Re: [Gems-users] ordering in buffers


Date: Wed, 30 Mar 2005 11:50:19 -0600 (CST)
From: Mike Marty <mikem@xxxxxxxxxxx>
Subject: Re: [Gems-users] ordering in buffers
To add to this, if the RANDOMIZATION flag in ruby/config/tester.defaults
is set to "true" (which is the default), then 100-cycle delays will be
randomly added to messages in unordered buffers to encourage race
conditions and protocol failures to occur for testing purposes.

--Mike

>
> Nauman,
>
> It depends on the cache coherence protocol whether messages on certain
> virtual channels can get out-of-order.  When a buffer is unordered, the
> tester will allow messages to bypass eachother on a point-to-point link.
> For instance, some protocols will deadlock if a GET request moves ahead of
> a PUT request to the directory, which could happen if you set the requests
> queue to be unordered.
>
> Look at the trace closer.  It is not always easy to tell what event causes
> the deadlock, but through some persistence, you should find it.
>
> Brad
>
>
>
> On Wed, 30 Mar 2005, Nauman Rafique wrote:
>
> > I am wondering how much difference does it make if we set some of the buffers to
> > be unordered?
> > I have set a couple of buffers to be unordered, and Tester is reporting a
> > deadlock. I am not really sure why deadlock occurs because from the trace,
> > everything looks fine. And I am simulating with only 1 processor.
> > If I set the buffer to be ordered, it works fine.
> >
> > Thanks.
> > --
> > Nauman
> >
> >
> > _______________________________________________
> > Gems-users mailing list
> > Gems-users@xxxxxxxxxxx
> > https://lists.cs.wisc.edu/mailman/listinfo/gems-users
> >
> _______________________________________________
> Gems-users mailing list
> Gems-users@xxxxxxxxxxx
> https://lists.cs.wisc.edu/mailman/listinfo/gems-users
>
[← Prev in Thread] Current Thread [Next in Thread→]