Newsgroups: comp.parallel.pvm
Path: ukc!uknet!EU.net!Germany.EU.net!netmbx.de!zib-berlin.de!uni-paderborn.de!urmel.informatik.rwth-aachen.de!newsserver.rrzn.uni-hannover.de!solaris.rz.tu-clausthal.de!zaphod.in.tu-clausthal.de!stub
From: stub@zaphod.in.tu-clausthal.de (Ulf Bartelt)
Subject: Strange PingPong-Timing on ALPHAs...
Message-ID: <1994May2.185406.15058@solaris.rz.tu-clausthal.de>
Sender: root@solaris.rz.tu-clausthal.de (Operator)
Organization: Techn. Univ. Clausthal
X-Newsreader: TIN [version 1.2 PL2]
Date: Mon, 2 May 1994 18:54:06 GMT
Lines: 24

Hi !

I just got a massive shock looking at some (u)long pingpong timing results...

Pingponging a 1Meg (u)long packet between two ethernetted Alphas coded 
PvmDataDefault ist nearly twice as fast as pingponging it Raw ... ???

I'm not familiar with XDR and such stuff, but the following stuff out
of /usr/include/rpc/xdr.h makes me think XDR takes (u)longs for (u)ints...

-=<cut>=-
 * N.B. and frozen for all time: each data type here uses 4 bytes
 * of external representation.
 */
#define IXDR_GET_LONG(buf)              ((int)ntohl((u_int)*(buf)++))
#define IXDR_PUT_LONG(buf, v)           (*(buf)++ = (int)htonl((u_int)v))
-=<cut>=-

How can this be if sizeof(long) returns 8 on this machines ?

Bye,
     Ulf.

P.S.: ...I'm running my programs under pvm-3.2.3...

