#Why is latency as measured by nabox so different than what Active IQ shows?

1 messages · Page 1 of 1 (latest)

cold chasm
#

Looking at node latency, Active IQ shows a maximum latency in the last hour of .718ms, nabox shows the same node with a high of 30.6ms.

Why such a difference?

rose turtle
#

hi @cold chasm what version of Harvest are you using and which dashboard/panel are you comparing?

cold chasm
#

3.2. cDOT / ONTAP:Cluster node latency vs AIQ node latency.

#

I'm looking at tukvvnetapp01-12 specifically, but I'm seeing the same on other nodes.

fading thorn
#

For one thing, that latency is averaged out in AIQUM because it does 5 minute intervals.

#

I'm not sure what the counter is for the node chart in Harvest, but I think AIQUM probably does more fancy stuff too.

#

(Though I can't imagine what on a node latency chart, maybe combining all workloads together?)

#

It might be measuring latency at dblade vs nblade on different charts (maybe AIQUM is nblade, Harvest dblade?). The difference is nblade latency is from the node the LIF is on, but if you have indirect i/o then the LIF and disks will be different.

cold chasm
#

95% of the IO here is FC to VMware vSphere. Seems like ALUA would minimize indirect I/O.
"latency is averaged out in AIQUM because it does 5 minute intervals." Yes, this makes sense. A momentary blip can get lost in a 5 minute interval. Any idea what nabox is doing to build this graph? If it's doing something like graphing the biggest latency in a 5 minute period I could make sense of the difference between the two.

fading thorn
#

No Harvest is 1 minute intervals

cold chasm
#

Thanks! That's helpful.

cedar dune
#

I believe ActiveIQ is only reporting read/write latency, while harvest also takes "other" ios into consideration ?

dawn plaza
cold chasm
rose turtle
cold chasm