16.10 - Tips for the Stream Operator - Parallel Transporter

Teradata Parallel Transporter Reference

Product
Parallel Transporter
Release Number
16.10
Published
July 2017
Content Type
Programming Reference
Publication ID
B035-2436-077K
Language
English (United States)
  1. High pack factors can increase the Stream operator’s throughput. When high pack factors cannot be used, increasing the number of sessions is another way to boost the Stream operator’s throughput if the client can support them.
  2. To reduce data load latency and improve real-time availability of single rows, reduce the pack factor.
  3. If input data contains errors, a low pack factor will reduce the overhead of rolling back the request that contains the error, and re-process all error-free rows.
  4. Speed up the Stream operator’s startup by using persistent macros and specifying the Stream operator’s recommended Pack factor from a previous/similar run.
  5. When selecting the number of sessions, consider the total system load at the time the TPT job using the Stream operator is run. When multiple TPT jobs using the Stream operator are running, consider a number of sessions equal to the number of AMPs in the system, or less.
  6. If multiple TPT jobs using the Stream operator may update rows from the same table with the same primary index value, manually partition the data on the primary index of the table, so all rows with the same PI value are directed to the same job. Then also specify SERIALIZE ON to force the rows with the same NUPI value to a single session within that job, further reducing possible contention.
  7. If a TPT job using the Stream operator is doing INSERT operations against a target table with a join index defined, consider directing the TPT job to insert into a non-indexed staging table. An insert/select from this staging table into the base table at regular intervals is likely to be a better-performing approach to updating a table when a join index is involved. Prior to the insert/select, a UNION can be used to make sure data recently inserted into the staging table is included in query answer sets.
  8. Assign the TPT user to a higher priority performance group when the TPT job using the Stream operator runs at the same time as decision support queries, if the TPT completion time is more critical than the other work active in the system.
  9. To ensure that the Stream operator is able to perform single-AMP operations on each input record, include the entire primary index value for the row being updated among the columns passed to the database.
  10. Use the latest versions of client TPT and Teradata CLIv2 and the latest version of Teradata Database.