merlimat commented on pull request #7499:
URL: https://github.com/apache/pulsar/pull/7499#issuecomment-930634997
I don't think that applies well to multi tenant env. In a multi-tenant
broker, the load is usually well spread across all IO threads. This could more
of an issue on a single-tenant, single-use-case cluster with very large
throughput and tiny number of partitions.
Also, I'm not yet convinced that it would be pose any reduction in
throughput in the above scenario in particular. eg: 30MB per thread will give a
decent amount of traffic. (and to my previous point, a broker with 30cores
should probably be assigned a lot more than 1 GB of direct memory, more likely
at least 32GB, which would leave this at 1 GB buffer per core).
Even then, I still believe we should go with thread-local approach, by
dynamically adjusting per-thread quotas.
--
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.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
