-
Notifications
You must be signed in to change notification settings - Fork 332
Performance
I'll start by stating the performance goals of Sneakers.
- Be as fast as Sidekiq - my go-to background processing framework, and if possible, faster.
- Keep the previous goal, while providing reliability, high-availability, and advanced messaging semantics - eventually these found their home with RabbitMQ (see Why i built it).
When ever I do not need (2) - I would probably use Sidekiq.
Further, if you do not need (2), for long-running background jobs, there's really no difference between Sneakers and things like Sidekiq, and in my opinion you should keep using Sidekiq as it has an immensely larger ecosystem.
While I can't really enumerate all of the possible use cases out there, mixing up I/O and CPU, I can specify the simplest, most bare bones microbenchmark there is in a spirit similar to the Computer Language Benchmark. To me, this kind of benchmark is simply a gut feeling of the plumbing of the framework. The next step is to set up a full fledged POC and use case benchmark.
Just as you can't conclude what language to use by those benchmarks, you can't conclude which background framework to use by my microbenchmark - you should test your own use case as it may behave differently.
For an empty job, Sneakers is mostly only limited by the broker speed (RabbitMQ). On a recent 2012 MBP it reaches 7000req/s for a silly microbenchmark.
A detailed benchmark of standard EC2 instances and comparison will follow here soon.
From my experience, in a workload that's typical of fast event processing, I get up to 3000req/s on Sneakers from an EC2-Large while Sidekiq sits tightly at 600 (combined - 2 separate Sidekiq processes in order to use both CPUs).
Note: Generic benchmarks are a misleading, sneaky thing. This page was revised to be more explicit, due to Mike Perham's insightful advice. See here for the original discussion - https://github.com/jondot/sneakers/issues/9