party Evaluation of Performance
25
http://www.storagereview.com/why_we_don_t_have_a_nutanix_nx8150_review

DataCore Performance
 The system averaged ~4,380 IOPS per VM (total
aggregate across 2 nodes was ~263K IOPS)
 DataCore on 2 Nodes ~263K IOPS
 DataCore on 3 Nodes ~395K IOPS
 DataCore on 16 nodes ~2,104K IOPS
 Nutanix on 16 Nodes ~342K IOPs
 The DataCore powered system was able to
sustain this level of performance on one node –
that’s 60 VMs running on one node averaging 4K
IOPs per VM
 No application and no additional hardware
changes required for this performance

Nutanix DataCore Advantage DataCore
IOPS per Host ~21K ~132K 6.3X more IOPS per host
VMs per Host ~7 30* 4.3X more VMs
IOPS per VM ~2.9K ~4.4K 152% more IOPS per VM
Latency (milliseconds) 1.36 0.5 272% Faster
RAM per Host 512GB 384GB 25% less RAM
27
Comparison: Nutanix vs DataCore
* Full N+1 redundancy (each host can support 60 VMs)
Conclusion: 4X to 6X less servers needed with DataCore

WITHOUT PARALLEL I/O
I/O processed sequentially…
DataCore Parallel I/O Technology
28

WITH PARALLEL I/O
I/O processed in parallel…
WITHOUT PARALLEL I/O
I/O processed sequentially…
4 nodes do the work of 20!
DataCore Parallel I/O Technology
29

Less Active Data Placed
on Slower Storage
More Active Data Placed
on Faster Storage
Data is Cached and Tiered
SAS SATAFlash Cloud
HOT COLDWARM ARCHIVE
RAM Cache
BURST
30

• Performance of an individual node determines the unit of measurement
• X number of VDI sessions, Y number of databases, etc
• Scale Out
• Add more nodes as workload increases
• Poor individual node performance means lots of nodes
• Scale Up
• Increase capabilities of individual nodes
• But, can’t be done for “integrated HCI”
What happens if your data is growing faster than compute needs?
31
Scalability: Out and/or Up

Independently Scale-out Storage Capacity:
Leverage External SAN with Auto-Tiering
DataCore
SANsymphony for
existing SAN
Lower Hardware Costs; Capacity added as needed
DataCore
Hyperconverged
Virtual SAN
Hot Warm ColdBurst

DataCore Benefits
Improved
performance by
3X or more
IMPROVE
PERFORMANCE
Reduced storage-
related downtime by
50% of more
REDUCE
DOWNTIME
Reduced storage-
related spending by
25% or more
INCREASE
UTILIZATION
In the first year,
positive ROI
SAVE
MONEY
33

30,000+ DEPLOYMENTS WORLDWIDE
10,000+ Customers 10th Gen Product
Companies in all Industries & Sizes
Software-defined Storage & Hyperconverged
Technology: Storage Virtualization & Parallel I/O
Main Offices
• Australia
• France
• Germany
• Japan
• UK
• USA
PROVEN. GLOBALLY.
34

• Try DataCore Hyperconverged Virtual SAN in your environment
• Speed up critical applications & consolidate your infrastructure
• Schedule a 15-minute live demo with one of our Solution Architects
http://info.datacore.com/LiveDemo
• Download Free Trial of DataCore™ Hyperconverged Virtual SAN
http://www.datacore.com/resources/free-hyperconverged-virtual-san
35
Next Steps

DataCore Case Study on Hyperconverged

Recommended

Recommended

More Related Content

What's hot

What's hot (20)

Similar to DataCore Case Study on Hyperconverged

Similar to DataCore Case Study on Hyperconverged (20)

More from Advantech Industrial Automation Group

More from Advantech Industrial Automation Group (8)

Recently uploaded

Recently uploaded (20)

DataCore Case Study on Hyperconverged