Newsgroups: comp.parallel.pvm
Path: ukc!uknet!pipex!sunic!EU.net!news.funet.fi!news.cs.tut.fi!cs.tut.fi!moga
From: moga@cs.tut.fi (Moga Alina)
Subject: deadlock causes?
Message-ID: <1994Feb28.133627.16466@news.cs.tut.fi>
Keywords: multiple communication among processes
Sender: usenet@news.cs.tut.fi (#Kotilo NEWS system )
Nntp-Posting-Host: ruskosotka.cs.tut.fi
Organization: Tampere University of Technology, Finland
Date: Mon, 28 Feb 1994 13:36:27 GMT
Lines: 23

I implemented an algorithm which uses the master/slave model.
And seldom it happens to end stuck in a deadlock.
The slaves are synchronized via messages by the master, 
but it's not a global synchronization.
Clusters of slave processes send messages with one tag
(equal with the current synchronization level) to the master.
At one moment of time there are many clusters sending messages 
with different tags, but the master receives only messages with
the minimum tag existing in the system.
Is it possible that the program end in deadlock cause to a big 
number of messages sent and not received ? 
I mean if those processes with bigger synchronization level, e.g. 10 
are faster than those with smaller one, e.g. 5 and send messages to
master which expects only messages with tag 5 to proceed, and then it 
accepts messages with 10, can produce a deadlock?
Sometimes I got the following messages in /tmp/pvml.<uid>

[t80080000] netinput() pkt from t4002b for t40040466 scrapped (not us)
[t80040000] loclinput() pkt src t40009 dst t40008 scrapped (no dst)

Please, send me any hint or idea about this.

Alina.

