[BioC] cannot allocate vector of size 762 Mo ! One more time !

[Kort, Eric]

See below.

Germain Paimparay writes:

> Sent: Wednesday, May 21, 2008 12:41 PM
> To: bioconductor at stat.math.ethz.ch
> Subject: [BioC] cannot allocate vector of size 762 Mo ! One
> more time !
>
>
> Hello,
>
> My name is Germain and I'm Bioinformatician.
>
> I have installed R (v 2.4) on RHEL5, and I try to read 93 CEL
> files with ReadAffy function.
>
> Despite 8Go RAM, I obtain this error :
>
> Error: cannot allocate vector of size 762 Mo ...
>
> I read all topics about this error and try all proposed
> solutions, but it still doesn't work.
>
>
> By unix "top" command, I can seen that % of use memory keep under 20%
>
> Note that I tried with R 2.7 on the same computer but it s
> doesn't work too.
>
>
> "ulimit" command return => unlimited
>
>
> gc command (in R) return
>
> (before ReadAffy)
> > gc()
>           used (Mb) gc trigger (Mb) max used (Mb)
> Ncells 222874  6.0     407500 10.9   350000  9.4
> Vcells  98794  0.8     786432  6.0   358096  2.8
>
> (After ReadAffy)
> > gc()
>           used (Mb) gc trigger   (Mb)  max used   (Mb)Ncells 464170 12.4     818163   21.9    597831   16.0
> Vcells 215236  1.7  224490510 1712.8 280335642 2138.8
>
>
>
> Session Info on Server with troubles
> > sessionInfo()
> R version 2.4.0 (2006-10-03)
> i686-pc-linux-gnu
>

< --8<--snip--->

> I tried with an other server (RHEL4) with 10Go RAM and the
> same R version (2.4) and it's work...
>
> Session Info on Server ok
> > sessionInfo()
> R version 2.4.1 (2006-10-03)
> x86-64-unknow-gnu

< --8<--snip--->

Which one of these two computers is not like the other?

A 32 bit machine can only address memory (at least for a single process...such as R) up to about 4 GB because that is the limit of a 32 bit address.

A 64 bit machine can address over 16 million terabytes (that would be quite a few arrays)--if only you could find a place to put all those RAM sticks.

So regardless of how much memory you have, you are not going to get more than 4GB for your R session on a 32 bit processor (and maybe quite a bit less depending on the OS).

Therefore, it is no wonder you had more success on the 64 bit machine.  Indeed, this week I normalized 1300 hgu133plus2 CEL files via justRMA on my 64 bit server.

So, as has been mentioned previously, you might try justRMA--perhaps with the destructive=TRUE option--if you need to normalize a lot of 32 bit CEL files on a 32 bit machine (avoids the call to ReadAffy).  Or, do it in batches with something like mas5 and stitch the batches together.  But since you apparently have access to a 64bit box, just stick to that.

-Eric








This email message, including any attachments, is for th...{{dropped:6}}