Re: [myrinet] RE: Discarded Data

=> From: "Davis, Harvey" <[email protected]>
=> Hello
=> 
=> I am running red hat linux 6.2, gm-1.2pre16 acroos a number of pentium class
=> machines.
=> 
=> I am sending out some data from the master box, receiving it at a number of
=> worker nodes, doing some processing and then returning the results to the
=> master.
=> 
=> Individually the results come back fine. However when all the machines fire
=> back their results at the same time, the master receives them but only yhte
=> first chunk of data is valid. The rest is filled with zeros.
=> 
=> I have allocated receive buffers equal in number to the number of workers
=> sending messages.
=> 
=> Do you have any ideas why this is occuring.
=> 


=> From: "Greg Lindahl" <[email protected]>
=> To: "Davis, Harvey" <[email protected]>, <[email protected]>
=> Subject: [myrinet] RE:  Discarded Data
=> Date: Mon, 8 Jan 2001 11:59:50 -0500
=> Status: R
=> 
=> > Individually the results come back fine. However when all the
=> > machines fire
=> > back their results at the same time, the master receives them but
=> > only yhte
=> > first chunk of data is valid. The rest is filled with zeros.
=> 
=> You've run out of memory. GM-1.2prewhatever doesn't test that error
=> condition correctly.

Most likely this is correct. gm_register_memory was NOT returning proper error
codes until gm-1.2. I would suggest upgrading to
gm-1.4pre45 (or whatever is current on the ftp site).

	ftp://[email protected]