On Thursday 21 February 2008 03:08:32 pm Stephan Seitz wrote:
> I wasn't able to get Windows XP Professional x64 running with gplpv until
> James released 0.8.0 if his great drivers.
>
> So my answer is a bit delayed:
Ok, here's my results.
Equipment: core duo 2300, 1.66ghz each, sata drive configured for UDMA/100
System: fc8 32bit pae, xen 3.1.2, xen.gz 3.1.0-rc7, dom0 2.6.21
Tested hvm: XP Pro SP2, 2002, tested w/ iometer 2006-07-27 (1Gb \iobw.tst, 5min
run)
iobw.tst on linux seems to grow to all available disk space - mine was 7Gb for
the first runs w/
/gplpv, then just before I rebooted w/o /gplpv, I created a 4Gb file with dd:
dd if=/dev/zero
of=/iobw.tst bs=2048 count=2097152. Since the client doing the actual disk io
is always
local to the computer being tested, I would expect network io to be negligible,
and the pv
net drivers were always enabled. (Physical eth0 is 100MB/s.)
file backed vbd on local disk, w/ and w/o /gplpv (James Harper's 0.8.1):
pattern 4k, 50% read, 0% random (what do the minus signs mean?)
(dynamo is the windows or linux client doing the actual work)
dynamo on? | io/s | MB/s | Avg. i/o time(ms} | max i/o time(ms) | %CPU
domu w/gplpv| -83.1 | -0.32 | -51.25 | 0 | 45.99
dom0 | 1485.9 | 5.8 | -19.19 | 597.5 | 16.05
2nd run:
domu w/gplpv| 273.0 | 1.07 | 431.52 | 0 | 32.27
domu w/qemu | 251.6 | 0.98 | 10.05 | 0 | 28.44
dom0 w/7Gb | 1040.1 | 4.06 | 0.96 | 395.4 |
0(domu not involved)
dom0 w/4Gb | 808.1 | 3.16 | 1.24 | 977.1 | 0
(2nd dom0 numbers from when booted w/o /gplpv)
pattern 32k, 50% read, 0% random
domu w/gplpv| 161.6 | 5.05 | -3.32 | 0 | 19.80
domu w/qemu | 109.0 | 3.41 | -8.93 | 0 | 25.35
dom0 w/7Gb | 140.7 | 4.40 | 7.10 | 467.6 | 0
dom0 w/4Gb | 159.3 | 4.98 | 6.28 | 270.1 | 0
Conclusion: gplpv disk drivers severely lag dom0 disk performance for small
pattern io, but is similar for large pattern io.
Now for the same tests with Halsign. This time, the file backed vbd is on a
samba share mounted on dom0.
dynamo on? | io/s | MB/s | Avg. i/o time(ms} | max i/o time(ms) | %CPU
domu w/hals.| 331.3 | 1.29 | -31.78 | 0 | 39.02
domu w/qemu | 460.2 | 1.80 | -5.85 | 0 | 42.35
dom0 w/4Gb | 958.1 | 3.75 | 1.04 | 187.8 | 0
dom0 w/4Gb | 1080.5 | 4.22 | 0.92 | 192.2 | 0
(2nd dom0 numbers from when booted w/o /pv)
pattern 32k, 50% read, 0% random
domu w/hals.| 81.1 | 2.53 | 49.94 | 0 | 36.36
domu w/qemu | 74.6 | 2.33 | 10.11 | 0 | 32.00
dom0 w/4Gb | 138.9 | 4.34 | 7.20 | 340.8 | 0
dom0 w/4Gb | 148.0 | 4.62 | 6.76 | 228.6 | 0
Conclusion: Even tho' samba might be expected to be slower than a local disk,
the small pattern numbers are very similar to gplpv, possibly because the io
is being offloaded on the samba host. The domu 32k pattern numbers are less
than dom0, unlike gplpv, but that may be because my samba buffers are only
8k.
And now, just for grins, the results from putting a dynamo client on a pv fc8
domu (I had to disable the firewalls on both domus):
pattern? | io/s | MB/s | Avg. i/o time(ms} | max i/o time(ms) | %CPU
4k, 50% read| 931.2 | 3.64 | 1.07 | 294.0 | 0
32k,50% read| 191.8 | 5.99 | 5.21 | 389.8 | 0
The numbers are very similar to dom0.
_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users
|
Home