On Friday May 23 2008 09:52:22 am James Harper wrote:
> I have just uploaded a new binary release of the PV drivers for Windows.
Despite not having a keyboard when booting w/ /gplpv, I was able to perform
some benches comparing 0.8.9 to 0.9.1.
Since iometer is gui based, and my test config is saved, that was easy. Dom0
performance did not change, nor would it be expected to. On domu, for the 4k
pattern, there was a 20% loss on multi-threaded testing (workers on domu and
dom0 at the same time), and a 25% loss on single threaded operation (dom0
runs the 'idle' task). There was no loss on the 32k pattern.
For iperf, I had to setup some batch files ahead of time that I could click on
in My Computer. For a tcp test, 'iperf-1.7.0 -c dom0-name -t 60 -r':
domu->dom0: 31 Mb/s (no loss)
dom0->domu: 36 Mb/s
For a udp test, requesting a 10Mb/s bandwidth, 'iperf-1.7.0 -c dom0-name -t
60 -r -b 10000000' gave:
domu->dom0: 2.6 Mb/s
dom0->domu: 9.9 Mb/s
Someone else on the list noticed that dom0-> domu performance was faster (and
in fact, what was expected) than the reverse direction. However, dom0 warned
that it did not receive the final ack on the dom0->domu test, and the domu
client hung, so there was probably a lot of lost packets.
All in all, *very* good for a massive rewrite of the drivers.
Btw James, when booted w/ /gplpv, the Hardware tab of the Keyboard Control
Panel icon says there is no device/driver present. When booted w/o /gplpv,
it, and Device Manager, says 'Standard 101/102-Key or Microsoft Natural PS/2
Keyboard'. And yet, Device Manager doesn't report xenhide as one of the
Xen-users mailing list