Guys,

I'm fairly new to the world of linux so I need some help. I'm trying to determine the IOP's of a red hat server and I've found a tool that might help me its called collectl.

http://collectl.sourceforge.net/index.html

I have installed the tool and am testing it at the moment. It seems pretty good and does give a lot of information. I'm trying to figure out how to use it to get the IOP's has anyone used this tool before and if so does anyone know how to use it to get IOP's from it?

Have you read these?

And forgive my ignorance, but what is an IOP? I'm guessing Input/Output + Port, Processor or Process... And I've managed to find references for all three initialisations

I'm surprised Mark (the author of collectl) hasn't responded - maybe he has a real life and goes on holidays.

Looking at your other (duplicate) posts, you'll need to define (precisely) what you are looking for. And maybe why.
IOP is usually (in my world) used to refer to an I/O processor - as in real (FSVO "real") hardware. I see Redhat documentation that refers to IOP when describing I/Os per second. Another example of a vendor misappropriating terminology for its own uses.

I/Os ain't just I/Os. What the applications see as I/O (read and write) may have no relationship to the physical I/O that hits the disk. They may be combined and delayed for efficiency. This (of course) also affects the perceived rate per second. Then there are things like Oracle where the I/O may not go to disk at all, or if it does it'll be direct I/O and not involve the normal schedulers.
So it comes back to what do you really want to know.

Ok this is the situation.

The current policy in my organization is that prior to any virtualization of existing physical servers a detailed analysis is done. In the case of Microsoft servers the Microsoft Assessment and Planning Toolkit was used. However we have been asked to assess some existing red hat servers. Obviously the MS toolkitt isnt applicable in this instance. Also the current version of VMware guided consolodation no longer supports non-MS servers.

The current policy specifies that the IO boundries/threshold are as follows

exceed 100 sustained (more than 4 hours)IOPS
exceed 200 burst (less than 4 hours) IOPS

I just need to know is there a way I can use collectl to calcualte this figure? I dont particularly want to get into a discussion about the defintion of IOPS (not that I dont appreciate your input so far).

If you are unwilling, or unable, to define what you are seeking, I fail to see how anyone can help you. Certainly not I.
Hopefully others may have a better grasp on things.

My apologies I didnt mean to be vauge what I mean by IOPs is IO operations per second.

Again, sorry if I'm being thick, but I don't quite understand what you mean by this... IOPS is a rate (from your last post) - the way I read your conditions is "More than 100 I/O operations per second for every second of at least 4 hours" and "More than 200 I/O operations per second every second for a period less than 4 hours", which kinda makes sense for the first one but not so much for the second. Could you clarify a bit better? And is this per process or for the entire server? There's a "Process I/O Statistics" section on this page of collectl's docs, if that helps

hey syg00 - I would have jumped in sooner but just saw this.
personally I don't like the term IOPS because performance is a combination of IOs an blocksize. A system doing large block writes is actually doing more work than a system doing a lot of small I/Os.

In any event, collectl tracks the number of reads/writes, so if you add them together don't you get IOPS? purely a guess on my part.

-mark

Actually looks like I only need the writes per second. I know you can output the data to a raw file and read it back but I'm wondering is there anyway to output the data to a CSV file?

Thanks.

It should be easy to do in awk, the code will depend on the layout of the output you get from collectl.

However, are you really going to be writing the number of write operations per second... to a file? ...in a write operation? In my mind's eye I just see it going berserk as it writes more and more frequently to the file, recording its own write operations, which get faster and faster and faster... Luckily, I can't see how this could ever happen in practice ;-P

I'm getting lazy again, keeping up with posts. By default generates plot format in space-separated format for easy plotting or even loading into excel. But you really want commas, all you need is --sep. So the command to see disk writes/sec would be:

collectl -sd -P --sep ,

#Date,Time,[DSK]ReadTot,[DSK]WriteTot,[DSK]OpsTot,[DSK]ReadKBTot,[DSK]WriteKBTot,[DSK]KbTot,[DSK]ReadMrgTot,[DSK]WriteMrgTot,[DSK]MrgTot
20110226,07:42:31,0,0,0,0,0,0,0,0,0
20110226,07:42:32,0,12,12,0,452,452,0,101,101
20110226,07:42:33,0,0,0,0,0,0,0,0,0

will this work for you?
-mark

2 members found this post helpful.