ioping - 10 vs 25?

[kpmedia]

Which is better and why?


# ioping . -c 10
# ioping . -c 25

I prefer the sustained test of 25. I think the 10 is too brief to be accurate.

But is ioping possibly causing it to be slower (less accurate tests) beyond 10?

Opinions?

Aside: I do think it's good to use both it and dd together, not one or the other. See what they both report.

 

[tchen]

From my monitoring graphs, neither are good. I've resorted to count 1 every 5 mins. The time average then gives you a better view of what the nominal range is, what happens when an io storm hits, etc.


If you must do this manually, just do the 10 at different times throughout the day.

 

[kpmedia]

From my monitoring graphs, neither are good. I've resorted to count 1 every 5 mins. The time average then gives you a better view of what the nominal range is, what happens when an io storm hits, etc.


If you must do this manually, just do the 10 at different times throughout the day.

What would be the syntax to try that automatically?

 

[tchen]

What would be the syntax to try that automatically?

I'm using zabbix so I've set it up to do the usual ioping -c 1 command with a few awks to strip the output.  Zabbix graphs it for me.  I don't have any scripts to do it otherwise.

 

[VPSCorey]

I think people need to pay attention to how their application performs instead of relying on tools such as ioping and dd tests which are never used correctly.

If your app normally writes 4-8k per write and only does 4 iops for a forum per page load why would you use a 1G dd test to measure performance which usually serves no purpose other than negatively impacting node performance?

 

[Deleted]

You should stay away from such tools, they will /never/ give you accurate results, ever. On virtualized environments, you'll get unusual results that do not match up with reality. Timers within virtualized environments have to be emulated due to containers not having physical access to hardware timers (you need ring0 to get to the PIT/APIC/ACPI), and you'll have to also contend with interrupts being emulated as well. Some CPU instructions have to be emulated as well, and some VM software do so poorly (anyone remember vmwares bad emulation of cmpxchg?)