Scalability question

[jdelisle]

I have about 30 Linux hosts with SAN connected disk, approx 20 disks per host.

I need to graph disk IO, for each disk, for each system. That's around 600 disks to monitor total.

When Cacti polls a remote linux host for disk IO, does it poll it once for all disks, or poll it once for each disk? IE: Host has 20 disks, does it get 20 SSH sessions running iostat, or does it get one SSH session running iostat?

Any issues with scalability? I'm thinking about a 30minute poll interval for this.

[gandalf]

First, graphing iostat via shh tunnel is not a default cacti template/script. As implementation may vary, I can't answer your question about the number of connections.
But you do know the diskio support of net-snmp that makes "similar" data available via SNMP (at least if that support is available on the target device)? Templates have been posted to the 4th link of my sig
R.

[jdelisle]

First, graphing iostat via shh tunnel is not a default cacti template/script. As implementation may vary, I can't answer your question about the number of connections.

Ok that makes sense. Ideally I'd do this in a way that would allow me to use a template of some type across multiple systems. Chances are that would mean polling the info one disk at a time, where the disk is specified as a parameter.

But you do know the diskio support of net-snmp that makes "similar" data available via SNMP (at least if that support is available on the target device)? Templates have been posted to the 4th link of my sig
R.

I assume you're referring to this:

http://docs.cacti.net/usertemplate:data:host_mib:diskio

It looks like just what I want. Any tips regarding getting disk IO stats to be made available via SNMP?

[jdelisle]

But you do know the diskio support of net-snmp that makes "similar" data available via SNMP (at least if that support is available on the target device)? Templates have been posted to the 4th link of my sig
R.

Got it working, it's beautiful. Thanks.