Re: [OMPI devel] Affect of compression on modex and launch messages

[George Bosilca]

Ralph,

There are several studies about compressions and data exchange. Few years ago we integrate such mechanism (adaptive compression of communication) in one of the projects here at ICL (called GridSolve). The idea was to optimize the network traffic for sending large matrices used for computation from a server to a specific workers. Under some circumstances (few) it can improve the network traffic, and according to the main author in the worst case it doesn't harm. However, it is still unclear that there is any benefit when the data is reasonably small (which is the case in ORTE). The project is hosted at http://www.loria.fr/~ejeannot/adoc/adoc.html. It's a simple dropin for read/write so it is fairly simple to integrate. On the author webpage you can find some publication about this, publication that highlight the performances of this approach.

  george.

PS: One of these a reference to the paper is available on ACM:

E. Jeannot, B. Knutsson, M. Bjorkmann.



On Apr 4, 2008, at 12:52 PM, Ralph H Castain wrote:

Adaptive Online Data Compression, in: High Performance Distributed Computing (HPDC'11), Edinburgh, Scotland, IEEE, july 2002.

Hello all


are sent, I only implemented compression for those two exchanges.

I have found virtually no benefit to the compression. Essentially, the



the variations I see on this rather unstable machine.

I invite you to try it yourself. You can get a copy of the code via:

    hg clone http://www.open-mpi.org/hg/hgwebdir.cgi/rhc/gather

You will need to configure with LIBS=-lz.

Compression is normally turned "off". You can turn it on by setting:

-mca orte_compress 1

You can also adjust the level of compression:

-mca orte_compress_level [1-9]

If you don't specify the level and select compression, the level will

provided some small amount of better compression, but took longer.


the message and its compressed size so you can see what was done.


whether or not compression is worthwhile.

Thanks
Ralph


_______________________________________________
devel mailing list
de...@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/devel

Based on some discussion on this list, I integrated a zlib-based compression ability into ORTE. Since the launch message sent to the orteds and the modex between the application procs are the only places where messages of any size overhead consumed in compression/decompressing the messages pretty much balances out any transmission time differences. However, I could only test this for 64 nodes, 8ppn, so perhaps there is some benefit at larger sizes. Even though my test size wasn't very big, I did try forcing the worst-case scenario. I included all available BTL's, and ran the OOB over Ethernet. Although there was some difference, it wasn't appreciable - easily within default to 1. From my tests, this seemed a good compromise. The other levels With compression "on", you will get output telling you the original size of Please let me know what you find out. I would like to reach a decision as to

smime.p7s
Description: S/MIME cryptographic signature