Trimmomatic bad parallelize

[Rammaria]

Hello!
I have 4 physical and 8 virtual cores at my computer and I hoped Trimmomatic can use all of them. I tried to run Trimmomatic with '-threads 4', '-threads 8', '-threads 16' and without '-threads' option, but I had the same result every time: Trimmomatic created a lot of processes, but only one of them took near 100% of 1 CPU, and other processes took near 5 % of other CPU. How to make it to get all cores?
I use Ubuntu 14.04 and Trimmomatic 0.32.



Thank you in advance.

[Brian Bushnell]

Can you describe your input data and what operation you are doing to it?

[tonybolger]

Hello!
I have 4 physical and 8 virtual cores at my computer and I hoped Trimmomatic can use all of them. I tried to run Trimmomatic with '-threads 4', '-threads 8', '-threads 16' and without '-threads' option, but I had the same result every time: Trimmomatic created a lot of processes, but only one of them took near 100% of 1 CPU, and other processes took near 5 % of other CPU. How to make it to get all cores?
I use Ubuntu 14.04 and Trimmomatic 0.32.



Thank you in advance.

Hi.

If you see the threads created, then you can be reasonably confident you have used the arguments correctly. The 'threads' argument (if >1) is actually the number of worker threads, so you should see those, the main thread, and one thread for each input or output file, plus whatever threads the JVM creates for itself. Recent versions will create enough workers to occupy all CPU cores while previous versions defaulted to using just the main thread for everything.

The chances are that you a bottlenecked on I/O or compression/decompression rather than the calculations for trimming. The first can be checked using iostat (or equivalent), while the second can be checked by not using compressed files and see if it changes.

These low-cpu usage situations often occur if you are using a very lightweight trimming pipeline, but in such cases, the processing should be fast anyway.