Coda File System

Re: Client Cache Files Limit

From: Jan Harkes <jaharkes_at_cs.cmu.edu>
Date: Fri, 18 May 2001 09:48:18 -0400
On Fri, May 18, 2001 at 02:07:29PM +0100, Beckmann, John wrote:
> Hi Jan,
> Thanks for your response. This fixed the problem I was having, but now I
> have come across the same problem that Greg Troxel was having with too many
> files in a single directory.
> 
> I do not understand the reasoning that was given to Greg as to why you can
> handle 4000 files with long file name and ~= 8000 files with sort filenames.
> I could understand this, if the system came back and complained that it had
> reached it's limit, but to just crash and disable the whole system because
> of this, would idicate that the whole system does very little checking as to
> what resources it can and can't use. 
> 
> After reading a lot of documentation on the coda website, I was under the
> impression that this software could be used for FTP mirror sites! Nice idea,
> but in reality...

What do you consider reality?

Coda's development started in 1986, around the time that UNIX gave you
32 filedescriptor per process and laptops and terabyte filestorage were
hard to come by (or even imagine).

I've been working for the past 3 years trying to get Coda reliable and
usable, and I definitely agree that many of the arbitrary limits that we
are hitting are far too low for current day standards.

I would love to have the possibility to cache more than 200/300 MB in my
coda client without bringing a P233 to it's knees. But I also consider
that a few years ago a large client cache was about 20MB, and many of
the used algorithms weren't designed for more than 10000 objects in the
client cache and scale poorly. It would be great to be able to have more
than 4000 files in a directory, but only relatively recently have we
gotten to the point that it is even possible to reliably write that many
files into /coda.

There is a lot of work left to do and perhaps we'll not even be able to
catch up with Moore's law. But don't believe Coda is an already done and
finished product, it is a result of over a decade of very innovative
research and has proven the viability of many ideas that were considered
undoable.

Coda's ideas have been ahead of their time for many years. Even now
there are no alternatives that hide network connectivity problems using
techniques such as adaptation to available bandwidth, disconnected
operation, and server replication.

Sorry, but I'm not giving up just yet.

Jan
Received on 2001-05-18 09:49:40