[ENBD] Anyone got swapping over NBD-2.4.25 working?

[Peter T. Breuer]

"A month of sundays ago Alan Messer wrote:"
> Peter T. Breuer wrote:
> > I suspect that udp would not cause any memory deadlocks.
> 
> Certainly UDP would improve the performance. Does ENBD provide its 
> own recovery mechanisms? Since I'd hate to lose a page.. :)

No - that's why enbd goes over tcp. But if you really trust your medium
I see no reason why you shouldn't use udp. A duplex fast ethernet
connection over point to point crossed cable should be OK!

The implementor of udp would have to implement such recovery mechanisms
as seem appropriate. I'd vote for none. After all, you HAVE to trust
your memory.

> Indeed, that what I thought. Although may be it buys to a little 
> more time until it dies.

You can even get stuck trying to kill a stuck process in weird ways, or
trying to read its present directory by running across /proc.  Yes, you
can win some time, but it's not worth the effort, I think.

> I added the 'mlockall' to the nbd-client as you suggested and in 
> my initial tests it worked like a charm! Ran it under very heavy 
> load for about 2-3 hours and apart from running very slooowwwwily, 
> it didn't crash at all. However, you need to be careful to remove 
> the NBD swap partition before you do a reboot, since it'll kill 
> the nbd-client and leave your swapper with no where to swap. (Not 
> good!).

Interesting. I imagined it would lock up way too much memory to be
useful. If that is the case, try and work out how to limit it a bit
(to what?).

> I'll let you know if I have any other problems, but I suggest 
> others give it a try too. If it works fine, I'd definitely suggest 
> that you should be tempted to put it in the main code with a 
> '--swap' option.

Yep. I'm not very keen on mlockall because it's a libc6 -only
function, and I like to stay compatible with libc5.

Peter