> I renamed the /usr/src/linux/include/xen subdirectory and I was able to
> then compile the unmodified drivers. (The filename of the PCI .ko file
> was different than the one identified in the README file.) I assumed
> the README was a bit out of date.
It was, and it's now fixed, thanks.
> I then copied the four files over to my Linux VM. And did an insmod on
> the four .ko files. They installed seemingly OK. But when I did my I/O
> test I got no performance difference whatsoever. Did I miss something
> or are my expectations incorrect.
You should certainly notice something! One thing which commonly
catches people out here is that the PV drivers show up as separate
devices to the ioemu ones, and it's easy to end up accidentally
testing the old ioemu devices by mistake.
In the case of network devices, switching over is fairly easy. You
need to create a new entry in the vifs section of your xm
configuration which doesn't have type=ioemu, restart the domain, and
load the PV drivers. You should find you have an extra eth device,
and that should perform a lot better than the old one.
Block devices are a little more tricky, since they're configured to
appear as specific devices (e.g. hda) and the Linux IDE driver doesn't
like letting go. If you're just testing, you can create a new block
device called, say xvda, in your xm configuration file and use that.
The IDE driver won't see it, and so the PV driver can claim it.
Moving your root filesystem to PV devices is a little more involved,
and requires you to put the driver in an initrd and set something like
hda=noprobe on the kernel command line.
Hope that helps,
Steven.
signature.asc
Description: Digital signature
_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel
|
Home