Mailing List Archive

Rok Strnisa writes ("[Xen-devel] Network Throughput Guide"):
> Hi Xen users,
>
> We know that setting up an efficient network can be a daunting task. This is why we put together a network throughput guide:
> http://wiki.xen.org/xenwiki/Network_Throughput_Guide

Wow, that's pretty comprehensive - and complicated.

Perhaps we should also think about how we can make sure the software
works optimally by default.

> Any feedback is more than welcome. Please send it directly to me.

I have kept xen-devel in the CC.

Ian.

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

Hi Ian,

> > We know that setting up an efficient network can be a daunting task.
> This is why we put together a network throughput guide:
> > http://wiki.xen.org/xenwiki/Network_Throughput_Guide
>
> Wow, that's pretty comprehensive - and complicated.

It can be complicated, yes.

> Perhaps we should also think about how we can make sure the software
> works optimally by default.

Making it work optimally by default is, unfortunately, far more complicated. However, we are constantly looking for possible optimizations that would work in most scenarios (and mostly likely not worsen in the remaining ones). If anyone has any ideas regarding this that are not already in the guide, please let me know.

Modifying the TCP parameters within the VM can have a large impact --- I guess the PV tools could do that automatically?

Best regards,
Rok

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

Rok Strnisa writes ("RE: [Xen-devel] Network Throughput Guide"):
> Modifying the TCP parameters within the VM can have a large impact
> --- I guess the PV tools could do that automatically?

I was thinking that patches should be sent to make the guest (and
dom0) use the right tuning values automatically. For the dom0 this
might be done by the toolstack but for guests it should surely be done
by the guest kernel rather than fixed up later by some kind of
separately-distributed PV tools (which nowadays we don't expect modern
Linux guests to need or maybe even have).

Ian.

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

> Rok Strnisa writes ("RE: [Xen-devel] Network Throughput Guide"):
> > Modifying the TCP parameters within the VM can have a large impact
> > --- I guess the PV tools could do that automatically?
>
> I was thinking that patches should be sent to make the guest (and
> dom0) use the right tuning values automatically. For the dom0 this
> might be done by the toolstack but for guests it should surely be done
> by the guest kernel rather than fixed up later by some kind of
> separately-distributed PV tools (which nowadays we don't expect modern
> Linux guests to need or maybe even have).

Good point. I will push the issue forward for the toolstack, and ask around regarding guest kernels.

R

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