Re: [Nbd] Question about the expected behaviour of nbd-server for async ops

[Alex B. <al...@al...>]

Goswin,

--On 30 May 2011 17:53:12 +0200 Goswin von Brederlow <gos...@we...> 
wrote:

> But that means reading in requests from the socket.

Yup

> And those might just
> as well be more writes to disk 1 as reads from disk 2. And reading the
> request needs the memory to store writes that you don't have.

That's true. At some point you need to throttle all requests. But you can
respond to various degrees of memory pressure in different ways (e.g. drop
read cache, write out write cache, stop acking things before they are
written (a.k.a. don't expand the write cache but handle the requests
synchronously), stop processing requests entirely).

I am not saying you that there aren't times when you want to ACK
immediately, just that it isn't always the best strategy.

> Not sending ACKs
> in no way garanties the client won't send more requests for a specific
> disk.

Of course it doesn't guarantee it. However, I think you will find it
is a pretty effective feedback mechanism to slow things down.

-- 
Alex Bligh