VNX Block: Abnormal I/O spikes recorded

Alex_Rogov

Hello,

We have several EMC VNX arrays monitored by stor2rrd and sometimes it records huge spikes of I/O and throughput (hundreds of thousands I/Os, several gigabytes per second). I've seen it during firmware upgrade and some background activities such as volume migration. Could be a bug in the array code.

It's not a big problem for daily (24h) graphs, but it messes up all scaling for long range graphs. Normal values and long range trends become invisible. It doesn't happen very often, but one abnormal value is already enough to mess up a graph and we have to install firmware upgrades on every array once in a while.

Is there a way to remove such abnormal data values from a stor2rrd graph?

Pavel

Hello,

let me talk to the developer regarding of filtering peaks during insertion once he is back from his hollidays. (end of next week)

Removal of such data is possible, however it is not too automated process.

Anyway this is something what we do only for our customers.

Alex_Rogov

Ok, I get the hint ))

Karel

Hello Alex,

try this fix:

http://www.stor2rrd.com/download-static/LoadDataModule.pm.gz

Gunzip it and copy to /home/stor2rrd/stor2rrd/bin (755, stor2rrd owner)

-rwxrwxr-x 1 stor2rrd stor2rrd 1567388 Jul  9 12:52 bin/LoadDataModule.pm

If your web browser gunzips it automatically then just rename it: mv LoadDataModule.pm.gz LoadDataModule.pm

Assure that file size is same as above listed one

The new high peaks in graphs should not appear again.

Please let us know if that helped, thanks.

Alex_Rogov

I've missed the update somehow... Thank you, I'll try it out.

Alex_Rogov

GZIP contains a different file:

-rw-r--r--   1 root root 1511611 Oct 18 11:56 LoadDataModule.pm

Is it ok?

Pavel

the best is using the latest version, it is included in 2.40.

NoOne

So.. is there any way to remove old spikes? They really annoy and ruin last year graphs.

Pavel

it is possible but not easy, check internet, you can find a few ways to to proceed it