PATCH: sum over all disk statistics for linux

[lamont]

Attached is a patch to make an 'all' parameter be the default to vfs.dev.read[] and vfs.dev.write[] on linux which sums over all hd/sd/cciss devices on the system.

These will give aggregate disk i/o in blocks on the system:

vfs.dev.write[,sectors]
vfs.dev.read[,sectors]

Rationale: i've got 1500 webserver-class machines which have varying kinds of hardware and partitioning layouts, but typically have one-big-/, but the underlying hardware can vary within a cluster of machines so that some machines may have sd, some hd, some cciss and i really don't care about those differences. All I want to know is how much disk i/o the system is doing.

[lamont]

well, that approach works, but i just found a better way utilizing /proc/vmstat on 2.6.x and /proc/stat on 2.4.x

zabbix-developers: are you interested in patches for this or do you already have something under active development?

[Alexei]

Yes, we are interested!

[lamont]

okay, i'll work on getting pgpgin and pgpgout just like vmstat. that's a better approach than the patch in post #1.

somehow i'd like to be able to graph the %utilization of the disk which is getting hammered the hardest on the machine. right now zabbix doesn't even support graphing %utilization, so that would be a start.

finding the utilization of the disk that is getting hammered the hardest may be better handled through the zabbix client supporting a better algebra...

i just noticed that if i drop in items for sda1 and sdb1 that the zabbix server will just disable the items for the partitions that don't exist, so this may be an acceptable solution in the short term. i haven't tried generating graphs yet...