16 March,2011 by Jack Vamvas
Some thoughts on analyzing whether virtualization is a viable alternative to physical servers for SQL Server.The general conclusion is that it offers very good management flexibility particularly around DR – and for most applications I’ve tested on is acceptable.
The whole performance stack must be analysed - which is outside the scope of this post.
1) vSphere HA/DRS – HA is implemented by pooling virtual machines and host into a cluster. Hosts are monitored and if there is a host failure the virtual machine is restarted on an alternative host
2a) Failover clustering with MSCS - cross – host would present the 2 MSCS VM nodes residing on different hosts .
3)There are other alternatives – combination of virtualization and log shipping or even database mirroring
Disaster Recovery with Site Recovery Manager(SRM)
All database volumes and OS volumes would exist at the storage layer on the Primary site .Synchronous mirroring would retain mirrored copies on the Secondary site
An SRM relationship would be retained from Primary to Secondary. In the event of DR the VMs would be started on the secondary mapped to the storage layer.
1)HA and DR of vSphere built-in.
2)Hardware maintenance is efficient .Planned outages , manage with vMotion
3)Potentially less management overhead
1)vSphere HA/DRS is only aware of hardware failure or connection failure – whereas MSCS is application failure aware
2)MSCS clustering with VM option does not allow vMotion
Considerations in virtualizing
1)The same disk alignment should be available in virtualised as in physical
2)Need to define what failure is being protected against. If the focus is on physical server failure MSCS is potentially overkill.
If there is a necessity to protect against application level failure then application level clustering is required
Testing patterns for POC
Outline and test realistic application scenarios
Demonstrate operation tasks and performance testing
1) Comparisons will focus on a) Disk Transfers/sec, Average Disk Queue Length, Average Disk Write Queue Length, Average Disk Read Queue Length, and % Disk Time.
2) SQL server level testing – defined framework which is part of the overall benchmark acceptance testing
3) Consider testing with 1:1 and multiple guests to host. Look at 1: 4 ratio – as this aligns with SQL Server 2008 Enterprise Server licensing on virtualised environment
Consideration should be given to Third Party Vendors currently not supporting Virtualisation