[Bioclusters] Question about grid

Tim Cutts bioclusters@bioinformatics.org
Thu, 13 May 2004 14:41:17 +0100

On 13 May 2004, at 2:22 pm, Arnon Klein wrote:

> Doing this in Java is actualy pretty easy, since RMI lets you 
> transport an object containing both code and data over the network.

But you'll be bitten by the usual problem which is that the compute/IO 
ratio for most bioinformatics applications is very low, as has been 
discussed on this list before ad nauseam.  You'll spend so much time 
getting the data to the compute node that you might as well have run 
the compute locally.  And your network admin might have things to say 
about the whole idea, too.  :-)

The bandwidth of gigabit ethernet is *theoretically* about an order of 
magnitude less than the machine's own memory.  It's actually much worse 
than that in practice.  Especially if you've got a network of these 
little buggers all competing for the bandwidth trying to get each other 
to run stuff.

Leaving aside the political nightmare that grid ideas create.

The only reason Seti@home succeeded as a grid computing project is that 
the compute/IO ratio is *huge*.  Genomics applications just can't do 
this, although conceivably some things such as molecular dynamics 
simulations might.

<troll>And it's never going to perform decently in java anyway</troll>