Re: [Xen-users] Performance of live-migration on the xl-stack with 10GB-lines

[Ian Campbell <ian.campbell@xxxxxxxxxx>]

On Fri, 2015-05-15 at 11:12 +0000, Hildebrand, Nils (BIT II 9) wrote:
> Hi Ian,
> 
> netcat suffers from the same limitation: Just a single CPU.

netcat needs to do orders of magnitude less processing than ssh on the
traffic it forwards though. I'd expect netcat to not be the bottleneck
any more.

> I think setting up a socket pair on IP-level would be something that
> would work, since the IP stack is able to use multiple CPUs.

Perhaps what you want is a netcat alike program which instead of just
forwarding the data uses something like splice(2) to join the input fd
to the output socket fd and let the kernel do its thing?

I'm not sure if such a thing exists, or if it is even possible. I think
netcat gets pretty close though.

> Or is this the same as passing an empty "-s" option?

According to the docs -s with no option looks for something to run (in
$PATH I suppose) which is named for the target host.

> The manpage is not that clear in that point. What does xl expect "the
> other end" to do when specifying "-s"?

With or without -s the expectation is that the processes stdin/stdout
will be piped into the stdout/stdin of an "xl migrate-receive" process
on the other end.

> Is there a technical detail description about how "xl migrate" works
> on sending/receiving side?

I'm afraid not, I don't think.

Ian.


_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxx
http://lists.xen.org/xen-users