Ask HN: I'm working on some tests for Amazon EBS volumes. What do you think?
13 comments
We're using EBS, so I'm interested.
You might consider posting your results in phases. People will probably have good suggestions about what to test next.
You might consider posting your results in phases. People will probably have good suggestions about what to test next.
That's a good idea. I could keep the Bonnie results in a git repo and create a new branch for each version of the test configuration.
EDIT: Okay, I got a repo going with updated config and a report from a sample run. I'm running some more significant tests now (up to 25GB) and you'll be able to get the results from here: http://github.com/solutious/ebstest/
EDIT: Okay, I got a repo going with updated config and a report from a sample run. I'm running some more significant tests now (up to 25GB) and you'll be able to get the results from here: http://github.com/solutious/ebstest/
Here's an active link:
http://gist.github.com/123400
http://gist.github.com/123400
I stopped using EBS after the EC2 instance just stopped responding - twice - each time after a few weeks. I was using an EBS volume for MySQL data storage. I am not satisfied that this problem has been fixed. I suggest you look at this thread before using EBS for production: http://developer.amazonwebservices.com/connect/rss/rssmessag...
That thread appears to be related to Debian (Lenny) and XFS only. Were you using XFS or ext3?
Was using XFS on Ubuntu...
I'm just curious...what sort of workload are you dealing with that makes XFS a rational choice? It's dramatically less well-tested than ext3, or even ReiserFS, on Linux, and its performance characteristics are not particularly compelling except for a few pretty specific cases.
I have a history of tinkering with odd filesystems, but I wouldn't deploy anything other than ext3 to a production Linux server at this point in time.
I have a history of tinkering with odd filesystems, but I wouldn't deploy anything other than ext3 to a production Linux server at this point in time.
hm. I've been having the same problems when I try to do large postgres operations like index on my Ubuntu instance. I'd figured my configuration was messed up somehow. I'm really glad I came across this information!
That thread talks about XFS problems, not about EBS.
Yes, but the problems seem to have to do with XFS and EBS.
Wrong. According to the thread it's a kernel issue.
Maybe read your own links before spreading FUD?
I don't care if it's the kernel or whatever else. I just know that the dang thing blew up on me twice. So you go and think about kernels and stuff - and have fun with EBS - I'm not using it anymore.
So we have the classical PEBCAK here. You got bitten by technology that you don't understand. Don't blame the machine, blame yourself.
Here's what I came up with: http://gist.github.com/123400
Right now I'm only running Bonnie64 tests and only for small and large instances of Linux. There's also no fancy reporting, just Bonnie's txt output. My plan is to add other tests and then figure out something simple for reporting (Google Charts, probably). After that I'll add Solaris and other distros / machines sizes.
Do you have any suggestions for other tests/benchmarks to use? What do you think about the approach?