About the neighbour table overflow, can't be a Coda problem as it is
related to the kernel's ARP cache. CMU has a bridged Class B network
(close to 60K hosts) and I'm not seeing these overflows, so something
must be screwy with your network setup. See also,

    http://www.linuxgazette.com/issue65/tag/12.html


On Wed, Jul 03, 2002 at 04:00:49PM +0100, Jan Bruvoll wrote:
> On the sideline:
> I'm just a bit sad that the design choices made for Coda as far as I see
> it makes it unusable in a real-world environment where standard PCs are
> shipped out with 40-50Gb disk space. Would it be completely impossible
> to chuck out the current mmap-ing model and do something less resource
> intensive instead? (I'm sorry not to have any proper alternatives, I
> don't know enough about this.)

Many of the limiting design choices were made in 1986. When harddrive
sizes were in the order of 20MB and laptops didn't exist.

Not using RVM for metadata, or more likely using RVM as a journalling
metadata cache is definitely possible, but there is so much to do
already. Just keeping up with the bugs/quirks that are introduced with
new kernel versions, new libc's and new gcc's is keeping me quite busy
already.

Jan