Hi,
I'm doing some work attempting to improve blastn performance. Using top to examine the runtime of using -num_threads 6 threads on a dual-CPU Xeon server (10 cores/CPU) running blastn 2.2.25 shows 6 threads. numactl --physcpubind=1-6 appears to hold these threads to the specified cores.
But using -num_threads 6 with 2.2.29 has 12 threads created! And -num_threads 3 sees 6 threads created. 3 of the 6 threads appear to be getting destroyed and recreated continuously while blastn runs.
Using numactl --physcpubind=1-6 in this case sees the 6 threads occupying the expected 6 cores. But the CPU utilisation of the 3 cores occupied by the 2nd tranche of 3 threads is much lower which would accord with the PID's indicating that these 2nd 3 threads are being created and destroyed.
(For anyone unfamiliar with top, once the top command is running, just enter 1 and then H to see threads and CPU utilisation.)
Anyone had better luck with threading blastn searches with the current release?
Thanks
David
I'm doing some work attempting to improve blastn performance. Using top to examine the runtime of using -num_threads 6 threads on a dual-CPU Xeon server (10 cores/CPU) running blastn 2.2.25 shows 6 threads. numactl --physcpubind=1-6 appears to hold these threads to the specified cores.
But using -num_threads 6 with 2.2.29 has 12 threads created! And -num_threads 3 sees 6 threads created. 3 of the 6 threads appear to be getting destroyed and recreated continuously while blastn runs.
Using numactl --physcpubind=1-6 in this case sees the 6 threads occupying the expected 6 cores. But the CPU utilisation of the 3 cores occupied by the 2nd tranche of 3 threads is much lower which would accord with the PID's indicating that these 2nd 3 threads are being created and destroyed.
(For anyone unfamiliar with top, once the top command is running, just enter 1 and then H to see threads and CPU utilisation.)
Anyone had better luck with threading blastn searches with the current release?
Thanks
David
