Failures of public services generate a lot of publicity. Every EC2 blip generates too many blog posts, including this one, about how the cloud cannot be trusted to run business critical functions. It is a difficult debate to have because availability numbers for cloud services are easy to get (there are companies devoted to reporting this) but public numbers for internal deployments are rare. What is the typical uptime for a deployment of server software from the large providers of server software?
We could start by reporting uptime numbers for MySQL deployments. But there is a big problem here. Who is willing to report a lousy number?
Friday, October 24, 2008
Subscribe to:
Post Comments (Atom)
Posts
Like everything in business, you have to evaluate the risk and balance the cost of potential failure modes. How is EC2 going down any different than a load balancer, network switch, NAS device or other piece of hardware? Surely you have every conceivable contingency covered with redundancy, including georedundancy, right? The Amazon and Google folks certainly have, but even so, gremlins will happen.
ReplyDeleteSaying EC2 or other enterprise-grade cloud services cannot be trusted is akin to saying hardware cannot be trusted. Both are true statements.
We could build something into MySQL to allow easy reporting of this info, however in the end it's probably regarded as a business-strategic choice to divulge that kind of info in any way (even as unidentifiable aggregate).
ReplyDeleteThat might work as long as you don't expect it to log when it is down -- :)
ReplyDeleteIt's easy to trumpet performance claims when there is no measurement (public, at least) of performance. I beleive Stratus continues to be the only company willing to put it on the line (server and OS availability performance) every day, in the open. Would we do it if the numbers were bad? I don't know; we've never had to :-)
ReplyDelete