ECC vc non-ECC RAM for NGS Workstation

[swNGS]

Hi,

I am in the process of spec'ing a workstation for processing human FASTQ resequencing data from GAII and MiSeq analysers. All pipelines will be alignment to reference genome and variant calling only (no denovo assembly). I think I'll be looking in the region of 32 GB memory.

I've been looking at some of the Dell workstations, which all seem to use ECC ram as far as I can see. I understand that ECC ram produces less errors, but on the other hand is MUCH more expensive than non-ECC ram. Would it be a problem to not use non-ECC memory to reduce the cost ?

Thanks,

Chris

[gringer]

ECC is insurance against future damage to your data and/or results. ECC ram will save you quite a bit in debugging time (i.e. debugging money) if sometime down the line there are a few errors that play an important role in the outcome of an analysis. Bad RAM is much more likely when there's more of it, which is part of the reason why ECC is used.

[swNGS]

Does anyone hav any real world examples of either scenario?
Eg where using non-ECC memory has caused a problem?.
In terms of scale I doubt I'll go over 32GB

[gringer]

Here's a post about where a flipped bit created a problem building a program:

http://blogs.oracle.com/ksplice/entry/attack_of_the_cosmic_rays1

I remember reading another blog post that was more related to genomics that discussed a similar situation (which held up the release of their latest version of their program), but can't seem to find the post again. Maybe someone else on this forum also remembers....

[tonybolger]

I've been looking at some of the Dell workstations, which all seem to use ECC ram as far as I can see. I understand that ECC ram produces less errors, but on the other hand is MUCH more expensive than non-ECC ram. Would it be a problem to not use non-ECC memory to reduce the cost ?

Annoyingly, ECC doesn't add _that_ much to the cost of the memory itself, especially for large sizes (right now, 8GB ECC registered is actually cheaper than 8GB non-ECC unregistered), but many machine vendors like to mark it up because they know you really want it. Plus the motherboards / CPUs which support it are more expensive (at least for intel).

IMHO, ECC isn't optional, it's absolutely required for bioinformatics, where a single bit error is likely to go undetected and silently corrupt your results.

But you're right, you'll pay a lot more for the machine with ECC vs one without, but you'll usually get a better tier of machine across the board.

[BaCh]

Does anyone hav any real world examples of either scenario?
Eg where using non-ECC memory has caused a problem?.

Yes ... oh yes. Has cost me weeks of time and lots of nerves. I'll look for ECC on my next development machine.

Here's an account of what happened:

http://mira-assembler.sourceforge.net/docs/DefinitiveGuideToMIRA.html#sect_bonus_atinseq

B.

[gringer]

I remember reading another blog post that was more related to genomics that discussed a similar situation (which held up the release of their latest version of their program), but can't seem to find the post again. Maybe someone else on this forum also remembers....

http://mira-assembler.sourceforge.ne..._bonus_atinseq

MIRA, yes, that was it. Thanks.