[parsec-users] dedup

Polina Dudnik pdudnik at gmail.com
Tue Sep 9 10:19:26 EDT 2008


Thank you. You answered my questions.

Polina

On Mon, Sep 8, 2008 at 6:06 PM, Christian Bienia
<cbienia at cs.princeton.edu> wrote:
> Hi Polina,
>
> Dedup uses pipelining because it makes sense and because the real-world
> commercial programs after which dedup was modeled use pipelining, too. Other
> parallelization strategies are conceivable, but we believe this would result
> in a less realistic workload.
>
> I'm not sure what you mean by starting all threads at the same time. Dedup
> already does that. You also have to consider that the input is a sequential
> byte stream which first has to be fragmented to create independent work
> units for the threads. So you can't start working on the entire input at
> once unless you cheat and do the fragmentation offline. Does that answer
> your question?
>
> - Chris
>
>
> -----Original Message-----
> From: parsec-users-bounces at lists.cs.princeton.edu
> [mailto:parsec-users-bounces at lists.cs.princeton.edu] On Behalf Of Polina
> Dudnik
> Sent: Monday, September 08, 2008 5:16 PM
> To: parsec-users at lists.cs.princeton.edu
> Subject: [parsec-users] dedup
>
> Hi,
>
> I have a question about the dedup benchmark. It uses pipelining to
> parallelize the compression. Does dedup use pipelining because the
> suite needed a pipelined benchmarks, or is pipelining crucial to the
> implementation? Is it possible to start all threads at once and let
> them run in parallel, or would that produce wrong results? Thank you.
>
> Polina
> _______________________________________________
> parsec-users mailing list
> parsec-users at lists.cs.princeton.edu
> https://lists.cs.princeton.edu/mailman/listinfo/parsec-users
>
> _______________________________________________
> parsec-users mailing list
> parsec-users at lists.cs.princeton.edu
> https://lists.cs.princeton.edu/mailman/listinfo/parsec-users
>


More information about the parsec-users mailing list