Teradata Performance Issue.

[Ultramundane]

Hi,

We are using Ascential DS 7.5.1A with the following patches applied:
75003,
80895,
80559,
80036,
76017,
20959

Occassionally the teradata enterprise stage (about 20% of the time), will have a major performance issue. The stage will suddenly process at 4000 rows/second for a job instead of 250,000 rows/second. When the stage does this it will run horrible for the entire job or until it is killed and restarted. What could cause this? The workload on the Ascential server is 10% and on Teradata it is about 20%. Does anyone know a workaround? The difference in job performance is about 1 hour vs. 6.25 hours.

Ascential has been working on this issue for weeks. The have been able to reproduce, but they don't know what is broken or how to fix it.

[Ultramundane]

Ooops,

The difference is 1 hour vs ~60 hours.

[ArndW]

I've noticed similar issues but have never been able to reproduce. A job that usually processes 15K records per second will, on one invocation, process just a couple of hundred. Nothing is obvious when looking at the system while it is running, but a stop/reset/re-run will get it back up to normal speed and I can't seem to find any possible cause. This happens with server jobs and only occurs very infrequently. It's been bothering me quite a bit but I can't find a way to get a handle on it.

How did Ascential manage to reproduce your problem? Running it often enough to see the effect or have they found some type of way to trigger a "molasses" job?

[Ultramundane]

We had wrote up two cases for Teradata. This was the second case. But, they found this same issue while they were testing for the first case: if the sessions and players are not equal n = n, which forces only one player , the job will just hang.

We encounter these issues with parallel jobs. I don't know enough about the internal workings of each stage to know whether or not they could be caused by the exact same thing.