ACI latency measurements - looking for information about

[ckaplan]

I am trying to understand where and how to best leverage ACIs' new integral packet counters and latency measurements when developing an overall performance monitoring solution.

• Can anyone point me to a document with more detail on how these metrics (latency in particular) are calculated?
  • Is it measuring port to port as just a dumb physical link
  • Is it measuring a tradition 3 way handshake
  • Is it per connection vs. per packet
• Can I get this data per connection or only averaged over the sample window?

Visibility, latency and atomic counter are mentioned in every doc and PPT I have found, but never with any detail.  I suspect a doc exists will all kinds of info on this, but I can't manage to find it.

Thanks for the pointer

CK

[dpita]

Hello

Just wanted to let you know I have started looking into your question. I'll ask around internally and try and dig up documentation or write something customer facing if there isn't already. It's a very good question but sadly I don't know the answer. 

I am researching and will get back to you. Thanks for posting. Have a nice day!

[ckaplan]

I was curious if you had uncovered any information about where those ACI performance metrics are measured and how the data is exposed.

Thanks

CK

[dpita]

Hello,

To answer the question regarding latency. This is what i found out:

Latency measurement is currently not supported for telemetry or health score impact. It is implemented on the fabric solely for load-balanding. 

In the future, from what i understand, each packet will have a latency measure on ingress and egress. 

Hope this helps! Sorry for the late reply!

[Corne Cornelius]

Hi

It's now 2 years later - has the future arrived ? Is it possible to measure latency on ingress/egress yet ?

thanks

cornec

[randreetta]

2020 the future is arrived !!

Now latency can be seen, a matrix one, from leaf to leaf.

Unfortunately, the average value equals the maximum value, with no variance and jitter, with a precision that reaches 1/10 of nanosecond (something like 100.4567us). With ptp the clock precision reaches nanoseconds, so that's not maybe so much strange.

But i see around 106us of latency between devices that are 80km far away from each other. Considering the speed of light of 210.000km/s in the fiber, this gives me 380us of delay. So it looks like, in some way, they have found a way to push the light beyond its limits (Einstein was wrong) !