Linux mprime client v22.8 problem
 

George,

I am running the above client on several nodes, including some servers (IBM DB2 7.3 latest patch level, Apache, Tomcat and others). On the DB2 server I had to remove the client as the whole machine froze up. On the Tomcat machine it slows down the response time by something like 20 milliseconds (=20%), but that is within acceptable limits, on the Apache machine there are no noticable effects. I tried to run with only one client, but same results. I also tried running factorization only and again the same results.

All the machines are running Redhat 7.3 with the latest Redhat patch levels (inc libraries). These machines have a large amount of Ram; 1.5GB up to 4GB and 2 cpus. The DB2 machine is a dual PIII Xeon with 1MB cache, the other 2 servers are dual 1GHz PIIIs.

I believe I had the same problem with earlier versions of the client, but then the machine were not in production, so a freeze didn't bother me too much. Now they are in production and I cannot have machines freezing at any time.

Is there any way I can figure what goes on with the client ( a debug log or something) ?

You're much more a Linux expert than I am. I don't know why a nice'd client would cause these problems.

I can send you an mprime with symbols in it or you can build an mprime from scratch using the source at http://www.mersenne.org/source.htm
I presume either of these might help you debug mprime.

Out of curiosity, if you set UsePrimenet=0 in prime.ini do you get the same troubles? That would tell us if we have a communication problem or an LL problem.

I am fairly certain it is not a communication problem, as the freezing of the machine happens fairly "slowly and gradually". The first to go is keyboard and mouse, then screen, and then after a while the communication system. If I have a remote window open on the machine and spot it happening I can save the situation by killing the mprime client and restarting it.

On average it does take up to a week for the machine to freeze.

Alf

It sounds like mprime is leaking resources (memory, file handles, sockets, etc). Do you know of any linux tools that might clue us into the problem?

I'll check it out, I know there are some, but I need to see if they'll work in my setup

Alf

George,

[quote]It sounds like mprime is leaking resources (memory, file handles, sockets, etc). Do you know of any linux tools that might clue us into the problem?[/quote]

There are several tools for Linux, as one could expect. Some of them seems to be highly specialized and some are more generic.

I would think the best thing would be to try them one by one, ie for memory leaks, file handles, network handles etc. That would make tracing over such a long period easier as well.

I'll run the traces on one of the systems if you send me an mprime with symbols in it.

Alf

I believe the tool you want to try is called "Valgrind" and is available from Source Forge or Freshmeat.