The ability to seamlessly migrate live virtual machines is the foundation for true hybrid cloud experience by being able to move workloads between on-premises and public clouds. Live virtual machine migration involves transferring both virtual machine and the corresponding state between hosts without disrupting any running services. The VMware vSphere vMotion feature provides an important capability to migrate the active state of virtual machines from one physical ESXi host to another.
However, testing against vSphere 4 and earlier versions showed that vMotion migrations introduced cluster stability issues and significant performance degradation of VMware Tanzu GemFire. But how do more recent versions of vMotion perform?
VMware’s Nabarun Nag has conducted tests of live migrations of GemFire clusters on a recent version of vMotion (6.7) to see if things have improved. The verdict:
“In running heavy read and write operations consistent to meet load performance against GemFire and vSphere 6.7, all operations completed successfully, there was no data loss, and the cluster nodes remained stable. As a result of this testing, we are able to state that GemFire is compatible with vMotion.”
Highlights of the white paper:
• In these experiments, we were able to see that when vMotion migrations occur, there is an expected temporary drop in the performance in both read-operation and write-operation workloads.
• These workloads resumed their normal rate of operation once the vMotion migration of the servers were completed.
• No members were removed or disconnected during the vMotion migrations.
• No data loss occurred.
• On the client side in both experiments, write operations began to take longer than 1 second during the vMotion migration.
• During vMotion migrations of one virtual machine, the average throughput decreased by 20% and average latency increased by 1-10 ms under the specific conditions of the test environment.
• During vMotion migrations of all four virtual machines, the average throughput decreased by 40% and average latency increased by 20-50 ms under the specific conditions of the test environment.