Active IQ Unified Manager Discussions

Volume latency

SanAdmin
4,317 Views

When running the  "qos statistics volume latency show" command, what numbers for performancce would be considered bad? I have latency that exceeds 500us, network that exceeds 500us, data exceeds 500us and disk that is 500us. Not sure where I should start to troubleshoot .

1 ACCEPTED SOLUTION

paul_stejskal
4,017 Views

Bad is what is defined by application thresholds. If the application performs horribly at 10 ms, then 10 ms is the bar. If 100 ms is when the users complain, that is when you need to worry.

View solution in original post

3 REPLIES 3

SpindleNinja
4,281 Views

There's a lot more that just looking at the latancy.   

What's the config (hardware, network and protocal) ? 

What's the app(s)? 

How much IO are you driving? 

 

500us is .5ms (which... is pretty good)    

paul_stejskal
4,018 Views

Bad is what is defined by application thresholds. If the application performs horribly at 10 ms, then 10 ms is the bar. If 100 ms is when the users complain, that is when you need to worry.

EWILTS_SAS
3,994 Views

"bad" latency is very subjective - what's good for one application is horrible for another.

 

You also need to factor in what you actually bought.  If you're expecting 500us latency on a SATA drive, you're going to be disappointed, all day, every day, and there's nothing you can do to achieve that latency target.

 

Survey your users and ask them to tell you when latency is acceptable and use that as an initial baseline.  If they haven't be complaining for the past week at 5ms latency, don't set your target at 0.5ms.  If you're getting 1ms latency today and they're screaming bloody murder, then you've got some serious investigation to do and perhaps user education on what is realisticallyo possible, assuming you don't have a blank check to write to your vendors.

Public