Check the recorded presentation during the architecture group meeting

Meet Google Drive - One place for all your files

Checkout the JIRA Epic

Improve throughput and performance under stress

Architecture Slack channel

Test setup?

• travelDemo.htm
• on DX-core
• on public instance (cache/anon access enabled)
• JProfiler in sampling mode
• wrk to generate the load (some threads each maintaining an open connection and bombarding magnolia w/ requests)
  • this means a quite serious stress inflicted on a mere public instance (some k's of reqs vs 200 threads in the pool)
• disclaimer: this may or may not be similar to the load test setup that NYT had!
  • we still need to find out which pages the partner is testing Magnolia against, which caching strategies they use, which modules are involved in the setup and so on
  • from the looks of it, the fix for MAGNOLIA-7981 will do the trick for NYT, but it only means they aren't stepping on other bottlenecks

wrk -c200 -t200 -d600s --timeout 20s --latency --header "User-Agent: wrk" <http://localhost:8080/magnoliaPublic/travel.htm>
Running 10m test @ <http://localhost:8080/magnoliaPublic/travel.htm>
  200 threads and 200 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency   323.71ms  748.67ms  12.39s    91.33%
    Req/Sec    17.49     18.05   340.00     87.45%
  Latency Distribution
     50%   92.05ms
     75%  240.47ms
     90%  809.91ms
     99%    4.08s
  1326415 requests in 10.00m, 30.83GB read
Requests/sec:   2210.23
Transfer/sec:     52.61MB