Affects Version/s: 6.5.0
Fix Version/s: 7.0.0
As a follow up of our stress testing of the new proxy distribution mechanism, I ran some perf comparison tests *rate tools through the proxy vs. directly on the directory server.
On authrate and searchrate we saw a significant difference in throughput between using the proxy or not, and should investigate if this can be tuned in configuration, or is a problem in the proxy.
The setup used to get the appended results:
machine1 (AWS m4.4xlarge):
SDK as a client for the * rate tools
PROXY with static discovery mechanism, proxy_backend with 1 shard pointing to the DJ instance
machine2 (AWS m4.4xlarge):
DJ with default configuration, 100k auto-gen entries
Approx results (full graphs apended):
System metrics for the proxy server also included (authrate/searchrate tests are the first 4 spikes)