Re: [Nbd] fua, trim, etc

[Wouter Verhelst <w@...112...>]

On Tue, Sep 13, 2011 at 12:25:21PM +0100, Alex Bligh wrote:
> There are actually subtly different questions here:
> 
> 1. Should the server advertise flush and fua support by default
>    (the answer here is pretty obviously yes in my book - though
>    currently it doesn't).

My policy on that is that I don't enable new features by default until
they've had some real-world exposure for a while. But yeah, eventually
it might make sense to change these defaults.

[...]
> Note, when considering defaults, that FUA is not as efficient as
> it might be, because in essence it does a flush().

Well, almost. I've been working on a multithreaded implementation (which
still requires some debugging) where FUA and FLUSH would be implemented
differently (FUA would cause fsync() to be called after the write with
which it is sent, and would not care about other writes, whereas FLUSH
(obvioulsy) would.

> Better would be to have a completely separate file opened with O_SYNC
> (see archives).  I think we should at least measure the performance
> implications before we change the defaults.

Yes, absolutely, as above.

-- 
The volume of a pizza of thickness a and radius z can be described by
the following formula:

pi zz a