vmamidi commented on pull request #6949: URL: https://github.com/apache/trafficserver/pull/6949#issuecomment-649550757
> Under a pure event-driven design, there would be no reason to have more threads than cores. But we don't really have that. In olugins, and I think in a significant number of places in the core, we do blocking syscalls. If we had a single continuation queue feeding all the threads (my suggestion that was already shot down), having more threads than cores would be an effective work-around for continuations whose functions blocked. But we have per-thread continuation queues. However, having more threads means that, on average, there are fewer continuations queued ahead of yours, and thus fewer continuations whose inappropriate blocking syscalls can block your continuation. So having more threads than cores may reduce latency and increase throughput. can you give me an example of inappropriate blocking syscalls? ---------------------------------------------------------------- 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]
