loxilb-io / loxilb

eBPF based cloud-native load-balancer for Kubernetes|Edge|Telco|IoT|XaaS.
https://www.loxilb.io
Apache License 2.0
1.36k stars 108 forks source link

cicd, scale and long-run testing effort #87

Open TrekkieCoder opened 2 years ago

TrekkieCoder commented 2 years ago

Scale and performance are the defining factors for load-balancers. Although we have some performance numbers with wrk, we need to get some scale numbers eg. how many sessions , how many CTs etc. We need to find a proper open-source tool for this testing e.g Trex

CloudyBytes10 commented 1 year ago

I tried running long running iperf sessions as per the blog

.. root@ccc65af764bf:/# iperf -c 123.123.123.1 -p 55001 -i5 -t5000 .. .. [ 3] 2565.0-2570.0 sec 6.59 GBytes 11.3 Gbits/sec [ 3] 2570.0-2575.0 sec 7.16 GBytes 12.3 Gbits/sec [ 3] 2575.0-2580.0 sec 6.79 GBytes 11.7 Gbits/sec [ 3] 2580.0-2585.0 sec 7.01 GBytes 12.0 Gbits/sec [ 3] 2585.0-2590.0 sec 5.93 GBytes 10.2 Gbits/sec write failed: Connection reset by peer [ 3] 0.0-2591.6 sec 2.98 TBytes 10.1 Gbits/sec

After restarting loxilb dockers randomly, sometimes the traffic stream gets reset and stops. It is not exactly reproducible and works well most of the times but might need further analysis.

luisgerhorst commented 9 months ago

I would be interested in benchmarking loxilb in a common real world scenario. Maybe I can build it in such a way that it can be reused for your CI.

What is the most common use case for loxilb? Is it just HTTP you think it is also used for something like redis (https://docs.redis.com/latest/rs/networking/cluster-lba-setup/)?

luisgerhorst commented 9 months ago

https://github.com/giltene/wrk2/ might be preferable to https://github.com/wg/wrk (via https://marioskogias.github.io/docs/heels.pdf)