Re: [tlug] Making better use of SSDs?

[Nava Whiteford <new@example.com>]

> On Tue, May 29, 2012 at 7:27 AM, Nava Whiteford <new@example.com> wrote:
> > You'll obviously get some benefit from using SSDs for Bioinformatics, but I know of no Bioinformatics tools that currently take advantage of the constant time random access offered by SSDs.
> 
> 
> Yes -- if they were handing out SSDs out with tissue packs at the
> train station, then no harm in trying them out!  :-)  Especially for
> just scratch space.  But, they'll have to be purchased and if they
> are, then it's a matter of whether or not they're worth the price.

:)

Give it a couple of years, they might well be.

> > For example, the alignment tools I know of tend to create reference indexes in memory, and align reads sequentially. So your reference genome and reads both get read sequentially.
> 
> Yes, you're right!  Unlike table access in a database, bioinformatics
> software do tend to be sequential.  It isn't like being open-source
> would help; the algorithms operate sequentially.
> 
> The many reasons by others here against using it as swap space are of
> course valid.  Many of them I never thought of.  So, using it as a
> disk seems to be the best option and, if so, the benefit is going to
> be marginal.

yes, particularly in the case of alignment the algorithms will be doing a lot of, small, random access (in memory). Random access on SSDs is fast, but obviously nowhere near that of RAM. This means each time you hit the swap you'll need to read a whole SSD block (256k), to access only a few bytes.

> At least, until someone develops bioinformatics software with SSDs in mind...

Yes, I think about writing alignment software that more effectively uses SSDs quite often. Not sure how much advantage it would be at the moment.

Right now, I tend to kit Bioinformatics boxes out with an SSD (for people (i.e. me :) ) to try things out) and loads of RAM.