From: hjstein@bfr.co.il (Harvey J. Stein)
Newsgroups: comp.parallel.mpi,comp.parallel.pvm
Subject: Re: Poor MPI performance with collection of Linux workstations
Date: 17 Mar 1999 10:27:39 +0200
Organization: Unspecified Organization
Message-Id: <m2r9qor05g.fsf@blinky.bfr.co.il>
References: <36ED325A.69D8C9B8@rrsg.ee.uct.ac.za>
    <7cn7up$4tq$1@grandprime.binc.net>
Cc: hjstein@bfr.co.il
Xref: ukc comp.parallel.mpi:4764 comp.parallel.pvm:8131


Leigh Orf <orf@go.spam.yourself.com> writes:

 > Mark Gebhardt <mark@rrsg.ee.uct.ac.za> wrote:
 > 
 > > Hi all
 > 
 > [big snip]
 > 
 > > Under certain conditions, my MPI implementation gives wildly varying
 > > execution times - something that the PVM version never did. Here's
 > > what happens:
 > 
 > I don't have an answer to your problem, but I have a similar problem.
 > I'm running Linux and have just recently upgraded all machines to the
 > 2.2.3 kernel, hoping my problems would be solved. My problems persist. I
 > have nine machines, one of which is a 4-processor box.

Are they small packets?  If so, here's a patch from the linux-kernel
mailing list that might help:

   From: Andrea Arcangeli <andrea@e-mind.com>
   Date: Fri, 26 Feb 1999 18:18:25 +0100 (CET)
   Subject: Re: Linux 2.2.2 TCP delays every 41st small packet by 10-20 ms

   On Thu, 25 Feb 1999, Josip Loncaric wrote:

   >The conclusion is this: in Linux TCP, every N-th small packet is delayed
   >by 1-2 "jiffies" (defined by the 100 Hz system clock).  For Linux kernel

   This 1-2 jiffies made me thing again at the PSH thing. Probably I am
   completly out but could you try this patch?

   Index: tcp_input.c
   ===================================================================
   RCS file: /var/cvs/linux/net/ipv4/tcp_input.c,v
   retrieving revision 1.1.2.13
   diff -u -r1.1.2.13 tcp_input.c
   - --- tcp_input.c	1999/02/26 17:39:46	1.1.2.13
   +++ linux/net/ipv4/tcp_input.c	1999/02/26 17:40:14
   @@ -55,6 +55,7 @@
     *					work without delayed acks. 
     *		Andi Kleen:		Process packets with PSH set in the
     *					fast path.
   + *		Andrea Arcangeli:	Quickack if received a tiny PSH packet.
     */

    #include <linux/config.h>
   @@ -133,12 +134,8 @@
	   /* Tiny-grams with PSH set artifically deflate our
	    * ato measurement, but with a lower bound.
	    */
   - -	if(th->psh && (skb->len < (tp->mss_cache >> 1))) {
   - -		/* Preserve the quickack state. */
   - -		if((tp->ato & 0x7fffffff) > HZ/50)
   - -			tp->ato = ((tp->ato & 0x80000000) |
   - -				   (HZ/50));
   - -	}
   +	if(th->psh && (skb->len < (tp->mss_cache >> 1)))
   +		tcp_enter_quickack_mode(tp);
    } 

    /* Called to compute a smoothed rtt estimate. The data fed to this


   Andrea Arcangeli


   Please read the FAQ at http://www.tux.org/lkml/

-- 
Harvey J. Stein
BFM Financial Research
hjstein@bfr.co.il