On 1/18/24 10:33, Jens Axboe wrote:
On 1/18/24 11:31 AM, Bart Van Assche wrote:
On 1/18/24 10:04, Jens Axboe wrote:
If we attempt to insert a list of requests but someone else is already
running an insertion, then fallback to queueing it internally and let
the existing inserter finish the operation.
Because this patch adds significant complexity: what are the use cases
that benefit from this kind of optimization? Are these perhaps workloads
on systems with many CPU cores and fast storage? If the storage is fast,
why to use mq-deadline instead of "none" as I/O-scheduler?
You and others complain that mq-deadline is slow and doesn't scale,
these two patches help improve that situation. Not sure why this is even
a question?
How much does this patch improve performance?
Thanks,
Bart.
