> On Saturday April 26 2008 10:12:33 pm James Harper wrote:
> > Hmmm... uninstall.bat is a work in progress and probably shouldn't
have
> > been included. I suspect that if you rebooted after running
> > uninstall.bat but before running install.bat you would have had an
> > unbootable system.
>
> Well, for the first time since 0.8.4, I got winxp to boot w/ /gplpv,
using
> 0.8.9. I booted w/o /gplpv and then ran uninstall.bat. Because of the
> warning
> above, I ran install.bat right away w/o a reboot. The final screen
came up
> and said all drivers were updated except xennet, which was 'Ready to
Use'.
> Then I rebooted w/ /gplpv, and the Find New Hardware Wizard came up
> automatically and guided me through installing xennet w/o a hitch.
>
> I had previously disabled all the new features in Device Manager's Xen
Net
> Device Driver Properties' Advanced Tab. (This was done by copying over
the
> 0.8.4 files so I could boot w/o a BSOD.) An 'iperf -c dom0-name -t 60'
> came up with 27.3 Mbits/s. I then proceeded to turn on each feature
one at > a time in the Advanced tab, and rebooting w/ /gplpv. Device
Manager
> invariably hung after enabling each feature, which caused the reboot
to
> hang.
Ouch. I was confident I'd fixed all of those hanging problems. I can
enable and disable those settings all I like and it never misses a beat
- the driver shuts itself down and restarts without a problem. Can you
tell me more about your test system?
> All measurements are with vcpus=2, unless noted otherwise.
>
> After adding enabling Checksum Offload, iperf gave 30.5 Mb/s.
>
> After adding setting 61440 for Large Send Offload, iperf gave 25.3
Mb/s.
>
> After adding enabling Scatter/Gather, iperf gave 25.1 Mb/s.
Scatter/Gather disabled isn't really a tested configuration, but the
others should work, although LSO disabled when Dom0 has it enabled could
be problematic...
What system are you running? My main test machine is a 1.8GHz dual core
AMD 1210 system, and I get 1GBit/sec TX and 200MBit/sec RX with all
features enabled. So your figures are lower than I'd expect, given that
any system that supports HVM should be at least comparable to my test
machine in terms of horsepower...
With LSO disabled my testing drops to 283 TX and 6 RX...
> So there are minor variations with and w/o the various options, but on
the
> whole, much better than the last version I could test, 0.8.4.
Well... that's something :)
>
> Doing 'iperf -c dom0-name -l 1M -w 1M' gives 28.8 Mb/s, and reversing
the
> direction (winxp as iperf server) gives 30.2 Mb/s.
You can achieve the same results with -r to do a TX test followed by an
RX (eg client/server roles reverse) or -d to do the TX and RX tests
concurrently.
> Going down to vcpu=1, dom0 as server gives 27.1 Mb/s, and domu as
server
> gives
> 35.7 Mb/s, so there is not a lot of difference between 1 and 2 vcpus
for
> me.
With all features enabled?
> Nice improvements. I will test disk i/o w/ iometer later.
I'll be interested in the results, but the disk stuff hasn't changed in
a while.
Thanks for the feedback!
James
_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users
|
Home