When I was on UML, I could look at /proc/io_status, and I sort of understood the whole token limiter thing.

Now I'm on Xen, and I enabled e-mail alerts for my node. IO rate alert was triggered for me , yesterday:

Your Linode has exceeded the notification threshold for disk io rate by averaging 620.49 for the last 2 hours.

How is disk io rate calculated?
How does the token limiter work for Xen guests?
Where can I see what my limit is? (I have a Linode 720)

i.e. How do I determine if I care about this alert that was triggered?

<bump>

No one?

I think there are other posts in the forums explaining this.

I think that UML was io requests and Xen is blocks of io. I think in another post caker recommended quadrupling the alert threshold. I could be wrong though. I'd look around the forum.

Xen uses the ionice feature of the CFQ disk scheduler instead of the token-limiter patch that caker wrote for UML. I don't know how it is configured.

Each count for calculating disk i/o is an i/o operation issued by your Linode. The default level for triggering an alert is 100 ops/sec (average over two hours) - compared with the 512 tokens per second refill rate under UML. Most Linodes operate the the 0 - 10 i/o ops per second range (averaged).

Your alert is not the end of the world but you probably want to see what your Linode was doing at that time - in case unusual activity is a warning of trouble to come. Linking big libraries is the only thing that really gets my i/o rate up.

_________________
/ Peter

The measured units of IO are different between UML and Xen. It's IO operations on UML, and blocks of IO under Xen.

An IO request could contain more than one block, so you probably want to increase your IO alert threshold under Xen.

-Chris

Thanks for replies! Threshold now increased.