Amazon now has bad-ass EC2 virtual servers. I am curious how powerful these new options really are though. We use Amazon’s single core $0.10/hr EC2 instances within our data backups system to do localized processing on the backup data store. They work fine for this purpose. However, we have tried in the past to run a CPU intense app on EC2, but that brought our EC2 instances to their knees compared to "equivalent" real servers.
For example, we toyed with the idea of building a spam filtering overflow system on EC2 to dynamically add capacity to our spam filtering cluster here at Rackspace as a precautionary measure if our customers’ inbound mail volume were to ever spike beyond what our real servers could handle. We actually got as far as coding the complete system for scaling-out the appropriate number of EC2 virtual spam filtering servers, rerouting sub-sets of our mail stream, and scaling-in after the traffic spike ended. However we never could get this system to perform the way we needed it to. The EC2 instances could not process near the amount of mail needed to make this cost efficient. With the CPU pegged it took 4 EC2 virtual servers to equal the spam processing power of one real single CPU Athlon 3200+ server with 1 GB RAM.
So we shelved this project and just added a ton of dual-opteron boxes in order to be able to sustain large traffic spikes.
Amazon says that their $0.10/hr EC2 instances are equivalent to an "early-2006 1.7 GHz Xeon processor", but for our app one of these EC2 instances was equivalent to 1/4 of a 2005 Athlon processor. So their 4-core instance should now just about equal a single-cpu Athlon box from our perspective. But at $0.40/hr the cost is the same as before.
Disclaimer: We did not spend a lot of time fine tuning our EC2 image, so perhaps we could have seen better performance if we did. However, we built our image by starting with their standard Red Hat base image, which is what the majority of EC2 customers run.