On Fri, 5 Jun 1998 thoth_at_purplefrog.com wrote:

> I never could get the server to run on my machine, but I'm a fossil,
> still using non-esperimental kernels.

Ahh, but that's just you, and not really coda as a whole. I know way back
I had problems getting a server going, too, but I haven't tried recently.
Once the 2.2 kernel comes out and I get Ethernet access, I'll attempt to
set up a full-time coda server and see how much it has improved.

> I'm not confident about the write performance of clients through slow
> links.

Peter said at his LinuxExpo talk that it trickles back according to the
available bandwidth...

> It's a grievous resource hog. 

Mostly an implementation issue - use of C++ doesn't help matters I
suppose, and it wouldn't hurt to run coda through purify once or twice...

> Linux doesn't yet have the "raw" disk devices coda needs for
> ultra-reliable operation.

What is it with these people that keep repeating "raw device, raw device" 
as if it were a magical divination, but have never heard of fsync(), which
gives you (compared to raw devices) great performance _and_ guaranteed
reliability? 

> Resolution after a partitioned server is bandwidth-expensive.  I have
> doubts that you can dynamically add a server to a volume (of course,
> Peter could bust out with a 3-line shell example and prove me wrong :)
> It doesn't have good built-in security (and that giant suck-ass ITAR
> will probably prevent it from having any that is exportable). 

Your FUD is funny :)

The builtin security is waaay better than what NFS has - it's obviously
not perfect, but

-- Elliot					http://www.redhat.com/
Chicken Little was right.