Here are some read and write performance numbers and graphs for Windows guests.
This was done with "iozone -a -i 0 -i 1 -Rb c:\output.wks" against a pair of virtual drives from within an XP SP2 guest. The host drive is a WD SATA-I directly attached to the motherboard's controller (no RAID). I created an 8 GB partition on the disk, and then created a vmdk in that partition, so that each test used the same portion of the same disk, with the raw and vmdk drives being mapped to the guest as IDE 0:0 and IDE 0:1 respectively. I used the "default" cluster size for each of the various format operation.
Direct access to NTFS host partition
Write peak 440.7 MB/sec
http://www.ntrg.com/misc/vmware/ntfs-raw-writes.jpg
Read peak 2,774.7 MB/sec
http://www.ntrg.com/misc/vmware/ntfs-raw-reads.jpg
NTFS vmdk on NTFS host partition
Write peak 464.0 MB/sec
http://www.ntrg.com/misc/vmware/ntfs-vmdk-writes.jpg
Read peak 2,469.9 MB/sec
http://www.ntrg.com/misc/vmware/ntfs-vmdk-reads.jpg
Direct access to FAT32 host partition
Write peak 533.6 MB/sec
http://www.ntrg.com/misc/vmware/fat32-raw-writes.jpg
Read peak 2,749.7 MB/sec
http://www.ntrg.com/misc/vmware/fat32-raw-reads.jpg
FAT32 vmdk on FAT32 host partition
Write peak 467.3 MB/sec
http://www.ntrg.com/misc/vmware/fat32-vmdk-writes.jpg
Read peak 2,743.6 MB/sec
http://www.ntrg.com/misc/vmware/fat32-vmdk-reads.jpg
All of the disk types produce consistent read performance, mostly due to the OS caching (note that the numbers are higher than the SATA-I maximum performance), although raw NTFS reads are the fastest by an edge.
Write performance shows that raw FAT32 is the fastest (an earlier instance of the write test produced +600 KB/sec, which is about 33% higher than all other virtual drive types).
updated performance numbers posted July 2, 16:47 GMT
This was done with "iozone -a -i 0 -i 1 -Rb c:\output.wks" against a pair of virtual drives from within an XP SP2 guest. The host drive is a WD SATA-I directly attached to the motherboard's controller (no RAID). I created an 8 GB partition on the disk, and then created a vmdk in that partition, so that each test used the same portion of the same disk, with the raw and vmdk drives being mapped to the guest as IDE 0:0 and IDE 0:1 respectively. I used the "default" cluster size for each of the various format operation.
Direct access to NTFS host partition
Write peak 440.7 MB/sec
http://www.ntrg.com/misc/vmware/ntfs-raw-writes.jpg
Read peak 2,774.7 MB/sec
http://www.ntrg.com/misc/vmware/ntfs-raw-reads.jpg
NTFS vmdk on NTFS host partition
Write peak 464.0 MB/sec
http://www.ntrg.com/misc/vmware/ntfs-vmdk-writes.jpg
Read peak 2,469.9 MB/sec
http://www.ntrg.com/misc/vmware/ntfs-vmdk-reads.jpg
Direct access to FAT32 host partition
Write peak 533.6 MB/sec
http://www.ntrg.com/misc/vmware/fat32-raw-writes.jpg
Read peak 2,749.7 MB/sec
http://www.ntrg.com/misc/vmware/fat32-raw-reads.jpg
FAT32 vmdk on FAT32 host partition
Write peak 467.3 MB/sec
http://www.ntrg.com/misc/vmware/fat32-vmdk-writes.jpg
Read peak 2,743.6 MB/sec
http://www.ntrg.com/misc/vmware/fat32-vmdk-reads.jpg
All of the disk types produce consistent read performance, mostly due to the OS caching (note that the numbers are higher than the SATA-I maximum performance), although raw NTFS reads are the fastest by an edge.
Write performance shows that raw FAT32 is the fastest (an earlier instance of the write test produced +600 KB/sec, which is about 33% higher than all other virtual drive types).
updated performance numbers posted July 2, 16:47 GMT