Want to evaluate impact of header size on L7 load-balancing throughput


I’m thinking to approach this evaluation by

  1. configuring ingress gateway to load-balance based on header content (since I read somewhere that by default istio does not buffer requests - but this is exactly what I need it to do)
  2. creating several test suites with varying size of header “payload”
  3. tuning istio ingress gateway request size buffer down to minimum for each test run
  4. measuring both throughput achieved and system resource utilization of the node

I’m sure there are few things that are wrong with this approach. I would appreciate community’s help to straighten it all out.

A discussion of and call for these type of test scenarios was a topic of discussion in today’s Service Mesh Performance meeting (recording just posted - Service Mesh Performance Community Meeting (December 2nd, 2021) - YouTube). The scenarios that you have listed here would be great input into the ongoing efforts in the CNCF Service Mesh Performance project - https://smp-spec.io. I encourage you to join in.