istio.io performance numbers

[linsun]

@danehans noticed the performance numbers seems odd between 1.21 and 1.22, for sidecars.

https://istio.io/latest/docs/ops/deployment/performance-and-scalability/#latency-for-istio-121 https://preliminary.istio.io/latest/docs/ops/deployment/performance-and-scalability/#latency-for-istio-122

We'd expect the sidecars numbers to be similar given tested on the same hardware. @Shuayb1 - iirc, when testing sidecars, you used ambient profile which has HBONE enabled, vs the default profile. Can you please confirm this? If so, could you rerun with default profile for sidecars?

[Shuayb1]

Yes, I used the ambient profile.

These are the graphs for the sidecars P90

p99

Below is also p90 graph from the ambient test with the default profile:

[linsun]

Thanks Shuayb! So the higher latency is caused by running sidecars with HBONE while with ambient profile. Could you generate the graph with the new numbers for sidecars (when using the default profile)? cc @craigbox

[craigbox]

What's the best number to use for describing "average" latency in general, for use on an overview page for example? p50? Can we have those numbers too please (maybe not in a graph; maybe I just need a pointer to the source data)

[Shuayb1]

What's the best number to use for describing "average" latency in general, for use on an overview page for example? p50? Can we have those numbers too please (maybe not in a graph; maybe I just need a pointer to the source data)

This is the data for the sidecar test using the default mode: sidecar.csv

and this is for the ambient test using the default mode: ambient.csv

[craigbox]

How do you generate the graphs? Should we put p50 on the page too? @howardjohn may have an opinion

[Shuayb1]

It is as discussed in the Slack chat. Thanks.