Multithreading (option "-t 20" ) does not work in vg giraffe.

[Aasesss]

Multithreading (option "-t 20" ) does not work in vg giraffe.

[jltsiren]

Giraffe uses the specified number of mapping threads and one additional output thread. When you run it in paired-end mode, Giraffe starts by mapping some reads using a single thread in order to determine fragment length distribution. This normally takes around one second.

It looks like your run is stuck in that process, with many reads taking a long time to map. That often means one or both of the following:

• The structure of your graph is not what Giraffe expects, and mapping will be very slow.
• Your reads are sorted, with unmapped reads placed in front. Fragment length estimation may be slow, and it will eventually fail.

There is a little more information in the wiki.

[Aasesss]

Giraffe uses the specified number of mapping threads and one additional output thread. When you run it in paired-end mode, Giraffe starts by mapping some reads using a single thread in order to determine fragment length distribution. This normally takes around one second.

It looks like your run is stuck in that process, with many reads taking a long time to map. That often means one or both of the following:

• The structure of your graph is not what Giraffe expects, and mapping will be very slow.
• Your reads are sorted, with unmapped reads placed in front. Fragment length estimation may be slow, and it will eventually fail.

There is a little more information in the wiki.

Thank you very much for your detailed reply!

Acturally, I think this problem may not caused by graph or reads data.

I need to map multiple samples. It works fine when I run sample after sample, but when doing multiple mapping processes at once, only the first process can successfully Implement multithreading. However, these processes all take up the same amount of memory (about 17GB). I'm not sure why this happened.

[jltsiren]

The distance index could be the reason for that. It is a memory-mapped file, and some systems may have performance issues when multiple jobs memory-map the same file. Especially if it's on a network drive.

These are some solutions that could work:

1. Make the distance index (.dist) file read-only.
2. Create a private copy of the distance index file for each process.
3. Create private copies of all index files for each process.

[Aasesss]

The distance index could be the reason for that. It is a memory-mapped file, and some systems may have performance issues when multiple jobs memory-map the same file. Especially if it's on a network drive.

These are some solutions that could work:

1. Make the distance index (.dist) file read-only.
2. Create a private copy of the distance index file for each process.
3. Create private copies of all index files for each process.

Thank you very much! Your advices are really useful.

The multithreading tasks worked well when I make the distance index (.dist) file read-only.