Timescale performs poorly on last-loc

[slhuang]

We generate and load IOT data into Timescale following exactly the readme.md (https://github.com/timescale/tsbs). However, we observe the query performance is very different from what reported in https://www.timescale.com/blog/how-to-benchmark-iot-time-series-workloads-in-a-production-environment/

For instance, the last-loc query takes around 14 minutes to finish on our server (with shared_buffer=128G, work_mem=4G, and maximum_worker_number=64). While the running time is around 1s in the blog. May we ask what else we need to tune in order to get that performance? Thanks!

FYI, we tried different physical layout configurations during data loading:

1. --in-table-partition-tag=true --chunk-time=8h --field-index-count=1 --do-create-db=true --do-abort-on-exist=false
2. --in-table-partition-tag=true --chunk-time=30m --field-index-count=1 --do-create-db=true --do-abort-on-exist=false
3. --in-table-partition-tag=true --partitions=10 --chunk-time=30m --field-index-count=1 --do-create-db=true --do-abort-on-exist=false The first configuration is the same as what is in the readme.md. However, none of the above configurations leads to comparable performance as reported in the blog post. Would you please kindly provide some guidance to us? Thank you!

[dssysolyatin]

I suspect that TimescaleDB benchmark code still create wrong index for --in-table-partition-tag=true I described it before inside https://github.com/timescale/tsbs/issues/182

[slhuang]

I suspect that TimescaleDB benchmark code still create wrong index for --in-table-partition-tag=true I described it before inside #182

I see. Thanks for the information! Were you able to fix it and speed the query up? e.g., manually create the correct index (tag_id, time).