Benchmarking using wrk the smallest linode (1024 now 2048) serving a page from an untuned Rails application using nginx/passenger getting almost no other traffic. Hard to compare of course given the various other factors, but produced slightly lower performance after the upgrade. Serving a page from nginx directly (no Rails) had no appreciable difference in performance, I guess the Rails web serving is more vCPU bound?

Before Upgrade:

  Running 30s test @ http://...
    5 threads and 20 connections
    Thread Stats   Avg      Stdev     Max   +/- Stdev
      Latency   308.91ms  135.01ms 985.82ms   80.00%
      Req/Sec    14.15      4.61    24.00     66.36%
    2206 requests in 30.00s, 28.51MB read
  Requests/sec:     73.53
  Transfer/sec:      0.95MB

After Upgrade:

  Running 30s test @ http://..
    5 threads and 20 connections
    Thread Stats   Avg      Stdev     Max   +/- Stdev
      Latency   321.74ms  102.45ms 957.74ms   87.32%
      Req/Sec    12.02      2.18    17.00     80.75%
    1858 requests in 30.01s, 24.03MB read
  Requests/sec:     61.92
  Transfer/sec:    819.98KB