[Spice-devel] Is Nagle disabled?

Yaniv Kaul ykaul at redhat.com
Mon Jun 27 21:54:17 PDT 2011


On 06/28/2011 01:42 AM, John A. Sullivan III wrote:
> Hello, all.  I put on my network engineer hat today and started doing
> packet traces of SPICE to see what we could do from a systems
> perspective to reduce the perceived latency.
>
> SPICE does not appear to be fragmenting - in fact the do not fragment
> bit is set so I assume we are doing MTU discovery somewhere along the
> way.
>
> Somewhere in the back of my mind, I remembered something about
> optimizing iSCSI connections by disabling the Nagle algorithm.  I don't
> think we can do this from the command line in Linux as it is not exposed
> in sysctl (at least I didn't see the no_delay option in /proc).  I
> gather it is set when the socket is created.  Does SPICE disable Nagle
> in both client and server when it creates the socket? Thanks - John
>
> _______________________________________________
> Spice-devel mailing list
> Spice-devel at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/spice-devel

If you'll grep the code for TCP_NODELAY, you'll find it is doing it in 
several places. The input channel, for instance, unconditionally sets it.
I seem to remember it is done when we detect WAN connection - but 
perhaps the code has changed since and for some cases it is done 
unconditionally.
Y.



More information about the Spice-devel mailing list