> it would be great if there was also a performance test case that simulates 
> the traffic patterns you have in mind. 

The pathology that this improvement addresses is quite intuitive to understand, 
in absence of a benchmark or performance test. That said, I do think we have 
broad agreement among our openwhisk community and in the broader serverless 
community that we need to establish dom benchmarking methodologies and a suite 
of tests for performance. This needs a concerted effort. I think the work 
Markus started (and if I recall Tyson has investigated) could be the start of 
such an effort.

A plan to pursue further within our code base is to allow the load balancer to 
ingest (real) traffic patterns and simulate load. This requires more 
refactoring to enable us to separate the load a balance and simulator los more 
easily (or to allow others to inject their own loadbalancer). I think the 
benefits of this is that we can accelerate investigation of next algorithms.

The road to for better load balancing is long and we are still early in the 
journey. This is where I see much of the future innovation and optimization. 

-r




Reply via email to