Ask HN: 10k+ node _windows_ cluster in AWS?

If you're looking to experiment -- I think the best answer is talk to AWS support as this isn't a normal case that is allowed by AWS accounts.<p>Check out <a href="http://aws.amazon.com/ec2/faqs/#How_many_instances_can_I_run_in_Amazon_EC2" rel="nofollow">http://aws.amazon.com/ec2/faqs/#How_many_instances_can_I_run...</a>.<p>To quote it: " You are limited to running 20 On-Demand or Reserved Instances, and running 100 Spot Instances per region. Cluster Compute Eight Extra Large instances are limited to running 8 On-Demand instances per region and Cluster GPU Quadruple Extra Large instances are limited to running 2 On-Demand instances per region (Currently cc2.8xlarge and cg1.8xlarge instances are only available in US East (N. Virginia)) "<p>However, these are the default limits -- AWS will up your account limits by quite a bit if you just ask. This said, max I've ever seen AWS approve was 400 on-demand instances per one region (I think this means 2800 instances total over 7 regions... never saw if the limit applied to all regions). Also, likelihood of having a 10000 spot instance request fulfilled is probably near zero if I were to guess even if they let you make a request that big without first clearing it with them.<p>I take it you're also familiar with the cost of running that? The absolute minimum you could possibly be charged using small instances is $1,200 -- partial hours are rounded up.

I was at Super Computing 11 this past November.<p>I haven't done it personally. But I saw a demo at the CycleComputing booth that looked pretty slick.<p>They've brought up to 30,000 linux nodes. Not sure if they have an offering for windows or azure.<p>There are entire companies being built to solve this problem.

An AWS (support) forum may be a better resource. I don't see why this wouldn't work - maybe the Amazon guys would appreciate a heads-up, though.