Thanks for your comment. My impression is that CPU performance is ok,
but the disk access suffers. 90% of time spent in kernel is telling that
the problem is probably in disk virtualization or windows driver. Let's
recall that non-kernel mode is running natively on CPU (or am I
mistaken?). It would be nice if someone could make the same test (Qt
compilation on MSVC2005 compiler) on the processor with full VT. If he
will not get better results then there is a serious performance problem
in VBox disk access (Ubuntu host, WinXP guest).
John
Riaan Kok wrote:
Hi folks,
Thanks for the numbers Jan.. was interesting to see. My feeling has
always been that Virtualbox is noticeably slower on some CPU-intensive
operations.. nothing scientific to go by, but as a longtime Gentoo
user you get to develop a gut feel for how long updates should take.
(no jokes, please.. :) ) Nevertheless, Virtualbox is still
impressive and all respects to the team.
Jan, the report that both compile runs had exactly the same duration
suggests that the compiler is not configured to parallelise when
possible.. so it runs in a single 'thread' by default. Using gcc and
make as an example, you'd have to use 'make -j2' to modify the job
engine. The rule of thumb is to use "number of CPUs + 1" if you want
to minimise compile time.
That aside, if you really want to improve the compilation
benchmarking, choose something smaller than Qt (so that the
compilation temporary files don't purge the source files from the disk
cache.. and you don't waste your time!), and run it a few times: keep
the first result separate and average the rest, just to confirm that
there are no weird fluctuations. Keep the compilation
singlethreadedly, that's a good enough test for comparison I think.
Lastly, minimise the window where it is running or hide the
compilation output, make it 'silent', so that you don't mix the
desktop window's scrolling performance in with the compilation test.
regards,
Riaan
------------------------------------------------------------------------
_______________________________________________
vbox-users mailing list
[email protected]
http://vbox.innotek.de/mailman/listinfo/vbox-users
_______________________________________________
vbox-users mailing list
[email protected]
http://vbox.innotek.de/mailman/listinfo/vbox-users
