[mira_talk] Re: miramem under-estimate
- From: Laurent MANCHON <lmanchon@xxxxxxxxxxxxxx>
- To: mira_talk@xxxxxxxxxxxxx
- Date: Mon, 22 Nov 2010 22:08:49 +0100
--
Genome assembly is a big problem moreover when you have more and more reads,
i just got a genome of 400 Mb finished and I can tell i was monitoring
the RAM of the server for many weeks...
I have used the latest ABYSS with 320 millions of reads.
As for other solutions:
- there is a new release of Velvet that should work with large data amounts
- SOAPdenovo yields good results, although the distributed binary does not
allow for all options described in the documentation (some of those are
simply hard coded).
I have tested Mira for Paired-end assembly and with million of reads
mira is a beast that consumes a lot of RAM ...and has a bad memory
footprint.
Laurent --
Bastien Chevreux a écrit :
On Montag 22 November 2010 Tony Travis wrote:
[...]
The machine I'm running the job on has 64GiB RAM, so 55GiB was not a
problem, until the MIRA memory allocation increased to 96.5GiB RAM and
the server started thrashing
Oh, and I forgot: for a couple of hundred Euro, buy a large and fast SSD, put
swap onto that ... and memory trashing is just a minor nuisance and slowdown,
not the big problem it is with hard disks :-)
B.
Other related posts:
