I attended an interesting panel discussion the other night about cloud computing. Six startup CTO’s were there to talk about their experiences with Amazon’s services:

• Prakash Khot from Dimdim
• Aaron White from Doink
• Mark Pascarella from Gotuit
• Joe Berkovitz from Noteflight
• Philip Jacob from Stylefeeder
• Brian Whitman from The Echo Nest

I really liked this panel because they were all technical and heavily involved in real day-to-day dealings with cloud computing. It’s easy to find breathless hype about how EC2 is going to revolutionize startups. This was a much more balanced view of what it’s really like.

I took some sketchy notes. Perhaps because the speakers knew everyone had heard the positives, or perhaps because I already knew the positives, these notes skew toward the negatives.

• All used Amazon services, and most if not all of them seemed to use RightScale to manage them.
• Cloud is great for quickly deploying new servers.
  • Load testing: set up two machines, let them pound each other.
  • Handling unexpected demand.
• It’s scalable, but not like normal computing:
  • Higher latency.
  • Higher error rates.
  • You have to pay much more attention to your error handling code.
• Cloud limitations:
  • Can’t send lots of emails, since you need a spam-whitelisted server.
  • Disk I/O in the cloud is a lot slower than in real machines (“punishingly slow”).
  • Want a db server with 32Gb RAM? Amazon doesn’t offer it.
  • Want dual hardware load balancers? Amazon doesn’t have it.
  • PCI (credit card) compliance is a problem: use a 3rd-party cart or PayPal instead of doing it yourself in the cloud.
• Amazon is clearly the leader.
  • Microsoft has a service (Azure) in beta, but it’s very Microsoft-specific.
  • Intuit is creating a service.
  • No one mentioned Google App Engine.
• The lock-in is latency:
  • Transferring data within the Amazon services is free.
  • Transferring data to an Amazon competitor: not free.
• Cost: cloud is more expensive than real machines.
  • Cloud is good for elastic computing, not for high constant demand.
  • Cloud is good for getting started without capital outlay. Many of the six startups represented got started before being funded.
• Cloud further enables virtual companies. A bunch of guys with laptops can move office wherever they want, don’t worry about where the servers are.
• They all muttered knowingly about the day Amazon’s cloud was down.
  • “You could read about it on Twitter, but all the avatars were missing, because Twitter kept them on S3.”
• Cloud encourages queue architectures.
  • Pipelined role-based “machines”.
  • The work of the app is broken into smaller pieces, distributed onto a broader network of virtual boxes.
  • Number of machines in each role can be altered dynamically to deal with the actual load right now.
• You need monitoring services external to your cloud!

One interesting side-node: the panel was at the Vilna Shul, a historic temple on cramped Beacon Hill. They could invest in some cushions for their 1840 high-backed pews, but other than that, it was a nice change of pace to be hearing tech discussion not among monitors and keyboards but torah holders and time-worn paint.