[Spread-users] TCP or UDP, spread-to-spread, spread-to-client, TCP_NODELAY

[Jonathan Stanton]

Daemon-Daemon uses UDP
Daemon-Client uses Unix Sockets or TCP sockets

Since TCP_NODELAY is a TCP specific socket option, it doesn't apply to unix domain 
sockets. I don't think anything like NAGLE would be used on a local connection 
like unix sockets are.

The very simple 'r.c' and 's.c' (compiled into spsend and sprecv by using "make 
spsend and make sprecv) that are included in the spread source can be used for UDP 
packet debugging. General purpose tools like netperf are also useful.

Jonathan

On Tue, Nov 28, 2006 at 05:22:46PM -0600, Matt Garman wrote:
> I know I read this before, but can't remember where.  What protocol
> (TCP or UDP) does spread use for daemon-to-daemon communication and
> for daemon-to-client communication?  My (very foggy) memory says it's
> UDP for the former, and TCP for the latter.  Is this correct?
> 
> Also, I noticed (via looking at libspread/sp.c) that the TCP_NODELAY
> socket option is set for internet sockets, but not for Unix sockets.
> Just curious: does anyone know if nagling is always disabled for Unix
> sockets (at least on Linux)?  (My intuition says that it's really not
> applicable for a Unix socket.)
> 
> We're not having any problems with spread, but we are having more
> general latency-related (but very mysterious) network problems.  We
> thought spread would be a good application to look at to get an idea
> of how sockets are generally created and what impact TCP_NODELAY has
> on them.
> 
> Thank you,
> Matt
> 
> _______________________________________________
> Spread-users mailing list
> Spread-users at lists.spread.org
> http://lists.spread.org/mailman/listinfo/spread-users

-- 
-------------------------------------------------------
Jonathan R. Stanton         jonathan at cs.jhu.edu
Dept. of Computer Science   
Johns Hopkins University    
-------------------------------------------------------