[Gluster-devel] Why integrate both client and server into a single executable?

Gordan Bobic gordan at bobich.net
Mon Feb 2 16:31:14 UTC 2009


On Tue, 3 Feb 2009 00:07:41 +0800, Kirby Zhou <kirbyzhou at sohu-rd.com>
wrote:
> I'm sorry, I had taken a mistake here. You are right, we can distinguish
> between client/server by the symbolic linked filename. But the program
> itself do not distinguish its role at least until 1.3.12. It means that
> somebody careless would set up a server daemon by the client process
name.
> It should be a problem.

If it's a server-only (or client-only) setup, then there's only one process
anyway, so I don't see a problem. If a machine is both a client and a
server, then having both in a single process results in a considerable
speed-up because it avoids a process switch and doesn't have to push all of
the communications through the socket layer and/or the IP stack. The
performance boost (and the CPU usage drop) is quite noticeable on my
systems.

> I do not know the actual situation of 2.0rc1, because of the
> 'NO_CONTENT_TYPE' problem I have noticed in some prior mails.
> About point 2, I will check my APT system later, why the fuse-lib depends
> on kmod-fuse.

What distribution are you talking about? I don't remember this being the
case on RedHat based distributions.

Gordan





More information about the Gluster-devel mailing list