On Sat, 2 May 2009 21:34:38 -0400, Scott T. Harder <[email protected]> wrote:
>Wow, Pat. Is there anything you can do to try to create an environment >such as is with the long x-missions? I hate to be trite, but it sounds >like you have too many things going on and you need to take them one >at a time. >... My description was probably not clear. There really isn't too much going on except in my words. Let me try to restate the problem. 1. Last week transmissions of large datasets between 2 nodes were taking nearly 10 times longer than they had a couple weeks earlier. 2. On the same day as these slow transmissions, the same dataset sent to the same remote node from a "development" C:D went at the old, fast speed. 3. The only difference we could find between out production and development C:Ds was checkpointing. The prod C:Ds have it turned on by default. The dev C:Ds do not. 4. We turned checkpointing on on one of our dev C:Ds and the transmission rate was slow - same as the rate from the the prod C:D. 5. We turned checkpointing back off but the transmission rate stayed bad. Other environmental conditions: Neither end was cycle constrained. The dev and prod C:Ds run on 2 LPARs on the same processor. Only a few very large transmissions were (noticably) affected. There were no changes to CONNECT:Direct over the past few weeks. I don't know enough about CONNECT:Direct checkpointing to know whether the test in #4 was valid. There transmissions replaced existing copies of the same dataset, and most of these transmissions were not allowed to go to completion during the test. (After all, the slow transmissions took over 3 hours to complete.) Perhaps the retransmission of a previously checkpointed dataset uses checkpoint even if not specified in the retransmission - the receiving end probably knows it is getting another copy of a previously incomplete transmission that had been checkpointed. Maybe someone that understands CONNECT:Direct checkpointing would comment on that. Pat O'Keefe ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [email protected] with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
