More Predictable Shared Dyno Performance

Summary

Heroku describes a year-long effort to make CPU access more predictable for shared dynos in the Common Runtime. Previously, individual dynos could burst CPU use relatively freely when capacity was available, improving utilization but sometimes allowing misconfigured or malicious “noisy neighbors” to cause response-time spikes or H12 timeouts. The platform introduced improved resource isolation for eco, basic, standard-1X, and standard-2X Dynos, limiting bursts while preserving the ability to burst and aiming for fairer access to shared CPU resources. The change does not alter the number of dynos per instance, and Performance, Private, and Shield type Dynos are excluded because they run on dedicated instances.

Takeaways

1. 1

   Shared dynos can still burst CPU use, but less extensively than before, prioritizing fairer and more predictable access to shared resources.

2. 2

   Noisy neighbors could cause random request-response spikes or H12 timeouts when other dynos consumed much more than their fair share.

3. 3

   Performance, Private, and Shield type Dynos are unaffected because they run on dedicated instances.