VMworld, profile picture
Uploaded byVMworld
PPTX, PDF1,099 views

VMworld 2015: Extreme Performance Series - vCenter Performance Best Practices

This presentation discusses vCenter performance best practices. It provides an overview of the vCenter architecture and how various components like the database, inventory service, and web client can influence performance. The document outlines factors that impact vCenter resource usage and strategies for optimizing deployment based on inventory size and workload. It also covers techniques for monitoring performance and tips for ensuring sufficient hardware resources.

Embed presentation

Downloaded 62 times
Extreme Performance Series: vCenter Performance Best Practices Ravi Soundararajan, VMware, Inc Dilpreet Bindra, VMware, Inc. Zhelong Pan, VMware, Inc. INF4764 #INF4764
• This presentation may contain product features that are currently under development. • This overview of new technology represents no commitment from VMware to deliver these features in any generally available product. • Features are subject to change, and must not be included in contracts, purchase orders, or sales agreements of any kind. • Technical feasibility and market demand will affect final delivery. • Pricing and packaging for any new technologies or features discussed or presented have not been determined. Disclaimer CONFIDENTIAL 2
Our Goals • Help you understand vCenter architecture • Help you understand what factors influence vCenter performance • Help you use this knowledge to guide vCenter deployment CONFIDENTIAL 3
VC 2015 Teaser: vCenter 5.5 vs. vCenter 6.0 Throughput 0 200 400 600 800 1000 1200 1400 Small Large Throughput(operations/minute) Inventory Size vCenter 5.5 vCenter 6.0 Windows 5.5 6.0 CONFIDENTIAL 4 50% higher 100% higher Substantial throughput improvements for high-churn workloads in 2015 5.5 6.0 Small inventory Large inventory
VC2015 Teaser: Web Client 5.5 GA vs. 6.0 • Sample latency improvements from 5.5 GA to 6.0 • Bottom line: please upgrade to 2015 CONFIDENTIAL 5 Operation Improvement in Latency Login 39% Action Menu 88% VM List 77% Host Summary 65% Create VM Wizard 73% DRS Cluster Settings 91%
Agenda 1 vCenter Deep Dive (+ Web Client) 2 Performance Considerations 3 Deployment Strategies 4 Concluding Remarks CONFIDENTIAL 6
In the Beginning CONFIDENTIAL 7 C# clients API clients vpxd DB vCenter server
However, this is Approximately vCenter (We Will Dissect This…) CONFIDENTIAL 8 ESXi + HostD + VPXA STORAGE NETWORK VPXD DB Web Client Server Health perfcharts vSphere Web Clients Update Manager vROps AD C# UI / API Clients Java Inv Serv … vCenter server SSO SPS Content Library
vCenter Deep Dive + Web Client
vCenter Architecture Components: High Level • Single Sign On/Identity management • Web Client • Vpxd • Inventory Service • Other Services – Content library, storage profiles, perfcharts, health services, workflow management, licensing, … CONFIDENTIAL 10
5.5 vs. 6.0 Installation CONFIDENTIAL 11 5.5: Components separately installed 6.0: Components installed together
SSO and Identity Management CONFIDENTIAL 12 AD vCenter server SSO • Global Roles and Permissions • Identity Management • Registered solutions visible to Web Client
SSO Plus the Web Client CONFIDENTIAL 13 Web Client Server vSphere Web Clients AD vCenter server SSO • 2 vCenters registered with SSO • Both visible from web client 1. Login 2. SSO Authenticates 3. After user is authenticated, user has access to all providers registered with SSO (e.g., vCenter)
Vpxd 14 VPXD DB Web Client Server vSphere Web Clients AD vCenter server SSO • Vpxd: performs main business logic • Sends tasks to appropriate hosts • Retrieves config changes from hosts • Pushes config updates to DB • Inserts stats into DB (5-minute intervals) • Satisfies queries from API clients Update Manager vROps C# UI / API Clients CONFIDENTIAL
Database CONFIDENTIAL 15 DB also performs these tasks: • Stats Rollups: VPX_HIST_STATX • 30 minutes, 2 hours, 1 day • Purging stats • Purging events (if auto-purge configured) • Purging tasks (if auto-purge configured) • TopN computation • 10 min, 30 min, 2 hours, 1 day For more info, see VSVC5234 from VMworld 2013 VPXD DB
Inventory Service CONFIDENTIAL 16 VPXD DB Web Client Server vSphere Web Clients AD Inv Serv vCenter server SSO Inventory Service • Embedded database • Cache of vCenter Inventory • Reduces load on vpxd Satisfies queries from web client Satisfies inventory queries from extensions Allows integrations with 2nd/3rd party tools
Inventory Service: Integrations with Extensions CONFIDENTIAL 17 Pros: • Reduced load on the vCenter Database • Seamless integration with other products (e.g., VROps) Cons: • Mem/IO resources • Proper sizing is key vROps
Other Services (1 of 2) CONFIDENTIAL 18 VPXD DB Web Client Server Health perfcharts vSphere Web Clients Update Manager vROps AD C# UI / API Clients Inv Serv vCenter server SSO SPS Storage Profile Service (SPS) • Storage management Health Services • Watchdogs, host and service health Perfcharts • Overview charts in web client
Other Services (2 of 2) CONFIDENTIAL 19 VPXD DB Web Client Server Health perfcharts vSphere Web Clients Update Manager vROps AD C# UI / API Clients Java Inv Serv … vCenter server SSO SPS Content Library Content Library • Shared VM repository across VCs • http://blogs.vmware.com/performance/2015/07 /efficiently-deploy-vms-vmware-vsphere- content-library.html Other • Licensing, ESX agent management, workflows, proxy
Windows vCenter Server Resource Usage: TaskManager CONFIDENTIAL 20 TaskManager sorted by memory usage Java processes: look at “User name” for some services…for others, use Windows Sysinternals Web client service Vdcs = content library SYSTEM? Use sysinternals
Sysinternals: Mapping Process to Service (1/2) CONFIDENTIAL 21 View Process Tree Find Process ID
Sysinternals: Mapping Process to Service (2/2) CONFIDENTIAL 22 Look at Service (VMwareSTS, Token service)
vCenter Appliance Resource Monitoring • vimtop CONFIDENTIAL 23 Similar to Linux top Shows services and resource usage Must enable ssh and troubleshooting mode
Agenda 1 vCenter Deep Dive (+ Web Client) 2 Performance Considerations 3 Deployment Strategies 4 Concluding Remarks CONFIDENTIAL 24
Performance Considerations
vCenter Server Performance Considerations Sufficient Resource Provisioning is Key • MINIMUM SYSTEM CONFIGURATIONS (Please check documentation): • Small setups (< 1000 VMs): 4 vCPUs, 16 GB • DB guidelines: 2 vCPU, 8GB • Medium setups (< 4000 VMs): 8 vCPUs, 24 GB • DB guidelines: 4 vCPU, 12GB – Large setups (> 4000 VMs): 8-16 vCPUs, 24-32GB • DB: 8 vCPU, 16GB • Ps. See documentation for “Tiny” configuration (test/dev) • MINIMUM – You may need more depending on your host/VM/network/storage configuration CONFIDENTIAL 26
Factors Influencing vCenter Resource Usage • ‘Churn’ – Out-of-the-box settings should be fine for ‘light’ workloads • Inventory size and host/VM configuration • # of extensions • Stats level and retention settings CONFIDENTIAL 27
Resource Usage Discussion • Examine resources under different inventory sizes and churn factors – CPU – Memory – Disk – Network CONFIDENTIAL 28
Impact of Inventory Size on CPU Usage: Idle Setup 0 10 20 30 40 50 60 70 0 200 400 600 800 1000 1200 IdleCPUUsage(%ofoneCPU) Number of ESXi Hosts CONFIDENTIAL 29 1000 hosts, 10K VMs 60% of 1 CPU used 100 hosts, 1000 VMs: < 10% of 1 CPU used Larger inventory, more CPUs needed Important: SSDs used in this experiment
Impact of Load on CPU Usage, 1000 Hosts, 10K VMs 0 200 400 600 800 1000 1200 1400 1600 1800 0 200 400 600 800 1000 1200 1400 1600 vCenterCPUUsage(%ofoneCPU) vCenter Throughput (operations per minute) CONFIDENTIAL 30 More load, more CPUs needed Important: SSDs used in this experiment 16 cores to achieve ~1600 ops/min. 7 cores to achieve ~700 ops/min.
CPU Version • Newer CPUs can impact performance • Example for large inventory: – 60% more throughput with E5-2670v3 Haswell-EP than with E7-8837 Westmere-EX CONFIDENTIAL 31
Impact of Inventory Size on Memory Usage: Idle Setup CONFIDENTIAL 32 0 2 4 6 8 10 12 14 16 18 20 0 200 400 600 800 1000 1200 Memory(nocache/buffers,GB) Number of ESXi Hosts vCenter Inventory and Memory Usage 1000 hosts: ~18GB 400 hosts: ~12GB Memory usage is strongly related to inventory size
Impact of Load on Memory Usage: 1000 hosts, 10K VMs CONFIDENTIAL 33 0 5 10 15 20 25 30 0 200 400 600 800 1000 1200 1400 1600 1800 Memory(nocache/buffers,GB) vCenter Throughput (operations per minute) vCenter Throughput and Memory Usage Ramp-up Steady state Memory varies more with inventory size than with load at scale First run of workload: ramp up Second run of workload: memory stays fixed
Top Consumers of Memory (Excluding DB) 1. Inventory service (java): inventory size 2. Vpxd (C++): inventory size • Inventory Service and vpxd often have similar usage 3. SPS (java): inventory size 4. vSphere Client (java): impacted by inventory size and extensions 5. Content Library and Transfer Service (java): concurrent transfers • Java services sensitive to heap size (especially Inventory service, vSphere client) CONFIDENTIAL 34
Inspecting Heap and Memory Usage • vCenter Appliance – Heap settings: cloudvm-ram-size –J <service name> – Memory per service: cloudvm-ram-size –l • vCenter Windows: same (make sure you are in correct directory) – chdir C:Program FilesVMwarevCenter Servervisl-integrationusrsbin – Heap settings: cloudvm-ram-size –J <service name> – Memory per service: cloudvm-ram-size –l • Changing heap size: 2 options 1. Resize VM and reboot: heaps are automatically resized (*new in 6.0*) 2. Resize individual service (No Reboot required): 1. cloudvm-ram-size.bat -C <newHeapSize> <name of service> 2. Restart CONFIDENTIAL 35
Impact of Increasing CPU and Memory • With more CPU/Memory, vCenter performance improves (Note: uses SSDs) • Example below: 16vCPU/32GB vs. 24vCPU/48GB • Bottom line: In 6.0, we scale better with more HW than in 5.5 CONFIDENTIAL 36 0 500 1000 1500 2000 2500 vCenter 5.5 vCenter 6.0 Windows vCenter 6.0 Server Appl. Throughput(operations/minute) vCenter Server Configuration 16 CPU, 32 GB RAM 24 CPU, 48 GB RAM 5.5: small delta 6.0: bigger performance improvement
Disk IO • Sample Write Breakdown: single cluster, 64H, 8K VMs, ~300 operations/minute, level 1 stats CONFIDENTIAL 37 0 2000 4000 6000 8000 10000 12000 14000 16000 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 Write KB/s swap core logs DB_data DB_trans SEAT netdump auto image IS DB transaction log DB Data + log  ~6MB/s Inventory Service ~7MB/s DB Data Total: ~3K write IOPs Database and Inventory Service largest contributors
Disk Characterization Traffic related to load, inventory size, and statistics level • Relational DB – Workload – Inventory Size – Statistics Level – Attached extensions causing vCenter Events • Inventory Service – Workload – Inventory Size • DB is involved in nearly all operations: latency is key CONFIDENTIAL 38
Disk IO as a Function of Inventory Size (Idle) 0 10 20 30 40 50 60 70 0 200 400 600 800 1000 1200 1400 0 200 400 600 800 1000 1200 DiskTransactionsperSecond DiskBandwidth(KBps) Number of ESXi Hosts Bandwidth Trans. per Sec. CONFIDENTIAL 39 1000 hosts: 1.2MB/s 400 hosts: ~500KB/s Modest disk IO when setup is idle. Note: increasing stats level increases disk IO
Disk IO as a Function of Operational Load: 1000 Hosts, 10K VMs 0 500 1000 1500 2000 2500 3000 0 5 10 15 20 25 30 35 40 0 200 400 600 800 1000 1200 1400 1600 DiskTransactionsperSecond DiskBandwidth(MBps) vCenter Throughput (operations per minute) vCenter Throughput and Disk Usage Bandwidth Trans. per Sec. CONFIDENTIAL 40 1600 VC ops/min 35 MB/s 2500 IOPs 700 VC operations/min 20MB/s 1500 IOPs With high churn, high IOPs capacity and low latency important  consider SSDs
Performance Considerations for IO • Traffic pattern – Write-mostly (inserts, purges, rollups, config updates), logs sequential, data random – SSDs critical for low latency and high bandwidth – Increase memory to ensure buffer caches capture most reads • Stats Level – Use proper stats level for your use case (more details next slide) • Physical Disk configuration: Use separate spindles for DB, Inventory service, core files, etc. – vCenter Appliance: ~10 VMDKs. Put DB and IS partitions on separate spindles – Windows: everything installed in one place. Use striped disks or SSDs under high churn • Managing DB disk growth (see next slides) • Location of DB (same node as VC, different node on same ESX host, etc.) CONFIDENTIAL 41
Impact of Stats Level on Database • Rough rules of thumb (your mileage will vary based on your setup) • Level 1 stats: per-VM and per-host aggregate stats • Level 2 stats: additional per-VM/per-host stats 4x or more stats than Level 1 depending on configuration • Level 3 stats: per-instance stats 6x or more stats than Level 2 depending on configuration • Level 4 stats: additional rollup types 1.4x more stats than Level 3 depending on configuration Recommendations: – Use the stats calculator in vCenter – Try to use higher stats levels only for temporary debugging – If the stat you want is at the wrong level, let us know – Consider vROps for more advanced stats functionality? CONFIDENTIAL 42
Latency to the DB Latency to DB important (often more so than ESX-to-VC latency) • Almost everything involves the DB… • Stats persistence • Certain UI queries • Updating configuration information • Historical queries (events, alarms, task history) • … Recommendation Place DB and vCenter close together (minimally, same geo if practical) Best case: same VM (if VM properly provisioned) Also good: different VM on same host (if host properly provisioned) Note: DB and vCenter on different hosts/VMs allows for independent sizing and tuning For more info, see VSVC5234 from VMworld 2013 CONFIDENTIAL 43
DB Performance Considerations (2 of 2) • Manage database disk growth – Majority of DB data is “SEAT” data (Stats, events, alarms, tasks): 80-85% (10s of GBs or more in big setups) – Inventory data: 10-15% of data (usually < 10GB for large inventories) – Choose stats levels wisely to avoid excessive growth – Utilize automatic purging of event/task tables if possible • Recompute DB stats on highly-volatile tables (at least once a day) – VPX_TOPN* 44 For more info, see VSVC5234 from VMworld 2013 CONFIDENTIAL
Network • Network carries configuration changes and statistics data • Network traffic is bursty due to periodic stats traffic: CONFIDENTIAL 45 0 10 20 30 40 50 60 0 5 10 15 20 NetworkUsage(Mbps) Time (minutes) vCenter Network Usage
Network Bandwidth Consumed as a Function of Load 0 50 100 150 200 250 0 200 400 600 800 1000 1200 NetworkUsage(Mbps) vCenter Throughput (operations per minute) vCenter Throughput and Network Usage Sent Received Total CONFIDENTIAL 46 Bandwidth requirements modest, but latency is key
Network • Link from vCenter to hosts and to DB impacts performance: latency more so than throughput: • Example from our lab: CONFIDENTIAL 47 8.23 8.85 9.23 10.24 12.74 26.39 0 5 10 15 20 25 30 Baseline 10Gbps, 1ms 100Mbps, 20 ms 10Mbps, 50 ms 1.5 Mbps, 100 ms 1.5 Mbps, 500 ms PowerOnLatency(secs) Median PowerOn Latency (secs)
Other Performance Considerations • Concurrency • API • Extensions • Windows vs. Linux CONFIDENTIAL 48
How Many Concurrent Operations Can I Perform? (1 of 2) • vCenter hard limits – 640 concurrent operations before incoming requests are queued – 2000 concurrent sessions (incoming requests plus remote console sessions) • Per-host or per-datastore limits – A host can perform up to 8 provisioning operations at once (provisioning = clone, VMotion, relocate) – If host is source and destination, host can only do 4 operations at once – A datastore can perform up to 128 VMotions at once – A datastore can perform up to 8 Storage VMotions at once – Limits can be changed, but changes are not officially supported • NIC configuration: 10Gb vs. 1Gb – 10Gb NIC allows a host to do 2x more VMotions at a time than 1Gb NIC CONFIDENTIAL 49
vCenter Concurrency (2 of 2) • Clone VM from host A to host B • Each host can participate in 7 other provisioning operations • Clone VM from host A to host A • Host A can only participate in 6 more operations CONFIDENTIAL 50 vCenter Host A VM 1 Host B VM 2 Cost to A: 1 Cost to B: 1 vCenter Host A VM 1 VM 2 Cost to A: 2 Do not use a single host as the source of all clones (i.e., spread out templates)  Better disk performance and better concurrency
API Performance Considerations: An Example • Example of a good vs. bad client in PowerCLI • PowerCLI: – Simple to use, but involves client-side filtering – Example: Get-VM gets all VMs from server, filters list @ client • $vmList = Get-VM –name “vm1”,”vm2”,”vm3”,”vm4” • Good: 1 server call, client throws away all but vm1,vm2,vm3,vm4 $nameList = “vm1”,”vm2”,”vm3”,”vm4” foreach ($name in $nameList) { Get-VM $name } Bad: 4 server calls, gets all VMs 4 times…excess client/server work For more info, see VSVC5234 from VMworld 2013 CONFIDENTIAL 51
Impact of Solutions • vROps – Increased network traffic to get stats from hosts – Increased CPU usage on hosts (to retrieve stats) – Increased CPU usage on vCenter to serialize data – Example at scale (1K hosts, 10K powered-on VMs) • Data transmitted without vROps: 0.21 Mbps • Data transmitted with vRops: 1.4 Mbps (7x change) • Data received without vROps: 0.39 Mbps • Data received with vROps: 1.35 MBps (4x change) – vCenter throughput reduced (example: 19% in a high-churn use case) • NSX – 400MB extra memory needed for vSphere web client service heap • vRA: increased concurrent workflows increases CPU on vCenter and network traffic CONFIDENTIAL 52 vROps vCenter ESX ESX ESX ESX ESXVC stats DB VC stats vROps vROps
Windows vs. vCenter Appliance • The vCenter appliance with embedded DB can support full vCenter Limits • The vCenter appliance and Windows have similar performance CONFIDENTIAL 53 0 200 400 600 800 1000 1200 1400 1600 Small Large Throughput(operations/minute) Inventory Size vCenter 6.0 Windows vCenter 6.0 Server Appl.
Web Client Performance Concerns
Web Client: Performance Considerations • If possible, browser machine should have 2 CPUs, 4GB • Faster CPUs help (on both client and server side) • Use browser in same geo as application server (RDP to a local machine?) • Make sure application server has sufficient heap size (may need to increase if plugins are installed) • Make sure Inventory Service has sufficient heap size CONFIDENTIAL 55
Agenda 1 vCenter Deep Dive (+ Web Client) 2 Performance Considerations 3 Deployment Strategies 4 Concluding Remarks CONFIDENTIAL 56
Deployment Strategies
Deployment Options • With increased scaling, our goal is that inventory size and churn are NOT the reasons you need to use multiple VCs • Possibilities – Fully Embedded – Embedded + external DB – External Platform Services Controller (PSC) – Multiple VCs with External PSCs, high-availability – Multiple VCs with External PSCs, multi-site • SSO on PSC replaces linked mode – Works on both Windows and Appliance – Allows global sharing of roles, permissions, tags, and licenses CONFIDENTIAL 58
Fully Embedded or Embedded with External DB • Good for most single vCenter configurations 59 DB AD VPXD Web Client Server Health perfcharts Java Inv Serv … SSO SPS Content Library AD VPXD DB Web Client Server Health perfcharts Java Inv Serv … SSO SPS Content Library
External PSC • Good if you anticipate multiple vCenters CONFIDENTIAL 60 AD VPXD DB Web Client Server Health perfcharts Java Inv Serv … SSO SPS Content Library PSC 2 vCPU, 2GB
Multiple vCenters: Use Cases • Increased Scale – Operations/s? • For some, 50 ops/min is where they want more vCenters – Large number of hosts/VMs? • For some, the single VC “sweet spot” is 200H/2000VMs • Business Considerations – Finance vs. Engineering – PCI-compliant vs. non-PCI-compliant racks – Server vs. Desktop Workloads • Multiple Geographies – Ok with single vCenter managing remote hosts? – vCenter per site or per group of sites? CONFIDENTIAL 61 1. Decide on one or more vCenters 2. Single vs. Multiple SSO sites?
Multiple vCenters, Single PSC Pro: Single Pane of Glass Pro: Shared Licenses, roles, permissions Con: Single point of failure (PSC) VPXD DB Web Client Server Health perfcharts Java Inv Serv … SSO SPS Content Library AD VPXD DB Web Client Server Health perfcharts Java Inv Serv … SSO SPS Content Library PSC CONFIDENTIAL 62
Multiple vCenters, Multiple PSCs with HA and Load Balancer Add a load balancer in front of PSCs 8 VCs per PSC pair Roles/Privileges/License replication VPXD DB Web Client Server Health perfcharts Java Inv Serv … SSO SPS Content Library AD VPXD DB Web Client Server Health perfcharts Java Inv Serv … SSO SPS Content Library PSC PSC LB CONFIDENTIAL 63
Multiple vCenters, Multi-site Mode with Multiple PSCs • Roles/Privileges/License replication across sites • No HA: must add LB for this VPXD DB Web Client Server Health perfcharts Java Inv Serv … SSO SPS Content Library AD VPXD DB Web Client Server Health perfcharts Java Inv Serv … SSO SPS Content Library PSC PSC CONFIDENTIAL 64
PSC: Performance Considerations • Default size (2 vCPU, 2GB) should be fine • Tag performance can be impacted by slow link between vCenter and PSC • Login may be slower for SSO vs. Standalone (contacting multiple vCenters) • Search is slower for external SSO (contacting multiple vCenters) • One slow vCenter may slow down Login/Search CONFIDENTIAL 65
Agenda 1 vCenter Deep Dive (+ Web Client) 2 Performance Considerations 3 Deployment Strategies 4 Concluding Remarks CONFIDENTIAL 66
Conclusions • vCenter 6.0 vastly improved in scalability and performance over 5.5 – Our goal: performance is not the reason you need multiple vCenters – Use PSCs for availability, not performance • For best performance, vCenter needs sufficient resources – CPU: scales with inventory size and churn – Memory: scales with inventory size – IO: scales with inventory size, churn, and stats level – Network: low-latency between VC and DB recommended • With 2nd or 3rd party solutions, resource requirements of vCenter will likely increase to manage the same inventory size • Use the appliance! – Windows vCenter and Linux appliance: similar performance, same scale limits – PSC provides sharing of roles/permissions/licenses CONFIDENTIAL 67
Extreme Performance Series – Break Out Sessions • INF4764 vCenter Performance Best Practices • INF5701 vSphere Compute & Memory • INF4936 Insight Into vSphere 6 vMotion • VAPP4639 Best Practices for Performance Tuning of Virtualized Telco and NFV • INF4853 Docker Containers on vSphere • VAPP5724 High Performance Panel - No App Left Behind • VAPP5165 Monster VM Database Performance • STO4949 Virtual SAN Performance Deep-Dive • EUC5802 Horizon View 6.x Performance and Best Practices • VAPP6537-GD Maximize Performance on vSphere 6 CONFIDENTIAL 68
Performance Hands On Labs • HOL-SDC-1604 vSphere Performance Optimization This Lab covers vSphere performance best practices and various performance related features available in vSphere 6. • SPL-CHG-1695 vSphere 6 Challenge Lab The vSphere 6 Challenge asks you to put on your thinking cap to save the day! Each module places you in a different fictional scenario to fix common vSphere operational and performance problems. • ELW-SDC-1604 vSphere Performance Optimization This expert led workshop will take you though the vSphere 6 performance best practices hands on lab with additional support and discussion. CONFIDENTIAL 69
Extreme Performance Series: vCenter Performance Best Practices Ravi Soundararajan, VMware, Inc INF4764 #INF4764

More Related Content

PPTX
VMworld 2015: Explaining Advanced Virtual Volumes Configurations
byVMworld
 
PPTX
VMworld 2015: Advanced SQL Server on vSphere
byVMworld
 
PPTX
VMworld 2015: Troubleshooting for vSphere 6
byVMworld
 
PPTX
VMworld 2015: Just Because You COULD, Doesn’t Mean You SHOULD – vSphere 6.0 A...
byVMworld
 
PPTX
VMworld 2015: vSphere Web Client- Yesterday, Today, and Tomorrow
byVMworld
 
PPTX
VMworld 2016: Enforcing a vSphere Cluster Design with PowerCLI Automation
byVMworld
 
PPTX
VMworld 2015: Virtualize Active Directory, the Right Way!
byVMworld
 
PPTX
VMworld 2015: Managing vSphere 6 Deployments and Upgrades
byVMworld
 
VMworld 2015: Explaining Advanced Virtual Volumes Configurations
byVMworld
 
VMworld 2015: Advanced SQL Server on vSphere
byVMworld
 
VMworld 2015: Troubleshooting for vSphere 6
byVMworld
 
VMworld 2015: Just Because You COULD, Doesn’t Mean You SHOULD – vSphere 6.0 A...
byVMworld
 
VMworld 2015: vSphere Web Client- Yesterday, Today, and Tomorrow
byVMworld
 
VMworld 2016: Enforcing a vSphere Cluster Design with PowerCLI Automation
byVMworld
 
VMworld 2015: Virtualize Active Directory, the Right Way!
byVMworld
 
VMworld 2015: Managing vSphere 6 Deployments and Upgrades
byVMworld
 

What's hot

PPTX
VMworld 2015: Site Recovery Manager and Policy Based DR Deep Dive with Engine...
byVMworld
 
PPTX
VMworld 2015: Automating Everything VMware with PowerCLI- Deep Dive
byVMworld
 
PPTX
VMworld 2015: Networking Virtual SAN's Backbone
byVMworld
 
PPTX
VMworld 2015: Managing Users: A Deep Dive into VMware User Environment Manager
byVMworld
 
PPTX
VMworld 2016: Getting Started with PowerShell and PowerCLI for Your VMware En...
byVMworld
 
PPTX
VMworld 2016: Virtual Volumes Technical Deep Dive
byVMworld
 
PPTX
VMworld 2015: VMware vSphere Certificate Management for Mere Mortals
byVMworld
 
PPTX
VMworld 2016: Troubleshooting 101 for Horizon
byVMworld
 
PPTX
VMworld 2015: Extreme Performance Series - vSphere Compute & Memory
byVMworld
 
PPTX
VMworld 2015: What's New in vSphere?
byVMworld
 
PPTX
VMware 2015: Next Horizon for Cloud Networking and Security
byVMworld
 
PPTX
VMworld 2015: Take Virtualization to the Next Level vSphere with Operations M...
byVMworld
 
PPTX
HBC9363 Virtualization 2.0 How the Cloud is Evolving the Modern Data Center
bydavehill99
 
PPTX
HBC8292 vCloud Air Recovery as a Service (RaaS) Deep Dive
bydavehill99
 
PPTX
VMworld 2015: Monitoring and Managing Applications with vRealize Operations 6...
byVMworld
 
PPTX
VMworld 2015: Virtual Volumes Technical Deep Dive
byVMworld
 
PPTX
VMworld 2016 - INF8036 - enforcing a vSphere cluster design with powercli aut...
byDuncan Epping
 
PPTX
VMworld 2015: The Future of Software- Defined Storage- What Does it Look Like...
byVMworld
 
PPTX
VMworld 2016: The KISS of vRealize Operations!
byVMworld
 
PPTX
STO7535 Virtual SAN Proof of Concept - VMworld 2016
byCormac Hogan
 
VMworld 2015: Site Recovery Manager and Policy Based DR Deep Dive with Engine...
byVMworld
 
VMworld 2015: Automating Everything VMware with PowerCLI- Deep Dive
byVMworld
 
VMworld 2015: Networking Virtual SAN's Backbone
byVMworld
 
VMworld 2015: Managing Users: A Deep Dive into VMware User Environment Manager
byVMworld
 
VMworld 2016: Getting Started with PowerShell and PowerCLI for Your VMware En...
byVMworld
 
VMworld 2016: Virtual Volumes Technical Deep Dive
byVMworld
 
VMworld 2015: VMware vSphere Certificate Management for Mere Mortals
byVMworld
 
VMworld 2016: Troubleshooting 101 for Horizon
byVMworld
 
VMworld 2015: Extreme Performance Series - vSphere Compute & Memory
byVMworld
 
VMworld 2015: What's New in vSphere?
byVMworld
 
VMware 2015: Next Horizon for Cloud Networking and Security
byVMworld
 
VMworld 2015: Take Virtualization to the Next Level vSphere with Operations M...
byVMworld
 
HBC9363 Virtualization 2.0 How the Cloud is Evolving the Modern Data Center
bydavehill99
 
HBC8292 vCloud Air Recovery as a Service (RaaS) Deep Dive
bydavehill99
 
VMworld 2015: Monitoring and Managing Applications with vRealize Operations 6...
byVMworld
 
VMworld 2015: Virtual Volumes Technical Deep Dive
byVMworld
 
VMworld 2016 - INF8036 - enforcing a vSphere cluster design with powercli aut...
byDuncan Epping
 
VMworld 2015: The Future of Software- Defined Storage- What Does it Look Like...
byVMworld
 
VMworld 2016: The KISS of vRealize Operations!
byVMworld
 
STO7535 Virtual SAN Proof of Concept - VMworld 2016
byCormac Hogan
 

Viewers also liked

PDF
VMworld 2014: vCenter Server Architecture and Deployment Deep Dive
byVMworld
 
PPTX
VMworld 2016: vSphere 6.x Host Resource Deep Dive
byVMworld
 
PPTX
vCloud Connector: гибридное облако на практике для vSphere
byitgrad
 
PPTX
What’s new in vSphere 5 and vCenter Server Heartbeat – Customer Presentation
bySuministros Obras y Sistemas
 
PDF
Новые возможности VMware vCenter Site Recovery Manager v5.0
byareconster
 
PPTX
VMworld 2015: How To Troubleshoot Using vRealize Operations Manager (Deep Liv...
byVMworld
 
PDF
VMworld 2013: vCenter Deep Dive
byVMworld
 
PPTX
VMworld 2015: Building a Business Case for Virtual SAN
byVMworld
 
PPTX
VMworld 2016: Ask the vCenter Server Exerts Panel
byVMworld
 
PPTX
VMworld 2016: Advances in Remote Display Protocol Technology with VMware Blas...
byVMworld
 
PPTX
VMworld 2015: vRealize Operations Insight: Manage vSphere and Your Entire Dat...
byVMworld
 
PPTX
Interface Definition Language
byInstitut supérieur des études technologiques de Radès
 
PDF
Introduction - vSphere 5 High Availability (HA)
byEric Sloof
 
PPTX
VMworld 2016: Migrating from a hardware based firewall to NSX to improve perf...
byVMworld
 
PPTX
VMworld 2016: What's New with Horizon 7
byVMworld
 
PPTX
VMworld 2016: Advanced Network Services with NSX
byVMworld
 
VMworld 2014: vCenter Server Architecture and Deployment Deep Dive
byVMworld
 
VMworld 2016: vSphere 6.x Host Resource Deep Dive
byVMworld
 
vCloud Connector: гибридное облако на практике для vSphere
byitgrad
 
What’s new in vSphere 5 and vCenter Server Heartbeat – Customer Presentation
bySuministros Obras y Sistemas
 
Новые возможности VMware vCenter Site Recovery Manager v5.0
byareconster
 
VMworld 2015: How To Troubleshoot Using vRealize Operations Manager (Deep Liv...
byVMworld
 
VMworld 2013: vCenter Deep Dive
byVMworld
 
VMworld 2015: Building a Business Case for Virtual SAN
byVMworld
 
VMworld 2016: Ask the vCenter Server Exerts Panel
byVMworld
 
VMworld 2016: Advances in Remote Display Protocol Technology with VMware Blas...
byVMworld
 
VMworld 2015: vRealize Operations Insight: Manage vSphere and Your Entire Dat...
byVMworld
 
Interface Definition Language
byInstitut supérieur des études technologiques de Radès
 
Introduction - vSphere 5 High Availability (HA)
byEric Sloof
 
VMworld 2016: Migrating from a hardware based firewall to NSX to improve perf...
byVMworld
 
VMworld 2016: What's New with Horizon 7
byVMworld
 
VMworld 2016: Advanced Network Services with NSX
byVMworld
 

Similar to VMworld 2015: Extreme Performance Series - vCenter Performance Best Practices

PDF
Dell EMC VxRAIL Appliance based on VMware SDS
byMarketingArrowECS_CZ
 
PDF
VMworld 2013: Extreme Performance Series: vCenter of the Universe
byVMworld
 
PDF
Nordic VMUG User Conference 2014 - Design VMware vCenter Server
byAndrea Mauro
 
PDF
VMworld 2014: What's New in vSphere
byVMworld
 
PDF
VMworld 2014: What's New in vSphere
byVMworld
 
PDF
VMworld 2013: vSphere Upgrade Series Part 1: vCenter Server
byVMworld
 
PDF
Presentation architecting a cloud infrastructure
byxKinAnx
 
PDF
Presentation architecting a cloud infrastructure
bysolarisyourep
 
PPTX
V center operations standard presentation
bysolarisyourep
 
PDF
VMworld 2014: Extreme Performance Series
byVMworld
 
PDF
VMUGIT UC 2013 - 04 Duncan Epping
byVMUG IT
 
PPTX
Whats new v sphere 6
byshixi wang
 
PDF
VMworld Europe 2014: What's New in vSphere?
byVMworld
 
PDF
Exploring VMware APIs by Preetham Gopalaswamy
byAlan Renouf
 
PDF
VMworld 2013: vCloud Powered HPC is Better and Outperforming Physical
byVMworld
 
PPTX
Track 1 Virtualizing Critical Applications with VMWARE VISPHERE by Roshan Shetty
byEMC Forum India
 
PDF
VMware vSphere Version Comparison 4.0 to 6.5
bySabir Hussain
 
PPT
Capito Ardoe House VMWare Presentation
byCapito Livingstone
 
PPTX
PCoIP and More - VMware View - Performance and Best Practices
byVMwarePR
 
PPTX
Virtualising Tier 1 Apps
byIwan Rahabok
 
Dell EMC VxRAIL Appliance based on VMware SDS
byMarketingArrowECS_CZ
 
VMworld 2013: Extreme Performance Series: vCenter of the Universe
byVMworld
 
Nordic VMUG User Conference 2014 - Design VMware vCenter Server
byAndrea Mauro
 
VMworld 2014: What's New in vSphere
byVMworld
 
VMworld 2014: What's New in vSphere
byVMworld
 
VMworld 2013: vSphere Upgrade Series Part 1: vCenter Server
byVMworld
 
Presentation architecting a cloud infrastructure
byxKinAnx
 
Presentation architecting a cloud infrastructure
bysolarisyourep
 
V center operations standard presentation
bysolarisyourep
 
VMworld 2014: Extreme Performance Series
byVMworld
 
VMUGIT UC 2013 - 04 Duncan Epping
byVMUG IT
 
Whats new v sphere 6
byshixi wang
 
VMworld Europe 2014: What's New in vSphere?
byVMworld
 
Exploring VMware APIs by Preetham Gopalaswamy
byAlan Renouf
 
VMworld 2013: vCloud Powered HPC is Better and Outperforming Physical
byVMworld
 
Track 1 Virtualizing Critical Applications with VMWARE VISPHERE by Roshan Shetty
byEMC Forum India
 
VMware vSphere Version Comparison 4.0 to 6.5
bySabir Hussain
 
Capito Ardoe House VMWare Presentation
byCapito Livingstone
 
PCoIP and More - VMware View - Performance and Best Practices
byVMwarePR
 
Virtualising Tier 1 Apps
byIwan Rahabok
 

More from VMworld

PPTX
VMworld 2016: How to Deploy VMware NSX with Cisco Infrastructure
byVMworld
 
PPTX
VMworld 2016: Virtualize Active Directory, the Right Way!
byVMworld
 
PPTX
VMworld 2015: The Best SDDC!
byVMworld
 
PPTX
VMworld 2015: Conversation with the VMware CIO Suggestions on being an IT Leader
byVMworld
 
PPTX
VMworld 2015: VMware NSX Deep Dive
byVMworld
 
PPTX
VMworld 2015: vSphere Distributed Switch 6 –Technical Deep Dive
byVMworld
 
PPTX
VMworld 2015: Introducing Application Self service with Networking and Security
byVMworld
 
VMworld 2016: How to Deploy VMware NSX with Cisco Infrastructure
byVMworld
 
VMworld 2016: Virtualize Active Directory, the Right Way!
byVMworld
 
VMworld 2015: The Best SDDC!
byVMworld
 
VMworld 2015: Conversation with the VMware CIO Suggestions on being an IT Leader
byVMworld
 
VMworld 2015: VMware NSX Deep Dive
byVMworld
 
VMworld 2015: vSphere Distributed Switch 6 –Technical Deep Dive
byVMworld
 
VMworld 2015: Introducing Application Self service with Networking and Security
byVMworld
 

Recently uploaded

PDF
Why Many Smart Device Platforms Fail to Scale?
byPromeraki Developments
 
PDF
Automated Governance for FME Flow: Smarter Admin at Scale
bySafe Software
 
PPTX
Tech Trends 2026: AI Agents, Quantum Computing, Robotics & Cybersecurity
byridwansassman
 
PDF
Explaining the flow of purpose-specific BOM
byakipii ogaoga
 
PPTX
Beyond the Algorithm: Designing Human-Centric Public Service with AI
byAlan Dix
 
PPTX
Celonis Optimizing your future with process
bydevjeremyappleyard
 
PDF
final.pdf
byomarbishtawi04
 
PDF
Session 1/5: Enhancing Automation with Screenplay & Business Rules
bysuhanisingh58689
 
PPTX
Workflow and decision Automation with Flowable
bychakib ch
 
PPTX
Do You Control the AI, or Does the AI Control You?
bymedhateltoukhy
 
PDF
Oracle Cloud Infrastructure 2025 Architect Professional (1Z0-1127-25) Master ...
byExamCert App
 
PPTX
Working Session — Build a Document Understanding Automation Using an OOTB ML ...
byDianaGray10
 
PPTX
Automation Without Apprentices: How AI Challenges the Open Source Way
byIcinga
 
PDF
How does MES(Manufacturing Execution System) work?
byakipii ogaoga
 
PDF
Escape from the Forbidden Zone: Smuggling green and inclusive tech past the g...
byBookNet Canada
 
PPTX
Computer architecture, Computer architecture ,Computer architecture
bywaheedqazi123
 
PDF
The Emergence of Empathic XR: AWE Asia 2026 keynote
byMark Billinghurst
 
PDF
Python Automation in Action: Real-World Use Cases and Practical Implementation
byDenys Smirnov
 
PDF
Writing GPU-Ready AI Models in Pure Java with Babylon
byAna-Maria Mihalceanu
 
PDF
February 2026 Patch Tuesday hosted by Chris Goettl and Todd Schell
byIvanti
 
Why Many Smart Device Platforms Fail to Scale?
byPromeraki Developments
 
Automated Governance for FME Flow: Smarter Admin at Scale
bySafe Software
 
Tech Trends 2026: AI Agents, Quantum Computing, Robotics & Cybersecurity
byridwansassman
 
Explaining the flow of purpose-specific BOM
byakipii ogaoga
 
Beyond the Algorithm: Designing Human-Centric Public Service with AI
byAlan Dix
 
Celonis Optimizing your future with process
bydevjeremyappleyard
 
final.pdf
byomarbishtawi04
 
Session 1/5: Enhancing Automation with Screenplay & Business Rules
bysuhanisingh58689
 
Workflow and decision Automation with Flowable
bychakib ch
 
Do You Control the AI, or Does the AI Control You?
bymedhateltoukhy
 
Oracle Cloud Infrastructure 2025 Architect Professional (1Z0-1127-25) Master ...
byExamCert App
 
Working Session — Build a Document Understanding Automation Using an OOTB ML ...
byDianaGray10
 
Automation Without Apprentices: How AI Challenges the Open Source Way
byIcinga
 
How does MES(Manufacturing Execution System) work?
byakipii ogaoga
 
Escape from the Forbidden Zone: Smuggling green and inclusive tech past the g...
byBookNet Canada
 
Computer architecture, Computer architecture ,Computer architecture
bywaheedqazi123
 
The Emergence of Empathic XR: AWE Asia 2026 keynote
byMark Billinghurst
 
Python Automation in Action: Real-World Use Cases and Practical Implementation
byDenys Smirnov
 
Writing GPU-Ready AI Models in Pure Java with Babylon
byAna-Maria Mihalceanu
 
February 2026 Patch Tuesday hosted by Chris Goettl and Todd Schell
byIvanti
 

VMworld 2015: Extreme Performance Series - vCenter Performance Best Practices

  • 1.
    Extreme Performance Series: vCenterPerformance Best Practices Ravi Soundararajan, VMware, Inc Dilpreet Bindra, VMware, Inc. Zhelong Pan, VMware, Inc. INF4764 #INF4764
  • 2.
    • This presentationmay contain product features that are currently under development. • This overview of new technology represents no commitment from VMware to deliver these features in any generally available product. • Features are subject to change, and must not be included in contracts, purchase orders, or sales agreements of any kind. • Technical feasibility and market demand will affect final delivery. • Pricing and packaging for any new technologies or features discussed or presented have not been determined. Disclaimer CONFIDENTIAL 2
  • 3.
    Our Goals • Helpyou understand vCenter architecture • Help you understand what factors influence vCenter performance • Help you use this knowledge to guide vCenter deployment CONFIDENTIAL 3
  • 4.
    VC 2015 Teaser:vCenter 5.5 vs. vCenter 6.0 Throughput 0 200 400 600 800 1000 1200 1400 Small Large Throughput(operations/minute) Inventory Size vCenter 5.5 vCenter 6.0 Windows 5.5 6.0 CONFIDENTIAL 4 50% higher 100% higher Substantial throughput improvements for high-churn workloads in 2015 5.5 6.0 Small inventory Large inventory
  • 5.
    VC2015 Teaser: WebClient 5.5 GA vs. 6.0 • Sample latency improvements from 5.5 GA to 6.0 • Bottom line: please upgrade to 2015 CONFIDENTIAL 5 Operation Improvement in Latency Login 39% Action Menu 88% VM List 77% Host Summary 65% Create VM Wizard 73% DRS Cluster Settings 91%
  • 6.
    Agenda 1 vCenter DeepDive (+ Web Client) 2 Performance Considerations 3 Deployment Strategies 4 Concluding Remarks CONFIDENTIAL 6
  • 7.
    In the Beginning CONFIDENTIAL7 C# clients API clients vpxd DB vCenter server
  • 8.
    However, this isApproximately vCenter (We Will Dissect This…) CONFIDENTIAL 8 ESXi + HostD + VPXA STORAGE NETWORK VPXD DB Web Client Server Health perfcharts vSphere Web Clients Update Manager vROps AD C# UI / API Clients Java Inv Serv … vCenter server SSO SPS Content Library
  • 9.
  • 10.
    vCenter Architecture Components:High Level • Single Sign On/Identity management • Web Client • Vpxd • Inventory Service • Other Services – Content library, storage profiles, perfcharts, health services, workflow management, licensing, … CONFIDENTIAL 10
  • 11.
    5.5 vs. 6.0Installation CONFIDENTIAL 11 5.5: Components separately installed 6.0: Components installed together
  • 12.
    SSO and IdentityManagement CONFIDENTIAL 12 AD vCenter server SSO • Global Roles and Permissions • Identity Management • Registered solutions visible to Web Client
  • 13.
    SSO Plus theWeb Client CONFIDENTIAL 13 Web Client Server vSphere Web Clients AD vCenter server SSO • 2 vCenters registered with SSO • Both visible from web client 1. Login 2. SSO Authenticates 3. After user is authenticated, user has access to all providers registered with SSO (e.g., vCenter)
  • 14.
    Vpxd 14 VPXD DB Web Client Server vSphere Web Clients AD vCenter server SSO •Vpxd: performs main business logic • Sends tasks to appropriate hosts • Retrieves config changes from hosts • Pushes config updates to DB • Inserts stats into DB (5-minute intervals) • Satisfies queries from API clients Update Manager vROps C# UI / API Clients CONFIDENTIAL
  • 15.
    Database CONFIDENTIAL 15 DB alsoperforms these tasks: • Stats Rollups: VPX_HIST_STATX • 30 minutes, 2 hours, 1 day • Purging stats • Purging events (if auto-purge configured) • Purging tasks (if auto-purge configured) • TopN computation • 10 min, 30 min, 2 hours, 1 day For more info, see VSVC5234 from VMworld 2013 VPXD DB
  • 16.
    Inventory Service CONFIDENTIAL 16 VPXD DB WebClient Server vSphere Web Clients AD Inv Serv vCenter server SSO Inventory Service • Embedded database • Cache of vCenter Inventory • Reduces load on vpxd Satisfies queries from web client Satisfies inventory queries from extensions Allows integrations with 2nd/3rd party tools
  • 17.
    Inventory Service: Integrationswith Extensions CONFIDENTIAL 17 Pros: • Reduced load on the vCenter Database • Seamless integration with other products (e.g., VROps) Cons: • Mem/IO resources • Proper sizing is key vROps
  • 18.
    Other Services (1of 2) CONFIDENTIAL 18 VPXD DB Web Client Server Health perfcharts vSphere Web Clients Update Manager vROps AD C# UI / API Clients Inv Serv vCenter server SSO SPS Storage Profile Service (SPS) • Storage management Health Services • Watchdogs, host and service health Perfcharts • Overview charts in web client
  • 19.
    Other Services (2of 2) CONFIDENTIAL 19 VPXD DB Web Client Server Health perfcharts vSphere Web Clients Update Manager vROps AD C# UI / API Clients Java Inv Serv … vCenter server SSO SPS Content Library Content Library • Shared VM repository across VCs • http://blogs.vmware.com/performance/2015/07 /efficiently-deploy-vms-vmware-vsphere- content-library.html Other • Licensing, ESX agent management, workflows, proxy
  • 20.
    Windows vCenter ServerResource Usage: TaskManager CONFIDENTIAL 20 TaskManager sorted by memory usage Java processes: look at “User name” for some services…for others, use Windows Sysinternals Web client service Vdcs = content library SYSTEM? Use sysinternals
  • 21.
    Sysinternals: Mapping Processto Service (1/2) CONFIDENTIAL 21 View Process Tree Find Process ID
  • 22.
    Sysinternals: Mapping Processto Service (2/2) CONFIDENTIAL 22 Look at Service (VMwareSTS, Token service)
  • 23.
    vCenter Appliance ResourceMonitoring • vimtop CONFIDENTIAL 23 Similar to Linux top Shows services and resource usage Must enable ssh and troubleshooting mode
  • 24.
    Agenda 1 vCenter DeepDive (+ Web Client) 2 Performance Considerations 3 Deployment Strategies 4 Concluding Remarks CONFIDENTIAL 24
  • 25.
  • 26.
    vCenter Server PerformanceConsiderations Sufficient Resource Provisioning is Key • MINIMUM SYSTEM CONFIGURATIONS (Please check documentation): • Small setups (< 1000 VMs): 4 vCPUs, 16 GB • DB guidelines: 2 vCPU, 8GB • Medium setups (< 4000 VMs): 8 vCPUs, 24 GB • DB guidelines: 4 vCPU, 12GB – Large setups (> 4000 VMs): 8-16 vCPUs, 24-32GB • DB: 8 vCPU, 16GB • Ps. See documentation for “Tiny” configuration (test/dev) • MINIMUM – You may need more depending on your host/VM/network/storage configuration CONFIDENTIAL 26
  • 27.
    Factors Influencing vCenterResource Usage • ‘Churn’ – Out-of-the-box settings should be fine for ‘light’ workloads • Inventory size and host/VM configuration • # of extensions • Stats level and retention settings CONFIDENTIAL 27
  • 28.
    Resource Usage Discussion •Examine resources under different inventory sizes and churn factors – CPU – Memory – Disk – Network CONFIDENTIAL 28
  • 29.
    Impact of InventorySize on CPU Usage: Idle Setup 0 10 20 30 40 50 60 70 0 200 400 600 800 1000 1200 IdleCPUUsage(%ofoneCPU) Number of ESXi Hosts CONFIDENTIAL 29 1000 hosts, 10K VMs 60% of 1 CPU used 100 hosts, 1000 VMs: < 10% of 1 CPU used Larger inventory, more CPUs needed Important: SSDs used in this experiment
  • 30.
    Impact of Loadon CPU Usage, 1000 Hosts, 10K VMs 0 200 400 600 800 1000 1200 1400 1600 1800 0 200 400 600 800 1000 1200 1400 1600 vCenterCPUUsage(%ofoneCPU) vCenter Throughput (operations per minute) CONFIDENTIAL 30 More load, more CPUs needed Important: SSDs used in this experiment 16 cores to achieve ~1600 ops/min. 7 cores to achieve ~700 ops/min.
  • 31.
    CPU Version • NewerCPUs can impact performance • Example for large inventory: – 60% more throughput with E5-2670v3 Haswell-EP than with E7-8837 Westmere-EX CONFIDENTIAL 31
  • 32.
    Impact of InventorySize on Memory Usage: Idle Setup CONFIDENTIAL 32 0 2 4 6 8 10 12 14 16 18 20 0 200 400 600 800 1000 1200 Memory(nocache/buffers,GB) Number of ESXi Hosts vCenter Inventory and Memory Usage 1000 hosts: ~18GB 400 hosts: ~12GB Memory usage is strongly related to inventory size
  • 33.
    Impact of Loadon Memory Usage: 1000 hosts, 10K VMs CONFIDENTIAL 33 0 5 10 15 20 25 30 0 200 400 600 800 1000 1200 1400 1600 1800 Memory(nocache/buffers,GB) vCenter Throughput (operations per minute) vCenter Throughput and Memory Usage Ramp-up Steady state Memory varies more with inventory size than with load at scale First run of workload: ramp up Second run of workload: memory stays fixed
  • 34.
    Top Consumers ofMemory (Excluding DB) 1. Inventory service (java): inventory size 2. Vpxd (C++): inventory size • Inventory Service and vpxd often have similar usage 3. SPS (java): inventory size 4. vSphere Client (java): impacted by inventory size and extensions 5. Content Library and Transfer Service (java): concurrent transfers • Java services sensitive to heap size (especially Inventory service, vSphere client) CONFIDENTIAL 34
  • 35.
    Inspecting Heap andMemory Usage • vCenter Appliance – Heap settings: cloudvm-ram-size –J <service name> – Memory per service: cloudvm-ram-size –l • vCenter Windows: same (make sure you are in correct directory) – chdir C:Program FilesVMwarevCenter Servervisl-integrationusrsbin – Heap settings: cloudvm-ram-size –J <service name> – Memory per service: cloudvm-ram-size –l • Changing heap size: 2 options 1. Resize VM and reboot: heaps are automatically resized (*new in 6.0*) 2. Resize individual service (No Reboot required): 1. cloudvm-ram-size.bat -C <newHeapSize> <name of service> 2. Restart CONFIDENTIAL 35
  • 36.
    Impact of IncreasingCPU and Memory • With more CPU/Memory, vCenter performance improves (Note: uses SSDs) • Example below: 16vCPU/32GB vs. 24vCPU/48GB • Bottom line: In 6.0, we scale better with more HW than in 5.5 CONFIDENTIAL 36 0 500 1000 1500 2000 2500 vCenter 5.5 vCenter 6.0 Windows vCenter 6.0 Server Appl. Throughput(operations/minute) vCenter Server Configuration 16 CPU, 32 GB RAM 24 CPU, 48 GB RAM 5.5: small delta 6.0: bigger performance improvement
  • 37.
    Disk IO • SampleWrite Breakdown: single cluster, 64H, 8K VMs, ~300 operations/minute, level 1 stats CONFIDENTIAL 37 0 2000 4000 6000 8000 10000 12000 14000 16000 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 Write KB/s swap core logs DB_data DB_trans SEAT netdump auto image IS DB transaction log DB Data + log  ~6MB/s Inventory Service ~7MB/s DB Data Total: ~3K write IOPs Database and Inventory Service largest contributors
  • 38.
    Disk Characterization Traffic relatedto load, inventory size, and statistics level • Relational DB – Workload – Inventory Size – Statistics Level – Attached extensions causing vCenter Events • Inventory Service – Workload – Inventory Size • DB is involved in nearly all operations: latency is key CONFIDENTIAL 38
  • 39.
    Disk IO asa Function of Inventory Size (Idle) 0 10 20 30 40 50 60 70 0 200 400 600 800 1000 1200 1400 0 200 400 600 800 1000 1200 DiskTransactionsperSecond DiskBandwidth(KBps) Number of ESXi Hosts Bandwidth Trans. per Sec. CONFIDENTIAL 39 1000 hosts: 1.2MB/s 400 hosts: ~500KB/s Modest disk IO when setup is idle. Note: increasing stats level increases disk IO
  • 40.
    Disk IO asa Function of Operational Load: 1000 Hosts, 10K VMs 0 500 1000 1500 2000 2500 3000 0 5 10 15 20 25 30 35 40 0 200 400 600 800 1000 1200 1400 1600 DiskTransactionsperSecond DiskBandwidth(MBps) vCenter Throughput (operations per minute) vCenter Throughput and Disk Usage Bandwidth Trans. per Sec. CONFIDENTIAL 40 1600 VC ops/min 35 MB/s 2500 IOPs 700 VC operations/min 20MB/s 1500 IOPs With high churn, high IOPs capacity and low latency important  consider SSDs
  • 41.
    Performance Considerations forIO • Traffic pattern – Write-mostly (inserts, purges, rollups, config updates), logs sequential, data random – SSDs critical for low latency and high bandwidth – Increase memory to ensure buffer caches capture most reads • Stats Level – Use proper stats level for your use case (more details next slide) • Physical Disk configuration: Use separate spindles for DB, Inventory service, core files, etc. – vCenter Appliance: ~10 VMDKs. Put DB and IS partitions on separate spindles – Windows: everything installed in one place. Use striped disks or SSDs under high churn • Managing DB disk growth (see next slides) • Location of DB (same node as VC, different node on same ESX host, etc.) CONFIDENTIAL 41
  • 42.
    Impact of StatsLevel on Database • Rough rules of thumb (your mileage will vary based on your setup) • Level 1 stats: per-VM and per-host aggregate stats • Level 2 stats: additional per-VM/per-host stats 4x or more stats than Level 1 depending on configuration • Level 3 stats: per-instance stats 6x or more stats than Level 2 depending on configuration • Level 4 stats: additional rollup types 1.4x more stats than Level 3 depending on configuration Recommendations: – Use the stats calculator in vCenter – Try to use higher stats levels only for temporary debugging – If the stat you want is at the wrong level, let us know – Consider vROps for more advanced stats functionality? CONFIDENTIAL 42
  • 43.
    Latency to theDB Latency to DB important (often more so than ESX-to-VC latency) • Almost everything involves the DB… • Stats persistence • Certain UI queries • Updating configuration information • Historical queries (events, alarms, task history) • … Recommendation Place DB and vCenter close together (minimally, same geo if practical) Best case: same VM (if VM properly provisioned) Also good: different VM on same host (if host properly provisioned) Note: DB and vCenter on different hosts/VMs allows for independent sizing and tuning For more info, see VSVC5234 from VMworld 2013 CONFIDENTIAL 43
  • 44.
    DB Performance Considerations(2 of 2) • Manage database disk growth – Majority of DB data is “SEAT” data (Stats, events, alarms, tasks): 80-85% (10s of GBs or more in big setups) – Inventory data: 10-15% of data (usually < 10GB for large inventories) – Choose stats levels wisely to avoid excessive growth – Utilize automatic purging of event/task tables if possible • Recompute DB stats on highly-volatile tables (at least once a day) – VPX_TOPN* 44 For more info, see VSVC5234 from VMworld 2013 CONFIDENTIAL
  • 45.
    Network • Network carriesconfiguration changes and statistics data • Network traffic is bursty due to periodic stats traffic: CONFIDENTIAL 45 0 10 20 30 40 50 60 0 5 10 15 20 NetworkUsage(Mbps) Time (minutes) vCenter Network Usage
  • 46.
    Network Bandwidth Consumedas a Function of Load 0 50 100 150 200 250 0 200 400 600 800 1000 1200 NetworkUsage(Mbps) vCenter Throughput (operations per minute) vCenter Throughput and Network Usage Sent Received Total CONFIDENTIAL 46 Bandwidth requirements modest, but latency is key
  • 47.
    Network • Link fromvCenter to hosts and to DB impacts performance: latency more so than throughput: • Example from our lab: CONFIDENTIAL 47 8.23 8.85 9.23 10.24 12.74 26.39 0 5 10 15 20 25 30 Baseline 10Gbps, 1ms 100Mbps, 20 ms 10Mbps, 50 ms 1.5 Mbps, 100 ms 1.5 Mbps, 500 ms PowerOnLatency(secs) Median PowerOn Latency (secs)
  • 48.
    Other Performance Considerations •Concurrency • API • Extensions • Windows vs. Linux CONFIDENTIAL 48
  • 49.
    How Many ConcurrentOperations Can I Perform? (1 of 2) • vCenter hard limits – 640 concurrent operations before incoming requests are queued – 2000 concurrent sessions (incoming requests plus remote console sessions) • Per-host or per-datastore limits – A host can perform up to 8 provisioning operations at once (provisioning = clone, VMotion, relocate) – If host is source and destination, host can only do 4 operations at once – A datastore can perform up to 128 VMotions at once – A datastore can perform up to 8 Storage VMotions at once – Limits can be changed, but changes are not officially supported • NIC configuration: 10Gb vs. 1Gb – 10Gb NIC allows a host to do 2x more VMotions at a time than 1Gb NIC CONFIDENTIAL 49
  • 50.
    vCenter Concurrency (2of 2) • Clone VM from host A to host B • Each host can participate in 7 other provisioning operations • Clone VM from host A to host A • Host A can only participate in 6 more operations CONFIDENTIAL 50 vCenter Host A VM 1 Host B VM 2 Cost to A: 1 Cost to B: 1 vCenter Host A VM 1 VM 2 Cost to A: 2 Do not use a single host as the source of all clones (i.e., spread out templates)  Better disk performance and better concurrency
  • 51.
    API Performance Considerations:An Example • Example of a good vs. bad client in PowerCLI • PowerCLI: – Simple to use, but involves client-side filtering – Example: Get-VM gets all VMs from server, filters list @ client • $vmList = Get-VM –name “vm1”,”vm2”,”vm3”,”vm4” • Good: 1 server call, client throws away all but vm1,vm2,vm3,vm4 $nameList = “vm1”,”vm2”,”vm3”,”vm4” foreach ($name in $nameList) { Get-VM $name } Bad: 4 server calls, gets all VMs 4 times…excess client/server work For more info, see VSVC5234 from VMworld 2013 CONFIDENTIAL 51
  • 52.
    Impact of Solutions •vROps – Increased network traffic to get stats from hosts – Increased CPU usage on hosts (to retrieve stats) – Increased CPU usage on vCenter to serialize data – Example at scale (1K hosts, 10K powered-on VMs) • Data transmitted without vROps: 0.21 Mbps • Data transmitted with vRops: 1.4 Mbps (7x change) • Data received without vROps: 0.39 Mbps • Data received with vROps: 1.35 MBps (4x change) – vCenter throughput reduced (example: 19% in a high-churn use case) • NSX – 400MB extra memory needed for vSphere web client service heap • vRA: increased concurrent workflows increases CPU on vCenter and network traffic CONFIDENTIAL 52 vROps vCenter ESX ESX ESX ESX ESXVC stats DB VC stats vROps vROps
  • 53.
    Windows vs. vCenterAppliance • The vCenter appliance with embedded DB can support full vCenter Limits • The vCenter appliance and Windows have similar performance CONFIDENTIAL 53 0 200 400 600 800 1000 1200 1400 1600 Small Large Throughput(operations/minute) Inventory Size vCenter 6.0 Windows vCenter 6.0 Server Appl.
  • 54.
  • 55.
    Web Client: PerformanceConsiderations • If possible, browser machine should have 2 CPUs, 4GB • Faster CPUs help (on both client and server side) • Use browser in same geo as application server (RDP to a local machine?) • Make sure application server has sufficient heap size (may need to increase if plugins are installed) • Make sure Inventory Service has sufficient heap size CONFIDENTIAL 55
  • 56.
    Agenda 1 vCenter DeepDive (+ Web Client) 2 Performance Considerations 3 Deployment Strategies 4 Concluding Remarks CONFIDENTIAL 56
  • 57.
  • 58.
    Deployment Options • Withincreased scaling, our goal is that inventory size and churn are NOT the reasons you need to use multiple VCs • Possibilities – Fully Embedded – Embedded + external DB – External Platform Services Controller (PSC) – Multiple VCs with External PSCs, high-availability – Multiple VCs with External PSCs, multi-site • SSO on PSC replaces linked mode – Works on both Windows and Appliance – Allows global sharing of roles, permissions, tags, and licenses CONFIDENTIAL 58
  • 59.
    Fully Embedded orEmbedded with External DB • Good for most single vCenter configurations 59 DB AD VPXD Web Client Server Health perfcharts Java Inv Serv … SSO SPS Content Library AD VPXD DB Web Client Server Health perfcharts Java Inv Serv … SSO SPS Content Library
  • 60.
    External PSC • Goodif you anticipate multiple vCenters CONFIDENTIAL 60 AD VPXD DB Web Client Server Health perfcharts Java Inv Serv … SSO SPS Content Library PSC 2 vCPU, 2GB
  • 61.
    Multiple vCenters: UseCases • Increased Scale – Operations/s? • For some, 50 ops/min is where they want more vCenters – Large number of hosts/VMs? • For some, the single VC “sweet spot” is 200H/2000VMs • Business Considerations – Finance vs. Engineering – PCI-compliant vs. non-PCI-compliant racks – Server vs. Desktop Workloads • Multiple Geographies – Ok with single vCenter managing remote hosts? – vCenter per site or per group of sites? CONFIDENTIAL 61 1. Decide on one or more vCenters 2. Single vs. Multiple SSO sites?
  • 62.
    Multiple vCenters, SinglePSC Pro: Single Pane of Glass Pro: Shared Licenses, roles, permissions Con: Single point of failure (PSC) VPXD DB Web Client Server Health perfcharts Java Inv Serv … SSO SPS Content Library AD VPXD DB Web Client Server Health perfcharts Java Inv Serv … SSO SPS Content Library PSC CONFIDENTIAL 62
  • 63.
    Multiple vCenters, MultiplePSCs with HA and Load Balancer Add a load balancer in front of PSCs 8 VCs per PSC pair Roles/Privileges/License replication VPXD DB Web Client Server Health perfcharts Java Inv Serv … SSO SPS Content Library AD VPXD DB Web Client Server Health perfcharts Java Inv Serv … SSO SPS Content Library PSC PSC LB CONFIDENTIAL 63
  • 64.
    Multiple vCenters, Multi-siteMode with Multiple PSCs • Roles/Privileges/License replication across sites • No HA: must add LB for this VPXD DB Web Client Server Health perfcharts Java Inv Serv … SSO SPS Content Library AD VPXD DB Web Client Server Health perfcharts Java Inv Serv … SSO SPS Content Library PSC PSC CONFIDENTIAL 64
  • 65.
    PSC: Performance Considerations •Default size (2 vCPU, 2GB) should be fine • Tag performance can be impacted by slow link between vCenter and PSC • Login may be slower for SSO vs. Standalone (contacting multiple vCenters) • Search is slower for external SSO (contacting multiple vCenters) • One slow vCenter may slow down Login/Search CONFIDENTIAL 65
  • 66.
    Agenda 1 vCenter DeepDive (+ Web Client) 2 Performance Considerations 3 Deployment Strategies 4 Concluding Remarks CONFIDENTIAL 66
  • 67.
    Conclusions • vCenter 6.0vastly improved in scalability and performance over 5.5 – Our goal: performance is not the reason you need multiple vCenters – Use PSCs for availability, not performance • For best performance, vCenter needs sufficient resources – CPU: scales with inventory size and churn – Memory: scales with inventory size – IO: scales with inventory size, churn, and stats level – Network: low-latency between VC and DB recommended • With 2nd or 3rd party solutions, resource requirements of vCenter will likely increase to manage the same inventory size • Use the appliance! – Windows vCenter and Linux appliance: similar performance, same scale limits – PSC provides sharing of roles/permissions/licenses CONFIDENTIAL 67
  • 68.
    Extreme Performance Series– Break Out Sessions • INF4764 vCenter Performance Best Practices • INF5701 vSphere Compute & Memory • INF4936 Insight Into vSphere 6 vMotion • VAPP4639 Best Practices for Performance Tuning of Virtualized Telco and NFV • INF4853 Docker Containers on vSphere • VAPP5724 High Performance Panel - No App Left Behind • VAPP5165 Monster VM Database Performance • STO4949 Virtual SAN Performance Deep-Dive • EUC5802 Horizon View 6.x Performance and Best Practices • VAPP6537-GD Maximize Performance on vSphere 6 CONFIDENTIAL 68
  • 69.
    Performance Hands OnLabs • HOL-SDC-1604 vSphere Performance Optimization This Lab covers vSphere performance best practices and various performance related features available in vSphere 6. • SPL-CHG-1695 vSphere 6 Challenge Lab The vSphere 6 Challenge asks you to put on your thinking cap to save the day! Each module places you in a different fictional scenario to fix common vSphere operational and performance problems. • ELW-SDC-1604 vSphere Performance Optimization This expert led workshop will take you though the vSphere 6 performance best practices hands on lab with additional support and discussion. CONFIDENTIAL 69
  • 72.
    Extreme Performance Series: vCenterPerformance Best Practices Ravi Soundararajan, VMware, Inc INF4764 #INF4764

Editor's Notes

  • #5 ‘Small’ Inventory: 100H, 1500VMs, 1000VMs powered on ‘Large’ Inventory, 1000H, 15K VMs, 10K VMs powered on
  • #6 Though we have more services in 2015, we’ve reduced memory usage of vpxd and web client enough to offset these added services, so the resource usage is the same for 2013.
  • #9 8
  • #13 12
  • #14 13
  • #15 14
  • #17 16
  • #19 18
  • #20 19
  • #21 vCenter server contains new services Windows Task Manager “Java” processes: Inventory Service (for linked mode and web client) vCO, storage policy-based management Windows Task Manager “tomcat” processes: SRS (Stats Reporting Service): Performance overview charts in VI client SMS (Storage Management Service): Storage Views in VI client … New services may require additional resources on vCenter server More CPU to accommodate bursts More disk bandwidth to accommodate Inventory Service (embedded DB) More memory for JVMs for Java/Tomcat
  • #55 Ravi Transition to Ameet for this set of slides.
  • #56 Ameet transition to Ravi to do this slide. Then, Ravi will transition to Ameet after slide is done. Ameet will talk about common customer performance complaints and what we are doing about it in an unspecified future, and then he’ll transition to some UE fixes. Question for the audience: if the Web Client were blazingly fast, would you still be bothered by navigational differences? Or are the two so intertwined that the question isn’t meaningful.
  • #58 Ameet transition to Ravi
  • #64 The synchronization latency between PSCs is ~30s, whether in a single site or multi-site configuration, according to Sriram.
  • #65 In a configuration like this, you can use a remote PSC if your local one crashes, but you need to manually run a script to repoint your PSC configuration to the remote PSC.