Avi Kivity

Unfortunately the patch doesn't help sufficiently, it's still slower.

Follow-up: https://github.com/scylladb/seastar/pull/2092

On ARM, these two patches change io_uring from ~ -12% to ~ +2%. So maybe there's hope.

On what account does this run?

On what account does this run?

On what account does this run? Also, why run on two test providers?

Did you check the return type of dma_write()? If it returned the amount of data actually written, then it didn't silently drop anything.

> But for that, there needs to be some way for the two compressors to communicate with each other, in finite time. Currently, there is no such way. Compressors can...

> > Is re-negotiation really needed? > > I mean, nothing is "really needed", but if you don't have the ability to send a message from the server to the...

> In any case, we don't have to commit to this change. We can consider it experimental, and revert it later if we aren't satisfied with it. I'm not against...