search

VEuPathDB / service-multi-blast

Async BLAST+ query service and supporting libraries.
Apache License 2.0
0 stars 0 forks source link

Why is mblast slower than ncbi and legacy blast? #112

Open Foxcapades opened 2 years ago

Foxcapades commented 2 years ago

Why is job 8712B1497978065E708AD3433906A4A8 so slow? This job was running subjobs > 50minutes in runtime before being manually killed. Why are these jobs so slow?

Additionally this full job broke NCBI's blast

slow_query.txt

Foxcapades commented 2 years ago

An individual query that was taking 45+ minutes before it was killed ran in under 2 minutes on NCBI.

Foxcapades commented 2 years ago

We are going to test on the non-active production server tomorrow to reproduce.

Foxcapades commented 2 years ago

Additional performance difference between mblast and the old blast is that the old blast is allowed to use at least 4 cores to execute where mblast is using a single core per blast process.

Foxcapades commented 2 years ago

ask bob for access to dev blast container and run the command from the above job id with the same query using n cores. test faster (4 cores) first to not use caching