usearch version, which one use to get results faster?

Question

[agata88]

Hi all,

I would like to parse 1.5GB database by USEARCH tool. Right now I am using 32bit version and need to divide database into 10 pieces. One part of database is analysed by pipeline 30-40 minutes. So if I run it one after the other I will need around 5h to finish.

Tell me please, If I buy a bigger version of USEARCH 64bit the time will be reduced? If so, how much?

Thanks in advance,

Best,

Agata

[mkulecka]

I don't know much about technical aspects of computers but I think 64bit programmes can use more memory than 32bit ones. The limit for 32bit is 4GB - if you have more RAM than that, you should benefit from 64bit version.

[agata88]

I have 96674MB of memory and 98292MB of Swap. So it will reduce the time of analysis?

[Giovanni M Dall'Olio]

Yes, the 64bit version will allow you to use all the RAM and probably be faster (http://drive5.com/usearch/manual/bitness.html). However 5 hours is not a long time for an analysis, it's up to you to decide whether it is worth. You may even try running the jobs in parallel.

[agata88]

I know it's not very long but I need to do it faster just for one analysis (required delivery time is very short) and I am looking for fast solutions. Thank you ! :)

[Giovanni M Dall'Olio]

What about running it in parallel?

Answer 1

[agata88]

Hi all,

I was talking to the author of usearch, who says that the 64bit version is not faster than 32bit version. So I need to find other solution, maybe skip some unnecessary steps...Thanks for your help.

Best,

Agata

[Istvan Albert]

Your are focusing too much on memory consumption or architecture as correlating with runtime speed. That is almost never the case.

As Giovanni M Dall'Olio stated a few times the key to doing anything faster is to run more of them at the same time. If your computer has more cores (almost all do) then chop the data in more pieces, run the search on each piece separately. Your jobs still complete in N amount of time but more pieces are done in the same amount of time.

[agata88]

Yes, you are right. I divided analysis into 8 cores and it went very fast :) Thanks!