The question often comes up, when should I use Raw Device Mappings (RDMs) instead of traditional VMFS/VMDK virtual disks. The answer for this usually is, always use VMFS unless you fit into specific use cases. An RDM is raw LUN that is presented directly to a VM instead of creating a traditional virtual disk on a VMFS datastore. Because a VM is directly accessing a LUN there is a natural tendency to think that a RDM would provide better performance as a VM is reading/writing directly to a LUN and not having to deal with the overhead of the VMFS file system. While you might think that, it’s not the case, RDM’s provide the same performance levels as a VM on VMFS does. To prove this VMware performed testing and published a white paper way back in 2007 on ESX 3.0 were they did head to head testing of the same workloads on a VM on VMFS and a VM using an RDM. The conclusions of their testing was:
- “The performance tests described in this study show that VMFS and RDM provide similar I/O throughput for most of the workloads we tested.”
They went on to list some of the use cases that one might want to use an RDM such as a clustering solution like MSCS or when you need to take array snapshots of a single VM. Some additional use cases might be around an application that needs to write SCSI commands directly to a LUN or if you may potentially need to move an application from a VM to a physical server where you can move the LUN. Performance should really never be a valid use case for using an RDM as they also have some disadvantages which include more difficult management and certain vSphere feature restrictions.
They also tested VMFS vs RDM again with vSphere 5.1 and once again the conclusions were the same:
- “What you should takeaway from this vSphere 5.1 data, as well as all previously published data, is that there is really no performance difference between VMDK and RDM and this holds true for all versions of our platform. +/- 1% is insignificant in today’s infrastructure and can often be attributed to noise.”
Well those older recommendations still apply today and to proof that VMware just published a new white paper that examine SQL Server performance on vSphere 6.5. As part of that white paper they chose revisit the VMFS vs RDM topic and included testing of SQL Server performance on both VMFS and RDM disk. Not surprisingly the results were the same, in fact they found VMFS performed ever so slightly better than an RDM disk. The conclusions of this recent testing was:
- “The results from DVD Store 3 running on the same host reveals that, as expected, the performance is virtually identical; in fact, just like the vSphere 5.1 case, the VMFS case slightly (1%) outperformed RDM.”
You can see the results in the below graph which validates VMware’s recommendation to use VMFS by default unless you have a special functional requirement that would require using an RDM.