[Spread-users] Limits
Ryan Caudy
caudy at jhu.edu
Thu May 20 13:14:28 EDT 2004
From reading your e-mail, I think you correctly understand the
architecture. The limit on client connections isn't specified in a
#define, but rather is OS and machine-specific. The limiting factor
right now is the event subsystem (see events.h/c), which uses select(),
and hence has a limit based on the maximum fd_set size. On Linux, I
believe this is 997 client connections (Spread uses some additional file
descriptors for internal communication). The may be some much lower
limit in terms of good performance on your system, depending on the
resources available and the types of applications. I'm not sure what
the connection exceptions you're getting in Java mean, without more
information.
Cheers,
Ryan
Paul DeMarco wrote:
> From the spread users manual:
>
> "By default Spread is compiled to support a maximum of 128 machines per
> segment
> and up to 20 segments with a total of 128 machines at most in a
> configuration.
> The 128 machine limit is a hard limit and the protocol has never been
> tested with more than 50-60 daemons. The parameters that control these
> limits are
> MAX PROCS SEGMENT, MAX SEGMENTS, MAX PROCS RING in spread params.h.
> MAX PROCS RING can never be more than 128, so there is not much point in
> changing it. However if you want to only allow one big segment you could
> change MAX SEGMENTS to 1. Or if you want more then 20 segments (say 25),
> each of which has only a few machines, you could set MAX SEGMENTS to 25 and
> MAX PROCS SEGMENT to 10.3"
>
>
> How about number of connections to a single spread daemon? Or am I
> missing it, and that was also provided. If I'm correct its spread
> segments, each with machines/daemon instances, and each daemon has
> connected clients.
>
> If theres no limit, is there a reliable or tested max, I was receiving
> connection exceptions (the java client) when going between 50-75
> connections to a single spread daemon instance.
>
> Thanks.
>
> --Paul DeMarco
>
> _______________________________________________
> Spread-users mailing list
> Spread-users at lists.spread.org
> http://lists.spread.org/mailman/listinfo/spread-users
>
--
Ryan W. Caudy
Center for Networking and Distributed Systems
Department of Computer Science
Johns Hopkins University
More information about the Spread-users
mailing list