VMWare performance difference between using physical disk and virtual disk (VMFS)

Question

Under Windows 7 with VMWare workstation 8, I want to run a Linux-VM with Ext3/4 as guest filesystem. I want the VM to have maximum read/write performance. So I'm thinking about the best hard drive setup for the VM. Should I give the VM access to a physical disk, so that the VM can access the disk natively using just Ext3/4?

The other option would be to create a virtual disk container. This would mean mapping the Ext3/4 filesystem inside the VM over the VMWare VMFS to a NTFS filesystem on the physical hard disk. This approach has some advantages in terms of backup and partitioning.

What is the impact of using VMFS mapping? Will it steal a lot of disk performance, or will I never notice the difference, so that I can also choose using a virtual-physical-mapped filesystem?

Answer 1

The impact greatly depends on many factors:

• Technology used: SCSI vs IDE; both native and virtualized
• Software used: VMWare version, Windows version, drivers, etc.
• Possible complications like antivirus on Windows scanning VMware containers, etc.
• Workload patterns: sequential vs random access, etc.
• Caching involved: Windows caches and pre-fetches some file reads, how this affects VMWare depends on what else is running on the Windows host.

Bottom line: nobody is going to tell you how it is going to affect your workloads, the only way to find out is to test both scenarios.

Moreover, doing direct disk access in VMWare is less secure, less portable and less convinient. E.g. you cannot migrate to another host easily, cannot resize disks without changing partitions, and there are possible security implications (read VMWare docs).

So, my suggestion is, go with the virtual disk container first. If you find the disk performance unacceptable, explore the direct disk access option.