The strategy to run daa2rma

Hello,

I am using daa2rma to run analysis on paired-end data. There is a upper limit for run-time in the HPC server and the daa2rma is running out of the upper bound run-time.

In this case, the analysis will not be completed within the period of time.

Is there any solution to get over this issue?

For example, support multiple threading to speed up or anything else?

Many thanks,

I am also interested in the same. Please let me know if there is any update.

Please don’t use daa2rma, please use daa-meganizer.
This is much faster and produces much smaller files.
In fact, this doesn’t produce a new file but rather extends the existing DAA files. The original content of the DAA file is not changed and you can remeganize the files again later, or use them for other purposes (for example you can “diamond view” them) after mechanization.