Release 0.8.9 of GPL PV drivers for Windows

Release 0.8.9 of GPL PV drivers for Windows

[James Harper]

Here's the latest version. Took me ages to get to the bottom of what
turned out to be a pretty simple problem - windows can give us more
pages in a packet than Linux can handle but Linux (netback) doesn't
complain about it, it just creates corrupt packets :(

Download from http://www.meadowcourt.org/WindowsXenPV-0.8.9.zip

I'll probably do another release very shortly, mainly to reduce memory
consumption on the tx side so that more interfaces can run at once.

>From the testing I've done, on a UP windows DomU, with iperf options '-l
1M -w 1M', with the iperf server running in Dom0, I get TX throughput of
about 1.5Gbits/second and RX throughput of about 0.5Gbits/second. When I
tried it under SMP it worked, but the performance was horrible. Probably
best if you don't run it under SMP for the moment :)

I did have a test performance of 2.5Gbits/second, but now that I have to
copy the windows buffers into my own buffers to reduce page usage, I
seem to only be able to get about 1.5Gbits/second out of it... This kind
of makes sense given that DomU to Dom0 network performance is going to
be CPU and Memory bandwidth bound.

James

Re: Release 0.8.9 of GPL PV drivers for Windows

[Emre ERENOGLU]

[-- Attachment #1.1: Type: text/plain, Size: 1716 bytes --]

Hi James, great news. As I'm in vacation, I won't be able to test it but
hope it works well.

Did you consider sending an email to the kernel dev guys to inform them
about this issue? maybe it's by design, maybe there's something that can be
improved.

Emre

On Sat, Apr 26, 2008 at 9:49 AM, James Harper <james.harper@bendigoit.com.au>
wrote:

> Here's the latest version. Took me ages to get to the bottom of what
> turned out to be a pretty simple problem - windows can give us more
> pages in a packet than Linux can handle but Linux (netback) doesn't
> complain about it, it just creates corrupt packets :(
>
> Download from http://www.meadowcourt.org/WindowsXenPV-0.8.9.zip
>
> I'll probably do another release very shortly, mainly to reduce memory
> consumption on the tx side so that more interfaces can run at once.
>
> >From the testing I've done, on a UP windows DomU, with iperf options '-l
> 1M -w 1M', with the iperf server running in Dom0, I get TX throughput of
> about 1.5Gbits/second and RX throughput of about 0.5Gbits/second. When I
> tried it under SMP it worked, but the performance was horrible. Probably
> best if you don't run it under SMP for the moment :)
>
> I did have a test performance of 2.5Gbits/second, but now that I have to
> copy the windows buffers into my own buffers to reduce page usage, I
> seem to only be able to get about 1.5Gbits/second out of it... This kind
> of makes sense given that DomU to Dom0 network performance is going to
> be CPU and Memory bandwidth bound.
>
> James
>
> _______________________________________________
> Xen-users mailing list
> Xen-users@lists.xensource.com
> http://lists.xensource.com/xen-users
>



-- 
Emre Erenoglu
erenoglu@gmail.com

[-- Attachment #1.2: Type: text/html, Size: 2340 bytes --]

[-- Attachment #2: Type: text/plain, Size: 137 bytes --]

_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users

RE: [Xen-users] Release 0.8.9 of GPL PV drivers for Windows

[James Harper]

> 
> Did you consider sending an email to the kernel dev guys to inform
them
> about this issue? maybe it's by design, maybe there's something that
can
> be improved.
> 

Not really. Changing it would have a fairly high impact (eg Dom0 and all
Linux DomU's on the system would have to support the higher number or
all sorts of strange things would happen).

The only thing you could thinking about defining as a bug is that
netback didn't complain about it anywhere, it just pretended everything
was okay and put corrupt packets onto the bridge.

James

Re: Release 0.8.9 of GPL PV drivers for Windows

[Florian Manschwetus]

[-- Attachment #1.1: Type: text/plain, Size: 1503 bytes --]

James Harper schrieb:
> Here's the latest version. Took me ages to get to the bottom of what
> turned out to be a pretty simple problem - windows can give us more
> pages in a packet than Linux can handle but Linux (netback) doesn't
> complain about it, it just creates corrupt packets :(
> 
> Download from http://www.meadowcourt.org/WindowsXenPV-0.8.9.zip
Sounds, nice, when i get the time these days, I'll sign this stuff to 
make it acceptable for my 64Bit 2008. Then lets have a look.

Florian
> 
> I'll probably do another release very shortly, mainly to reduce memory
> consumption on the tx side so that more interfaces can run at once.
> 
>> >From the testing I've done, on a UP windows DomU, with iperf options '-l
> 1M -w 1M', with the iperf server running in Dom0, I get TX throughput of
> about 1.5Gbits/second and RX throughput of about 0.5Gbits/second. When I
> tried it under SMP it worked, but the performance was horrible. Probably
> best if you don't run it under SMP for the moment :)
> 
> I did have a test performance of 2.5Gbits/second, but now that I have to
> copy the windows buffers into my own buffers to reduce page usage, I
> seem to only be able to get about 1.5Gbits/second out of it... This kind
> of makes sense given that DomU to Dom0 network performance is going to
> be CPU and Memory bandwidth bound.
> 
> James
> 
> _______________________________________________
> Xen-users mailing list
> Xen-users@lists.xensource.com
> http://lists.xensource.com/xen-users
> 


[-- Attachment #1.2: S/MIME Cryptographic Signature --]
[-- Type: application/x-pkcs7-signature, Size: 3673 bytes --]

[-- Attachment #2: Type: text/plain, Size: 137 bytes --]

_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users