Hi all,
We are experiencing a relative slow ODBC throughput from TD to Microstrategy and I am wondering if this is "as good as it gets" or if the throughput is below standard.
The challenge is that TD is performing nice, but the transfer time of the output of the query is taking a significant time of the total run time of the report. We have tested with the following setup:
- 500.000 rows
- 35 columns
Results:
- Roughly 33.000 rows/s
- Roughly 9.4 MB/s
- Query execution time: 2s
- Transfer time: 15s
- Processing time on Microstragy: 10
- Total time: 27s
Currently this is our environment:
- TD 2700 14.10.06.03 - connected with 2 nodes wtih 1 Gb to a 10 Gb core network. No firewall in between, in the same data center.
- Microstrategy 9.4.1, previously on a TD TMS, image moved to our own VM hosting with SLES 11. Microstrategy 10.x offers native support for TPT Fast Export, but we are not yet there...
- ODBC verson on the Microstrategy machine is 14.10.00.09-1 and is the latest available. Micostrategy can only use the 32-bit driver on Linux.
Tested is:
- scp with a random generated file from TD to Microstrategy shows a througput of 65 MB/s. Backup of TD via ABU reaches close to 200 MB/s, close to the limit of the physical network connection of the nodes
- Version of odbc (.9 gave a 1/3 improvement, in the release notes was a reference to better performance: "ODBC-12455 -- The ODBC Driver for Teradata now uses the -O2 compiler optimization level for better performance on the Linux platform"). But we are stuck on the 14.10.x stack as 15.x is not certifified with Microstrategy. I am aware that TTU 15.10 is backwards compatible with TD 14.10.
- MaxRespSize is on maximum, all kinds of settings were tested, but only on very low settings the transfer times increased
- Various settings such as NoScan, DisablePREPARE, TCPNoDelay etc did not influence the transfer time significantly (milliseconds difference, could be related to other things in the environment).
- Encryption did impact the transfer negatively (as expected).
- DNS or direct connection on ip-addres: no signicant difference
Should I check more things? Am I overlooking something? Or is this really the normal performance of odbc...
Forums: