shinrich edited a comment on pull request #6978: URL: https://github.com/apache/trafficserver/pull/6978#issuecomment-663228884
I did some side-by-side measurements in with three machines in one of our colo's between global pool, hybrid pool, and thread pool. I ran a python script over the 15 minute squid.log files to calculate the latencies for transactions that go to origin. The script also computed what percent of those transactions reused an existing origin connection. I combined that with the connection count information from ysar (which pulls traffic_server metrics once a minute). Reuse Model|Median Latency|90% Latency|99% Latency|% Trans reusing origin connection|concurrent open origin connections Global|26ms|293|964|71.48|1972 Hybrid|21ms|280|928|78.91|3668 Thread|21ms|284|949|78.36|7998 The global pool has the smallest number of origin connections, but the latency numbers are better for hybrid and thread. Those two models behave similarly, but the hybrid has far fewer open connections. ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected]
