Archives of the TeradataForum
Message Posted: Fri, 10 Mar 2006 @ 20:06:59 GMT
| Subj: | | Re: MLoad session calculations |
| |
| From: | | Bob Hahn |
Setting or defaulting session count equal to ampcount is too many sessions--usually this is quite a bit more sessions than are required for
optimal throughput. (This applies to Fastload, Multiload, and FastExport).
A graph of a test consisting of multiple utility job runs with varying sessions shows acquisition/return phase throughput improvement at first,
then it levels off and then it gets worse as the session count continues to increase. It usually levels off by 8 - 16 sessions.
One factor for optimal throughput is avoiding client waiting to send data because all the sessions are busy.
Another factor to consider in session count is node parallelism. Utility session messages are sent to the AMPs, not the PEs. All AMPs are
engaged but the receiving AMPs are doing some more work. Again, 8-16 sessions will tend to spread this around adequately.
I would recommend a few tests to validate..
| 
