Hi!
In dom0 (2.6.10-xen0), I''m able to write to lvm-volume at 35 - 40
MB/sec
(megabytes per second). Measured with:
"sync; date; dd if=/dev/zero of=/dev/vg0/lv0 bs=1024k count=100; sync;
date"
But, under domU (2.4.29-xenU), using that same lvm-volume as VBD, I only get
around
10 - 12 MB/sec write performance measured in the same way.
What could cause this? This is a test-box, and I was only running one xenU
domain, and both the dom0 and domU were idle (besides that dd performance
measurement).
Thanks!
-- Pasi Kärkkäinen
^
. .
Linux
/ - \
Choice.of.the
.Next.Generation.
-------------------------------------------------------
SF email is sponsored by - The IT Product Guide
Read honest & candid reviews on hundreds of IT Products from real users.
Discover which products truly live up to the hype. Start reading now.
http://ads.osdn.com/?ad_ide95&alloc_id396&op=click
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/xen-devel
> In dom0 (2.6.10-xen0), I''m able to write to lvm-volume at 35 > - 40 MB/sec > (megabytes per second). Measured with: > "sync; date; dd if=/dev/zero of=/dev/vg0/lv0 bs=1024k > count=100; sync; date" > > But, under domU (2.4.29-xenU), using that same lvm-volume as > VBD, I only get around > 10 - 12 MB/sec write performance measured in the same way.That''s very odd -- I see virtually no loss of throughput in a domU. However, 100MB isn''t really enough to get a decent measure as its likely to all still be in memory -- sync doesn''t do what you think. I usually do measurements using a transfer size of 10x the memory of the system. A couple of other things to try if you still experience poor peformance: * use a 2.6 domU * use a raw partition rather than LVM We do most of our testing using 2.6 for everything, so its possible there a weird 2.4 domU to 2.6 dom0 performance bug. Ian ------------------------------------------------------- SF email is sponsored by - The IT Product Guide Read honest & candid reviews on hundreds of IT Products from real users. Discover which products truly live up to the hype. Start reading now. http://ads.osdn.com/?ad_ide95&alloc_id396&op=click _______________________________________________ Xen-devel mailing list Xen-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/xen-devel
> Both the dom0 and domU were using 128 MB of RAM. I tested with: > "dd if=/dev/zero of=/dev/foo bs=1024k count=2048". That makes > 2 GB of data to > write, so that will definitely not fit in the RAM / buffers. I ran the > "test" three times on each kernel to be sure about the result. > The CPU in the test-box was 3 GHz P4 with 1 MB of l2-cache. > The box (both > the dom0 and domU) were idle besides running the test.Have you tried enabling hyperthreading and running the dom0 and domU''s on different threads? We haven''t done any testing with software raid 5 systems. It''s possible the extra CPU burnt by dom0 is causing bad scheduling interactions. Ian ------------------------------------------------------- SF email is sponsored by - The IT Product Guide Read honest & candid reviews on hundreds of IT Products from real users. Discover which products truly live up to the hype. Start reading now. http://ads.osdn.com/?ad_ide95&alloc_id396&op=click _______________________________________________ Xen-devel mailing list Xen-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/xen-devel
Pasi Kärkkäinen
2005-Feb-17 16:40 UTC
Re: [Xen-devel] 2.4.29-xenU VBD performance (problem)
On Wed, Feb 16, 2005 at 11:14:50PM -0000, Ian Pratt wrote:> > > Both the dom0 and domU were using 128 MB of RAM. I tested with: > > "dd if=/dev/zero of=/dev/foo bs=1024k count=2048". That makes > > 2 GB of data to > > write, so that will definitely not fit in the RAM / buffers. I ran the > > "test" three times on each kernel to be sure about the result. > > The CPU in the test-box was 3 GHz P4 with 1 MB of l2-cache. > > The box (both > > the dom0 and domU) were idle besides running the test. > > Have you tried enabling hyperthreading and running the dom0 and domU''s on different threads? >I enabled hyperthreading and ran dom0 with cpu0 and domU with cpu1. dom0 performance remained the same as previously (93 MB/sec). 2.6.10-xenU performance increased from 58 MB/sec -> 69 MB/sec. 2.4.29-xenU performance remained the same at 23 MB/sec. It seems that enabling hyperthreading and running domains with different (virtual) cpu:s increase performance. Hyperthreading enabled, but running domU with same (virtual) cpu as dom0 caused the performance to be the same as with hyperthreading disabled.> We haven''t done any testing with software raid 5 systems. It''s possible the extra CPU burnt by dom0 is causing bad scheduling interactions. >It might be.. anyway, 2.4-domU performance is really bad.. -- Pasi Kärkkäinen ^ . . Linux / - \ Choice.of.the .Next.Generation. ------------------------------------------------------- SF email is sponsored by - The IT Product Guide Read honest & candid reviews on hundreds of IT Products from real users. Discover which products truly live up to the hype. Start reading now. http://ads.osdn.com/?ad_ide95&alloc_id396&op=click _______________________________________________ Xen-devel mailing list Xen-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/xen-devel
Nivedita Singhvi
2005-Feb-17 17:22 UTC
Re: [Xen-devel] 2.4.29-xenU VBD performance (problem)
Ian Pratt wrote:>>But, under domU (2.4.29-xenU), using that same lvm-volume as >>VBD, I only get around >>10 - 12 MB/sec write performance measured in the same way. > > > That''s very odd -- I see virtually no loss of throughput in a domU. > > However, 100MB isn''t really enough to get a decent measure as its likely to all still be in memory -- sync doesn''t do what you think. I usually do measurements using a transfer size of 10x the memory of the system. > > A couple of other things to try if you still experience poor peformance: > * use a 2.6 domU > * use a raw partition rather than LVM > > We do most of our testing using 2.6 for everything, so its possible there a weird 2.4 domU to 2.6 dom0 performance bug.Right. And to some extent, comparing 2.4 to 2.6 is a bit of an apples to oranges comparison. There are other differences in the kernel path - not just the virtual block driver coming out of domU. thanks, Nivedita ------------------------------------------------------- SF email is sponsored by - The IT Product Guide Read honest & candid reviews on hundreds of IT Products from real users. Discover which products truly live up to the hype. Start reading now. http://ads.osdn.com/?ad_id=6595&alloc_id=14396&op=click _______________________________________________ Xen-devel mailing list Xen-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/xen-devel
> I enabled hyperthreading and ran dom0 with cpu0 and domU with cpu1. > > dom0 performance remained the same as previously (93 MB/sec). > > 2.6.10-xenU performance increased from 58 MB/sec -> 69 MB/sec. > 2.4.29-xenU performance remained the same at 23 MB/sec. > > It seems that enabling hyperthreading and running domains > with different > (virtual) cpu:s increase performance.It''s possible that what is happening heere is that the the MAX_PENDING_REQS is too small for your system with multiple disks in a s/w raid-5. Increasing this value is going to require us to do a little bit of work (as the ring buffer will no longer fit in a page). However, it would be instructive if you could *reduce* this by e.g. 10% and see how if effects performance. If the answer is "it goes to hell" then that''s a good sign that it''ll be worth our effort increasing it.> It might be.. anyway, 2.4-domU performance is really bad..The 2.4 blkfront code is very old and probably needs a spring clean -- all the effort has gone into 2.6 recently. Perhaps someone will volunteer to take a look? Ian ------------------------------------------------------- SF email is sponsored by - The IT Product Guide Read honest & candid reviews on hundreds of IT Products from real users. Discover which products truly live up to the hype. Start reading now. http://ads.osdn.com/?ad_ide95&alloc_id396&op=click _______________________________________________ Xen-devel mailing list Xen-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/xen-devel