doing large NFS jobs on Amazon EC2?

[ixfd64]

Would it be wise to do large NFS jobs on Amazon EC2, or any other cloud computing platform? Using cloud services is more costly than buying one's own hardware in the long run, but it's relatively cost-effective if one needs a lot of computing power within a short timeframe.

For example:

Factoring RSA-768 required about 1,700 CPU-years in total. This effort is equal to renting 100 "Eight Extra Large" instances (with 16 cores each) on Amazon EC2 for one year. Each "Eight Extra Large" instance costs $1,762 to $5,000 a year, depending on the utilization. Even at the upper end, I'd imagine that building a 1,600-core HPC cluster would cost a lot more than $500,000, not to mention the electric bills.

Speaking of which, does anyone know how much did EPFL spend on factoring RSA-768, in terms of electricity?

[bai]

this is a relevant report in case you haven't seen it.

[debrouxl]

And http://www.mersenneforum.org/showthread.php?t=16114 is a topic for some ready-made tooling usable in the cloud.
For RSA-768+-class-jobs, the lasieve4 sievers would have to be replaced by the lasieve5 sievers: lasieve4 ran out of steam on NFS@Home's sieving of M1061, so they had to finish on lasieve5, and yet, the M1061 SNFS sieving is easier than the RSA-768 GNFS sieving.

[fivemack]

Originally Posted by ixfd64

Would it be wise to do large NFS jobs on Amazon EC2, or any other cloud computing platform? Using cloud services is more costly than buying one's own hardware in the long run, but it's relatively cost-effective if one needs a lot of computing power within a short timeframe.

It makes sense to do it on a cloud platform if you're only going to do it once; on the other hand, you are very very rarely only going to do it once, and if you're doing it once and in an academic context then you 'just' have to convince a compute-power-awarding agency that you are adequately awesome, and can get to use computers where somebody else is paying for the computers and the electricity.