About the performance of single threaded DIAMOND runs.

[SidPalmal4029]

Has anybody observed that in case of single theaded diamond performance , the newest version (2.1.23) is quite slower than the older version (my comparision basis was version 2.1.13). i was using this to run a blastp for AAI dataset where there was a visible difference in the runtimes. Has any body observed any similar phenomenon?

[bbuchfink]

Please let me know your command line, type and size of data, and system you are running on.

[SidPalmal4029]

I am running some benchmarks of a custom pipeline that i am working on. The system that iam running my tests is a shared server of config: Dual-socket Intel Xeon Platinum 8260 system with 48 physical cores (96 threads) and 754 GB RAM. The code that i am running is like doing a reciprocal blast on 2 proteomes : diamond blastp -q A.faa d B.dmnd --evalue 1e-5 --max-target-seqs 1 --more-sensitive --threads 1 -f 6 qseqid sseqid pident -o A_vs_B.tsv --quiet and vice versa.

One thing I want to highlight is i ran this with 16 proteomes (120 calculations) for 3 - 4 times for both the versions, the older version gave me a constant of like completing the calculation in 02:25 - 2.30 mins with 1.21s/it - 1.3s/it , while the newer one give me a range of 02:08<00:00, 1.07s/it at lowest and 0.3:00<00:00, 1.5s/it.

Another dumb question in this regard, will the speed affect if the system has intermittent RAM usage spikes due to other jobs running simultaneously?

[bbuchfink]

Ok, I'll take a look.

Another dumb question in this regard, will the speed affect if the system has intermittent RAM usage spikes due to other jobs running simultaneously?

Yes, anything else running on the same system can definitely affect performance. For a clean benchmark you need the machine to yourself.

[SidPalmal4029]

Ok.. thanks for the insight !!

[bbuchfink]

I can confirm a ~10% performance reduction. Will look into it.

[bbuchfink]

Looked into it and could trace some potential causes but not everything, it may have improved in the latest version but likely not fully fixed for now.

[SidPalmal4029]

Fair enough. Thanks for the update!!

[bbuchfink]

It should be solved in the latest version at least according to my benchmarks, but feel free to let me know what you see. On a side note, setting -c1 always helps for better performance.

[SidPalmal4029]

Preliminary analysis looks like the issue is solved, the runtime difference is not much in the two versions. As of now, my concern was RAM utilization in a shared server . Recently got access to a dedicated compute node, so now i plan to run a full rarge size data set and compare. Once I am done will let you know.