The white paper, Virtualizing Microsoft SQL Server 2008 with Citrix XenServer, discusses the performance and scalability of Microsoft SQL Server 2008 when it is deployed within virtual machines (VMs) hosted by Citrix XenServer 5.0 Dell OEM Edition on a Dell™ PowerEdge™ R900 server with a Dell-EqualLogic PS5000 iSCSI storage array. DBHammer was used as the benchmarking tool was used to simulate the SQL workload. The following were the highlights of the test findings:

1. Virtualizing single instance SQL server farms using XenServer is possible with very little virtualization overhead.
  • XenServer overhead averaged 7.6% and 7.5% for the two single SQL Server vs. single VM tests respectively.
  • By comparison, Vendor X averaged 36.4% overhead in the single test that it was able to participate in.
2. The benefits of virtualization and the strength of XenServer specifically are most pronounced with multiple instances of SQL Server.
  • Multiple XenServer VMs are increasingly more productive than physical SQL servers as the number of instances increase.
  • XenServer out-performs both physical SQL servers running multiple SQL Server instances and vendor X running multiple VMs in the majority of test cases.
3. Virtualizing SQL Server 2008 on XenServer doesn’t require a Fibre Channel SAN. In these tests, we utilized standard, relatively inexpensive iSCSI-based storage.The complete paper is available in PDF format. This wiki page provides the highlights and a place to discuss and ask questions about the paper through the comments section at the bottom of the page.


Test Configuration


Hardware

Physical and virtual (XenServer & Vendor X) host server:
  • Dell® PowerEdge® R900, Four quad-core Intel® E7310 1.6Ghz CPUs, 64GB RAM

DBHammer for SQL load generation client servers (XenServer VMs):
  • Intel 64bit host server, 24 CPU cores (4 x 6 cores Intel E7450), 2.4Ghz/32GB
Storage
  • Dell EqualLogic PS5000, 16x138GB Disks, Raid 10/913GB
  • 50GB database, 50GB log LUNs per SQL Server/VM
Network
  • NetGear®, 24 port gigabit switch, model #GS724T
  • NICs: Intel gigabit VT Quad Port Server Adapter, 3 ports
CitrixXenServer-SQL2k8-TestSetup

Test Scenarios


The three major test groups for the study were:

Test group #1: Small – 4 Core/14GB physical or virtual host server
  • Single instance physical server vs. single VM server
  • Two instance physical server vs. two VM server
Test group #2: Medium – 8 Core/28GB physical or virtual host server
  • Single instance physical server vs. single VM server
  • Two instance physical server vs. two VM server
  • Four instance physical server vs. four VM server

Test group #3: Large – 16 Core/56GB physical or virtual host server
  • Single instance physical server (baseline only, no VM comparison test)
  • Two instance physical server vs. two VM server
  • Four instance physical server vs. four VM server
  • Eight instance physical server vs. eight VM server

Results: Small – 4 Core/14GB physical or virtual host server


DBHammer-Small
Results: One and two instances of SQL Server on virtual servers vs. the same number on one physical server
  • One instance of SQL Server
XenServer virtualized a single instance of SQL Server with a 7.6% virtualization overhead vs. 36.4% for Vendor X.

  • Two instances of SQL Server
XenServer out-performed the physical SQL Server by 30% and Vendor X by 31%.

Results: Medium – 8 Core/28GB physical or virtual host server


DBHammer- Medium
Results: One, two and four instances of SQL Server on virtual servers vs. the same number on one physical server

  • XenServer virtualized a single VM workload with only 7.5% virtualization overhead. Due to an inability to support more than four vCPUs, Vendor X was unable to compete with XenServer in this single instance vs. single VM, 8 core/28GB test.
  • XenServer out-performed the physical server in the two and four instance tests, 6.7% and 27.8% better, respectively.
  • XenServer out-performed Vendor X in the in the two and four instance tests, 15.7% and 15.5% better, respectively.

Results: Large – 16 Core/56GB physical or virtual host server


DBHammer - Large
Results: One, two, four and eight instances of SQL Server on virtual servers vs. the same number on one physical server

  • XenServer and Vendor X were unable to participate in the single instance, single VM test due to the inability of both to have sixteen vCPUs assigned to a single VM.
  • Vendor X was unable to participate in the two instance, two VM tests due to its inability to support eight vCPUs in a single VM.
  • The physical server performed best with a single instance of SQL Server. As additional instances were created and tested, performance declined steadily.
  • XenServer outperformed the physical server in the four and eight instance tests.
  • XenServer outperformed Vendor X in the eight VM test. Vendor X outperformed XenServer slightly in the four VM test, the only time Vendor X out-performed XenServer in these tests.
  • Vendor X performance declined by 10.5% when scaling from four to eight VMs.

Additional details, including system component-level testing of processors, memory, and NIC performance, are included in the full white paper: Virtualizing Microsoft SQL Server 2008 with Citrix XenServer

 Please use the comments area below for discussion and questions about this white paper.