VMWare server 1.0.6 37% faster than Virtual Box 1.6.2

[TerryE]

Bifferos, something else that you might care to try is the issue of cluster alignment. Most guest OSs and Host OSs write pages to disk in 4Kb clusters. To write 4Kb not cluster aligned you need to read 8Kb modify and write back. In general because of the VDI header + the "track" alignment of partitions, the clusters in the VDI file system are not aligned to the clusters of the host file system. So typically on LBA systems the sectors per track is set at the maximum of 63 and the VDI header is 512+4N (rounded up to 512 bdy) bytes long. If you wanted to keep all partitions cluster aligned then you'd want to specify a track size which is a multiple of 4Kb, say 54 sectors not 63, and keep your VDI sizes to (1024N - 128) Mb

I need to have a play here as well.

[tungsten2k]

I re-tried with VB 2.0.6 on Intel OpenSolaris. The result was a whopping 25% slower than Linux, although to be fair, I had problems to figure out how to add my second IDE disk to the ZFS pool, so both OS and virtual images were on the same physical disk for this test.

Judging by the fact that OpenSolaris 2008.11 seemed to respond like molasses on my P4 with 1GB ram, I'd say the extra disk probably wouldn't make much difference.

The ZFS file system requires ample RAM in order to function with acceptable performance. 1GB is the minimum recommended RAM for any deployment utilizing ZFS. Since your system had 1GB total, with the added RAM load of the VM, and the high disk activity of a compile within it, it is likely the host system was struggling for adequate resources. Re-running your benchmark using UFS as the file system, or with additional RAM to compensate for the ZFS file system would be a more appropriate (and interesting !) comparison.

...and agreed, we are all surely interested in your compile benchmark using Windows as the host OS.

-=dave

[bifferos]

...and agreed, we are all surely interested in your compile benchmark using Windows as the host OS.

I tried Windows XP SP2 host and didn't find that much different to Linux. Sorry I never got around to posting this result. I'm interested to know why my observations don't seem to tally with Terry's or other people's. Perhaps everyone else is using dual core machines (which I'm unable to test on at the moment), or perhaps my mistake is in using the clock on the guest machine to base my timing on. My theory is that any time fluctuations in the guest should balance out over a period of a couple of hours, but if there's a bug in that part of the virtualisation it would skew the tests.

PS: Thanks for the info on OpenSolaris.