Re: Inconsistent timing results of multithreaded program on an SMP machine.
From: Marcel Zalmanovici
Date: Sun Nov 20 2005 - 05:48:01 EST
<kernel@xxxxxxxxx To: Muli Ben-Yehuda <mulix@xxxxxxxxx>
rg> cc: Marcel Zalmanovici/Haifa/IBM@IBMIL, linux-kernel@xxxxxxxxxxxxxxx
Subject: Re: Inconsistent timing results of multithreaded program on an SMP
20/11/2005 12:39 machine.
On Sun, 20 Nov 2005 21:35, Muli Ben-Yehuda wrote:
> On Sun, Nov 20, 2005 at 09:28:13PM +1100, Con Kolivas wrote:
> > Ok I've had a look at the actual program now ;) Are you timing the time
> > it takes to completion of everything?
> > This part of your program:
> > for (i= 0; i<8; i++)
> > pthread_join(tid[i], NULL);
> > Cares about the order the threads finish. Do you think this might be
> > affecting your results?
> I don't see why it should matter. Depending on the order the threads
> finish, we will always wait in pthread_join until the last one
> finishes, and then do between 0 and 7 more pthread_joins that should
> return immediately (since the last one has already finished).
If it was instant it shouldn't matter. I'm aware of that in theory, but
have certainly been reports of pthread_join taking quite a while happening
a sort of lazy/sloppy way. I don't know why this is the case but I wondered
if it was showing up here.
I've looked through the detailed run results and found this:
Thread DS 0, TID = 7646
Thread DS 0, TID = 7645
Thread DS 1, TID = 7648
Thread DS 1, TID = 7650
Thread DS 2, TID = 7651
Thread DS 2, TID = 7652
Thread DS 3, TID = 7653
Thread DS 3, TID = 7654
Main exit ...
As you can see except the first 2 threads all finished in order they were
created. With the average being at about 16.5 this is a high result and
thread order was almost ideal.
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/