[Gems-users] Problems building the GEMS 2.1 checkpoints


Date: Fri, 10 Oct 2008 16:18:40 +0200 (CEST)
From: Antonio García Guirado <toni@xxxxxxxxxxx>
Subject: [Gems-users] Problems building the GEMS 2.1 checkpoints
Hi all!

I'm trying to make the checkpoints of the GEMS 2.1 release. Since the
scripts provided didn't work for me, I'm creating the checkpoints almost
by hand.

I'm having problems with Apache. I've adapted some of the scripts used to
build the Apache+Surge checkpoint so I can use them, but I've found that
the mechanism to count ended transactions doesn't work for me.

Surge, when executing, prints some statistics as it runs, showing how many
transactions has been already completed (well, really showing the average
transactions per second). In the code of Surge, in particular in the file
client11s.c, it executes the end_transaction_magic_call() function each
time a transaction completes. When end_transaction_magic_call() is
executed, the end_transaction_callback procedure in mfacet.py file is
eventually called, or at least it's supossed to be so.

I've modified this end_transaction_callback procedure in mfacet.py so it
prints the number of ended transactions.

When executing surge and apache, it seems to work fine at first until it
reaches a point where the number of copleted transactions shown by
mfacet.py and the number shown by client11s.c (Surge) don't match. The
number that client11s.c shows continues growing, but the number shown by
mfacet.py seems to grow much more slowly. I think that those numbers
should match, as each time a transaction is
completed in Surge, end_transaction_magic_call is called.

I am using 40 threads in Surge, and the number of ended transactions that
mfacet.py seems to receive before having problems is around 32000. From
then on, this number increases very very slowly, despite that Surge goes
on completing transactions as fast as before. It's like calls to
end_transaction_magic_call() get lost somehow.

Has anyone seen this behavior before?
Am I right that both numbers should match?
Any idea about how to solve this problem?

Thank you very much.


-- 
Antonio García Guirado
Grupo de Arquitectura y Computación Paralela
Departamento de Ingeniería y Tecnología de Computadores
Facultad de Informática. Universidad de Murcia.
Campus de Espinardo - 30100 Murcia (SPAIN)
email: toni@xxxxxxxxxxx

[← Prev in Thread] Current Thread [Next in Thread→]
  • [Gems-users] Problems building the GEMS 2.1 checkpoints, Antonio García Guirado <=