Thursday, August 14, 2008

General Policy - Virtualization of Elite Products

Thomson Elite has released an official policy regarding - Virtualization of Elite Products including Prolaw.
1.1 General Policy
Thomson Elite generally recommends against using virtualization environments (e.g., Virtual Machines from VMware or Microsoft Virtual Server) for primary production servers hosting Thomson Elite products.

Thomson Elite makes no performance warranties in relation to Thomson Elite applications hosted on Virtual Machines.

The use of Virtual Machines for test and disaster recovery (DR) environments may be appropriate, provided customers understand and accept sole responsibility for any performance issues related to virtualization.

1.2 Behind the Policy
While use of Virtual Machines can be an excellent way to consolidate server functions onto fewer boxes and reduce the time hardware is idle, the effectiveness of such consolidation relies, in part, on the functions consolidated either not requiring full CPU utilization or requiring it at different times. Since many Thomson Elite applications tend to share the same peak load times (e.g., month-end accounting), it is difficult to capitalize on the strengths of Virtual Machines without at least as much hardware as is recommend by Thomson Elite.

1.3 Production Use Against Recommendation
Customers running virtual environments for primary production use, against Thomson Elite’s recommendations, and who require support for performance issues related to using Virtual Machines, will be charged at standard support rates.

Thomson Elite will not be responsible for the performance of products running on Virtual Machines, since they fail to meet the requirements established by Thomson Elite.

1.4 Virtualization Performance Concerns

1.4.1 Processor Performance
In addition to the hardware that Thomson Elite recommends for an appropriate level of performance, software which allows the use of Virtual Machines imposes an additional overhead which can reduce performance by 15-20 percent or more in some cases.

1.4.2 Disk I/O Performance
In addition to processor cycles that may be consumed as overhead by the use of Virtual Machines, Disk I/O performance is also impacted by the added layer of virtualization. As a point of reference, using Microsoft’s SQLIO tool, limited testing on directly attached storage consisting of a six (6) spindle RAID 10 array, revealed the following:

* VMware ESX Server 2.5.2 imposed approximately a 15% sequential read performance penalty using a virtual SCSI disk.
* VMware GSX Server 3.2 imposed approximately a 20% sequential read performance penalty using a direct connection to the physical array.
* Microsoft Virtual Server 2005 imposed over a 30% sequential read performance penalty using a virtual SCSI disk.
* Microsoft Virtual Server 2005 imposed nearly a 70% read performance penalty using a virtual IDE disk.

1.5 Virtualization Products
While Thomson Elite recommends against Virtual Machines in primary production use, those customers who choose to proceed with virtual environments where “Not Recommended,, may find the following virtualization products more suitable than desktop versions, such as VMware Workstation or Microsoft Virtual PC: VMware GSX Server 3.2 (or VMware Server, the free successor to GSX Server expected in Q2 2006)

* VMware ESX Server 2.5
* Virtual Server 2005 Standard Edition
* Virtual Server 2005 Enterprise Edition

1.6 Future of Virtualization
Thomson Elite is aware of the growing popularity of virtualized solutions and will continue to evaluate it as the market, technology, supporting hardware and underlying platforms mature.

I've been told by other users that if you go ahead and run Prolaw on a virtual platform and have any sort of performance problem, the first thing Prolaw support will do is tell you to move the server from the virtual platform back to a physical server.

Does anyone have any experience running Prolaw in a virtualized environment?