How to get the max I/O for the disk ?

[Abdullah_2]

Dear All,
HI, I'd like to get the max I/O for the disk?
Where actually i used the glance but it give the resulte by 100%, where i what is the max I/O for the disk.
Example :
the result from galnce
it is equal to 45%
I want to know how much it is equal 100% (the maximum) ?

is it clear ?

many thanks

[Michael Tully]

The top level of glance only gives an average across all disks. You need to drill down further to get a better on-line view. Use the 'F5' key or until you see 'IOByDisk'

You could use 'sar -d 5 5' to view each disk

[Tom Geudens]

Hi,
Yes, I believe the question is clear. You want to know the maximum throughput (in Kbytes or Mbytes) possible for a given disk. Right ?

However, the answer is not as easy. There are however a couple of tools available that can help you there. For throughput I'd recommend IoZone. See http://www.iozone.org/

Note however that throughput does NOT equal responsetime. If you want to measure "performance" you'll need other tools.

Hopefully this gets you started,
Regards
Tom Geudens

[Tom Geudens]

Hi,
Yes, I believe the question is clear. You want to know the maximum throughput (in Kbytes or Mbytes) possible for a given disk. Right ?

However, the answer is not as easy. There are however a couple of tools available that can help you there. For throughput I'd recommend IoZone. See http://www.iozone.org/

Note however that throughput does NOT equal responsetime. If you want to measure "performance" you'll need other tools.

Hopefully this gets you started,
Regards
Tom Geudens

[Peg McMahon]

The metric in glance is does not deal with how much work a disk can do (how many bytes it can handle in a given time frame). What it does tell you is the percentage of time during the measurement interval that the disk was being used.

So the 45% in the example above says that 45 percent of the time the disk drive was in use - not that it was doing 45% of the work it is capable of doing.

For tuning purposes, we tend to feel that a disk is starting to feel pressure at 30 to 40 percent utilization.

You can confirm this in the GUI glance plus (command gpm& in an Xwindows session). Open the IO by disk report. If you click once on a disk to select it and then double-click to open it, there are metrics in the second popup screen on the percentage of jobs in the disk queue. This is vital information to me when tuning disk performance.

Best Regards,
Peg

[Steven E. Protter]

Here is a performance collection script set that runs background and collects this data. I attached it.

Your best bet to find out theoretical capacity is this:

ioscan -fnC disk

Then go to the manufacturer's web site and look up the specifications. You will find the actual performance of the disk somewhat lower than the specification.

Note: You profile has you at 0 for 23 on handing out points in answers to your question. While points are not mandatory, the system here is if you get help, to assign them. A practical way of looking at it is you encourage more people to try harder to answer your question by handing out points.

I hope my post helped you.

Peace.

SEP

[Tom Danzig]

You could do something like:

timex dd if=/dev/zero bs=1k count=10240 of=/dev/rdsk/cXtYdZ

This will give the total elapsed time required to write a 10MB file in blocks of 1k. You can then do the math to figure out MB per second.

If /dev/zero is not available, then execute
the following to create:

For 10.20:
# mknod /dev/zero c 3 0x000003

For 11.x
# mknod /dev/zero c 3 0x000004

[Tim D Fulford]

MeasureWare

If you plot BYDSK_PHYS_IO_RATE Vs BYDSK_UTIL
follow the trend till you hit 100% util.

To get this information do an extract
extract -xp -d -r -b -e
get MS Excel out & start plotting.

That said, the easiest thing to do is to use sar & see what the servcice time of the disk is. Say "sar -d 300 10" (get 5 minutes averages & 10 samples)

If you have a service time of 5.5ms for a single disk then it takes 5.5ms to return 1 IO. Thus 1000/5.5 is the number of IO's per SECOND it will do (182 IO/s). Bingo saturation IOP level. A word of warning, you do not want to design your disks to be running near 100% utilisation, disk queues will start to form & things get sticky, 75%-80% is the maximum.

Regards

Tim

[Tim D Fulford]

Just a couple of things....I dont want to contradict anyone but I've been working V hard in this area & come across the same illusions again & again.

1 - dd tests on disks only show how fast dd can run your disk at. Any one will tell you that
dd if=/dev/rdsk/cXtYdZ bs=512 of=/dev/null
will be really slow, where as
dd if=/dev/dsk/cXtYdZ bs=4096k of=/dev/null
will be better

2 - Terminology
Bandwidth is ammount of data, kB/s or MB/s
Throughput is IO/s & has NOTHING to do with bandwidth..
However, they are related by block size so
Bandwidth [kB/s] ~ block sz [kB] * IO/s

Generally speaking doing a single IO of 512bytes will take as long as 64kB (hence the point 1)

3 - Use your data
To take manufactrers figures & do calculations on them could be very misleading, your data on your current system is the best source. Of course if you want to switch from 10,000 rpm to 15,000 rpm disks you would expect a similar performance improvement (~35% more IO), but to use the service times they quote could mean the system is under or over specified.

Regards

Tim