Chinwag with Vinay Gaonkar
Vinay Gaonkar used to work at VMware, and since switched to Violin Memory – last year and this year VMware worked with Violin to smash through the psychological performance barrier of 1M IOPS to a VM. This year the guys at the performance team in VMware took a Violin Array and used it to do some Eric Sloof-Style “Myth Busting”, showing how there is no real performance difference between VMFS/VMDK and RDMs (Raw Device Mappings). It’s persistent myth that the “RAW” in RDM indicate some sort of “native throughput” that outperforms our VMFS/VMDK files. It’s something I’ve been saying since I was instructor in 2004/5. For me RDMs perhaps should have been called “Native Device Mappings”, because the allow the VM to natively speak to a LUN (iSCSI/FC) on the array mainly to get features or meet requirements for rather particular configurations – for example in the early days of MSCS Clustering inside a VM often you’d find either physical or virtual RDMs were a requirement to get to the shared/quorum volumes… or in technologies like EMC Recover Point you needed RDM to get some sort of “management LUN”. Both use case – really had nothing to do with performance….