Skip navigation

Description:  This document outlines considerations around optimisation of vCenter Server instances and best practice recommendations to maximise performance of your vCenter ecosystem. Each item listed should be addressed in the context of the target environment as there is no one solution to optimise the vCenter management environment. The following is simply a list of recommendations that should, to some extent, improve performance in most environments.

 

Prerequisites: 

             Microsoft Windows vCenter Server

             Microsoft SQL Database Instance

 

Part 1 – vCenter Server

                             

Description

Recommendations / Best Practices / KB Links

Virtual Server Sizing

Ensure that the vCenter virtual   system(s) are sized accordingly based in the inventory size. Where vCenter   components are separated and distributed across multiple virtual machines   ensure that all systems meet the sizing recommendation set out in the   installation and configuration documentation https://www.vmware.com/support/pubs/vsphere-esxi-vcenter-server-pubs.html

(vSphere 5.1 )

http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=2021202

Distribute vCenter Services across   multiple virtual machines

Depending on inventory size, multiple   virtual machines can be used to accommodate different vCenter roles. VMware   recommends separating VMware vCenter, SSO Server, Update Manager and SQL for   flexibility during maintenance and to improve scalability of the vCenter   management ecosystem.

Dedicated Management Cluster

For anything other than the smallest   of environments, VMware recommends separating all vSphere management   components onto a separate out-of-band management cluster. The primary   benefits of management component separation, include:

·           Facilitating quicker troubleshooting and problem resolution as   management components are strictly contained in a relatively small and   manageable cluster.

·           Providing resource isolation between workloads running in the   production environment and the actual systems used to manage the   infrastructure.

·         Separating the management components from the resources they are   managing.

vCenter to Host operational latency

The number of network hops between   the vCenter Server and the ESXi host affects operational latency. The ESXi   host should reside as few network hops away from the vCenter Server as   possible.

vCenter to SQL Server operational   latency

The number of network hops between   the vCenter Server and the SQL database affects operational latency. Where   possible, vCenter should reside on the same network segment as the supporting   database. If appropriate, configure an affinity rule to ensure that the   vCenter Server and database server reside on the same ESXi host, reducing   latency still further.

Java Max Heap Size

Ensure that the max heap size for   Java virtual machine is set correctly based on the inventory size.  Confirm heap size on JVM Heap settings on   vCenter, Inventory Service, SSO and Web Client are checked. Monitor Web   Services to verify. (vSphere 5.1)

http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=2021302

Client Connections

Attempt to limit the number of   clients connected to vCenter Server as this affects its performance. This is   particularly the case for the traditional Windows C# client.

Performance Monitoring

Use performance monitoring tools to   ensure the health of the vCenter eco system and troubleshot problems as they   arise. Where appropriate, configure a vC Ops Custom Dashboard for   vCenter/Management components.  Also   ensure appropriate alerts and notifications on performance monitoring tools   exist.

Virtual disk type

All management virtual machine VMDK’s   should be provisioned in an eagerZeroedThick format.  This   provides approximately a 10-20 percent performance improvement over the other   two disk formats.

vCenter vNIC  type

Ensure to employ the VMXNET3 paravirtualized network adaptor to maximise network   throughput, efficiency and reduce latency.

ODBC Connection

Ensure that the vCenter and VUM ODBC   connections are configured with the minimum permissions required for daily   operations. Additional permissions are required during installation and   upgrade activities but not for day to day operations. Please refer to the   Service Account Permissions section below.

vCenter Logs Clean Up

vCenter server has no automated way of purging old vCenter   Log files. These files can grow and consume significant disk space on the   vCenter Server. Consider a 3/6 monthly scheduled task to delete or move log   files older than the period of time defined by business requirements.

The example, the VBscript  below could be used to clean up old log files from vCenter. This script   delete files that are older than a fixed number of days, defined in line 9,   from the path set in line 6. This VBscript can be   configured to run as a scheduled task using the windows task scheduler.

Dim Fso

Dim Directory

Dim Modified

Dim Files

Set Fso = CreateObject("Scripting.FileSystemObject")

Set Directory = Fso.GetFolder("C:\ProgramData\VMware\VMware   VirtualCenter\Logs\")

Set Files = Directory.Files

For Each Modified in Files

If DateDiff("D",   Modified.DateLastModified, Now) > 180 Then Modified.Delete

Next


For more information, refer to KB article: KB1021804 Location of vCenter Server log files. 

For additional information on modifying logging   levels in vCenter please refer to KB1004795 and KB1001584.

Note:   Once a log file reaches a maximum   size it is rotated and numbered similar to component-nnn.log files and they   may be compressed.

Statistics   Levels

Statistic   collection intervals determine the frequency at which statistic queries   occur, the length of time statistical data is stored in the database, and the   type of statistical data that is collected.


As   historical performance statistics can take up to 90% of the vCenter server   database size, they are the primary factor in the performance and scalability   of the vCenter Server database. You can view the collected historical statistics   through the performance charts in the vSphere Web Client, through the   traditional Windows Client or through command-line monitoring utilities for   up to 1 year after the data was first ingested into the database.

Ensure   that statistics collection times are set as conservatively as possible so   that the system does not become overloaded. At the same it is equally   important to ensure that the retention of this historical data meets the   customer’s data compliance requirements.

Because   the statistics data consumes such a large proportion of the database, proper   management of vCenter statistics is an important consideration for the overall   database health. In addition, the processing of this data through a series of   rollup jobs to stop the SQL server becoming overloaded, is also a key consideration   for vCenter Server performance.


For   Instance:

Set   new DB Data Retention Period of 60 Days

Configure   DB to not retain performance data beyond 60 days.

Task   and Events Retention

Ensure   that Task and Events Retention levels are set as conservatively as possible   whilst meeting the customer’s data retention requirements. Every time a task   or event that is executed via vCenter, this is stored in the database. For   example a task is created when a user powers on or off on a virtual machine   or an event when something occurs such as vCPU  usage for a VM changing to red.

vCenter Server has a   Database Retention Policy setting that allows you to specify after how long   vCenter Server tasks and events should be deleted. This correlates to a   database rollup job that purges the data from the database after the selected   period of time.

Whilst   compared to statistical data these tables consume a relevantly small amounts   database space, it is good practice to consider this option for further   database optimisation. 

For   Instance, by default, vCenter is configured to store tasks and events data   for 180 days. However, it may be possible, based on the customer’s compliance   requirements, to configure vCenter not to retain Event and Task Data in the   database beyond 60 days.

vCenter   Server Backup Best Practice

In addition to scheduling regular backups of the vCenter database, the backups   for the vCenter Server should also include the SSL certificates and license   key information.

 

 

Part 2 – SQL DB Server

                                 

SQL   Database Server Disk Configuration

The   database data file generates mostly random I/O, while database transaction   logs generate mostly sequential I/O. The traffic for these files is almost   always simultaneous so it’s preferable to keep these files on two separate   storage resources that don’t share disks or I/O.

Therefore,   where a large inventory demands it, ensure that the vCenter Server database   uses separate drives for data and logs which, in turn, are backed by   different physical disks.

tempDB Separation

For   large inventories place tempDB on a different   drive, backed by different physical disks than the vCenter database files or   transaction logs.

Reduce   Allocation Contention in SQL Server tempDB database

Use   multiple data files to increase the I/O throughput to tempDB.   Configure 1:1 alignment between TempDB files and vCPUs  (up to eight) by spreading TempDB across at least as many equal sized files   as there are vCPUs.

For   instance, where 4 vCPUs exist on the SQL server,   create three additional TempDB data files, and make them all equally sized.   They should also be configured to  grow   in equal amounts

After   changing the configuration, restart the SQL Server instance. For more   information please refer to: http://support.microsoft.com/kb/2154845

Database   Connection Pool

vCenter server starts,   by default, with a database connection pool of 50 threads. This pool is then   dynamically sized according to vCenter workload. If high load is expected due   to a large inventory, then the size of the pool can be increased to 128   threads. This will increase memory consumption and load time of the vCenter   Server. To change the pool size, edit   the vpxd.cfg file,   adding: Where ‘128’ is the number of connection threads to be configured.


    
< vpxd>
  < odbc>
  < maxConnections>128</maxConnections>
  < /odbc>
  < /vpxd>

Table   Statistics

Update statistics   of the tables and indexes on a regular basis for better overall performance   of the database. Create an SQL job to carry out this task or   alternatively it should form part of a vSphere database maintenance plan.

http://sqlserverplanet.com/dba/update-statistics

Index   Fragmentation

(Not   Applicable to vCenter 5.1 or newer)

Check   for fragmentation of index objects and recreate indexes if needed. This   happens with vCenter due to statistic roll ups. Defragment after <30%   fragmentation. See this KB1003990.

Note: With the new enhancements and   design changes made in the vCenter Server 5.1 database and later versions,   this is no longer applicable or required.

Database   Recovery Model

Set   the transaction logs to SIMPLE  recovery. This model will reduce the disk space needed for the logs as well   decrease I/O load. 

Choosing   the Recovery Model for a Database:

http://msdn.microsoft.com/en-us/library/ms175987(SQL.90).aspx

How   to view or Change the Recovery Model of a Database in SQL Server Management   Studio

http://msdn.microsoft.com/en-us/library/ms189272(SQL.90).aspx

Virtual   Disk Type

Where the vCenter database server is   a virtual machine, ensure that all VMDK’s are provisioned in an eagerZeroedThick format.  This   option provides approximately 10-20 percent performance improvement over the   other two disk formats.

Verify   SQL Rollup Jobs

Ensure SQL Agent rollup jobs have   been created on SQL during the vCenter Installation.  For instance:

  • Past Day stats rollup
  • Past Week stats rollup
  • Past Month stats rollup

For the full set of   stored procedures and jobs please refer to the appropriate article below.

If necessary, recreate MSSQL agent   rollup jobs.

Note that detaching, attaching,   importing, and restoring a database to a newer version of MSSQL Server does   not automatically recreate these jobs. To recreate these jobs, if missing,   please refer to:  KB1004382.

KB 2033096 (vSphere 5.1   & 5.5)

http://kb.vmware.com/kb/2033096

KB 2006097 (vSphere   5.0)

http://kb.vmware.com/kb/2006097

Ensure that the myDB  references the vCenter Server database and not the master or some other   database. If these jobs reference any other database, you must delete and   recreate the jobs.

Ensure database jobs are running   correctly

Monitor scheduled database jobs to   ensure they are running correctly.

For more information, refer to KB   article: Checking   the status of vCenter Server performance rollup jobs: KB2012226

Verify MSSQL Permissions

Ensure that the local, SQL and AD   Permissions required are in place and align with the principle of least   privilege (see below)

If necessary, truncate all unrequired   performance data from the database (Purging Historical Statistical Performance   Data)

For more information, refer to   KB article: Reducing   the size of the vCenter Server database when the rollup scripts take a long   time to run KB1007453

To truncate all performance data from   vCenter Server 5.1 and 5.5:

Warning: This procedure   permanently removes all historical performance data. Ensure to take a backup   of the database/schema before proceeding.

Shrink Database

After purging historical data from   the database optionally shrink the database.    This is an online procedure to reduce the database size and to free up   space on the VMDK, however, this activity will not in itself improve   performance.

Shrinking the size of the VMware   vCenter Server SQL database  KB1036738

http://msdn.microsoft.com/en-us/library/ms189080.aspx

Rebuilding indexes to Optimize the performance of SQL Server

Configure regular maintenance job to rebuild indexes. KB2009918

VPX_HIST_STAT  

Table   Sizes

VMware   recommend a fill factor of 70% for the 4 VPX_HIST_STAT tables. If this is too   high for resources on the server, then it will need to take time splitting   pages, which equates to additional I/O.

If   you are experiencing high unexplained I/O in the environment, monitor the SQL   Server Access Methods object: Page Splits/sec. Page splits are expensive, and cause your table to perform   more poorly due to fragmentation. Therefore, the fewer page splits you have   the better your system will perform.

By   decreasing the fillfactor in your indexes, what you   are doing is increasing the amount of empty space on each data page. The more   empty space there is, the fewer page splits you will experience. On the other   hand, having too much unnecessary empty space can also hurt performance   because it means that less data is stored per page, which means it takes more   disk I/O to read tables, and less data can be stored in the buffer cache.

High   Page Splits/sec will result in the   database being larger than necessary and having more pages to read during   normal operations. 

Determining   where growth is occurring in the VMware vCenter Server database (1028356)

http://kb.vmware.com/selfservice/microsites/search.do?cmd=displayKC&externalId=1028356

Troubleshooting VPX_HIST_STAT table sizes in VMware   vCenter Server 5.1 KB2038474

Reducing   the size of the vCenter Server database when the rollup scripts take a long   time to run KB1007453

Monitor Database Growth

Monitor database growth over a period   of time to ensure the database is functioning as expected. For more   information, refer to KB article: Determining where growth is occurring in   the vCenter Server database KB1028356

schedule and verify regular database   backups

The vCenter, SSO, VUM and SRM servers   are by themselves stateless. The databases themselves are far more critical   since they store all the configuration and state information for each of the   management components. These databases must be backed-up nightly and the   restore process of each database needs to be tested periodically. 

Ensure that a schedule of regular   backups exists of the vCenter database and based on requirements of the   business, restore and mount databases from backup periodically onto a non-production   system to ensure a clean recovery is possible, should database corruption or   data loss occur in the production environment.

Create a Maintenance Plan for vSphere   databases

Work with your DBA’s to create a   daily and weekly database maintenance plan. For Instance:

·           Check Database   Integrity

·           Rebuild Index

·           Update Statistics

·           Back Up Database (Full)

·           Maintenance Cleanup Task

Note: DO NOT SHRINK DB IN   MAINTENACE PLAN UNLESS THERE IS A SPECIFIC REQUIREMENT TO RECLAIM DISK SPACE

http://msdn.microsoft.com/en-us/library/ms189080.aspx

 

 

 

Part 3 - Service Account Permissions (Least Privilege)

       

vCenter Service Account  

Required by the ODBC Connection for   access to the database. The vCenter service account must be configured with dbo_owner  privileges for normal operational use.    However, the vCenter database account being used to make the ODBC   connection also requires the db_owner role   on the MSDB System database during installation or upgrade of the vCenter   Server.  This permission facilitates the installation of SQL Agent jobs   for vCenter statistic rollups.

Typically the DBA would only grant   the vCenter service account the db_owner role on the MSDB System database when installing   or upgrading vCenter, then revoke that role when these activities are   complete.

RSA_DBO

Only Required for SSO 5.1 the RSA_DBA   account is a local SQL account which is used for creating the schema (DDL)   and requires dbo_owner  permissions.

RSA_USER

Only Required for SSO 5.1, the   RSA_USER reads and writes data (only DML).

VUM Service Account

Despite being a 64bit application,   VUM requires a 32bit ODBC connection from “C:\Windows\SysWOW64\odbcad32.exe”.   The VUM service account must be provide the dbo_owner permission on the   VUM DB. The installation of vCenter Update Manager 5.x with a Microsoft SQL   back end database also requires the ODBC connection account to temporarily   have db_owner  permissions on the MSDB System database. This was a new requirement in   vSphere 5.0.

As with the vCenter service account,   typically the DBA would only grant the VUM service account the db_owner role   for the MSDB System database when installing or upgrading the VUM component   of vCenter, then revoke that role when this task is complete.

SRM Service Account

Despite being a 64bit application,   SRM requires a 32bit ODBC connection from “C:\Windows\SysWOW64\odbcad32.exe”.   The SRM service account must be configured with the dbo_owner privilege on the SRM   database.

http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=1027973 

 

I’ve recently been working with a customer who has been a program of right-sizing vCPUs and Memory based on recommendations from vC Ops oversized VM reports. For many virtual machines, the vC Ops report has recommended a vCPU count of 3, and I’ve had to work hard to justify this to the customer, and realign their thinking away from the traditional pCPU architecture based on a 20=1, 21=2, 2^2=4, etc. system.


Just to clarify, vSphere 5 introduced VM Hardware Version 8 (AKA Hardware Compatibility Mode). This is the minimum version of virtual machine hardware that is required to allow odd numbered vCPUs on the guest.


Although vSphere 5 supports VMs with odd or/and even number of vCPUs, there is no specific recommendations about this from VMware. Basically, vC Ops recommends the number of required vCPUs depending on the requirements of the workload displayed over time, and  the most important thing is that you’re careful about allocating more vCPU's than physical cores per pCPU (for NUMA locality) and that you only allocate as many vCPU cores as the application needs.


  Whilst there is no recommendation for/against odd numbers of vCPU’s, we are all of course tuned to the physical world where we talk about 2, 4, 6, 8  CPUs etc. and 1024MB, 2048MB etc. of memory and it sometimes hard to unlearn what we’ve learned over such a long time.  Below is an article that covers this question in a bit more depth. http://blog.solori.net/tag/odd-core-counts-with-vmware/

Filter Blog