As reported around the web, it looks like several people have noticed performance issues with Amazons EC2 cloud computing service. While this was to be expected, some people are surprised nonetheless. For those that are surprised: please read “the truth about overselling“. It’s about web hosting, but much of the contents are true for cloud computing as well. Cloudkick has nice graphs showing the latency to their EC2 instances:
The truth behind EC2 and other services is that you are not “renting a server”; you are sharing capacity with other users on a host system. The only guaranteed resource is memory; CPU, memory bandwidth and disk I/O are shared with other users on the same host system. The same goes for network bandwidth. If you are on a busy host or network segment, performance might be less than expected.
Since Amazon is a for-profit, they will try to optimize the load in such a way that they get a minimal amount of complaints. Maximum performance for all customers is not their top priority and never will be; the key points used to describe the service are “Flexible, Reliable and Secure”. Guaranteed performance is not on the list. Users should also realize that if their virtual machines are running 24×7, buying dedicated machines will offer better performance and better value.
Keeping this in mind, here is a prediction for 2010: sometime during this year, we’ll start to see complaints about S3 and Elastic Block Store performance as well. These services run off the same infrastructure; the only difference between S3 and EC2 is that Amazon has a higher degree of control over where data hotspots are placed within their infrastructure. Presumably, moving some data is easier than migrating running virtual machines.
EBS performance is the most likely to start dropping first. Amazon are using some sort of networked storage for this; though they provide no details about the actual implementation I expect them to be using iSCSI. When more and more developers start using this service, their storage network will face the same scaling issues that EC2 is having. Nobody has ever built cloud infrastructure like this, so Amazon can only learn from their own experiences. Monitoring your traffic can get you a long way, but when you have absolutely no control over the services people are running on your infrastructure you will be faced with unpredictable peaks in usage now and then.
For those that are interested: here’s a recent report on S3 performance. Last year everything seemed to be fine, I’ll be monitoring the news for issues closely over the course of 2010.
HostedFTP.com – Amazon S3 Performance Report
Related posts:
Your RSS reader
Daily e-mail
Twitter
