[ENBD] 2.4.32 conclusions

[Peter T. Breuer]

"Also sprach Anders Blomdell:"
> > Arne at least says that [raw device] works. It's cumbersome, though.
> Not for production right now I think...

Well, it's the approved way in 2.4 of doing bufferless i/o. I don't see the problem.


> > What does top say is the process doing the work, and what does strace
> > say it is doing? :-).
> Top says it's working, strace says it's doing nothing (almost), but I get 
> about

What's the "it"? I would guess that it is enbd-server.

If it is doing nothing via strace, then it is an internal code loop and that is hard
to credit, especially if you see strace output from time to time (a select() call
every 5s?). There's nothing anyone can do - one has to run select on a socket to
see if anything comes down it. Maybe error conditions are not being picked up?

> 1/3 of writing speed when cross-writing as opposed to a pure 1master-2slave
> configuration. I'll leave it until I can experiment with the old servers..
> .

It sounds like a consequence of the server maxing cpu, which is the strange
item that needs to be investigated.

Peter