Slow LIST command

[Mika Iisakkila]

Rob Siemborski wrote:
[skiplist]
> No, it doesn't have a specific in-core cache, it relies on the filesystem
> cache alone (it mmap()s the db file and uses that directly).

Can you tell offhand if it does any of the mmap() stuff that's a no-no
on HP-UX platform (eg. same process mapping two overlapping areas, or accessing
the files directly while something is mmapped)?

I had to stay with Berkeley on HP-UX since skiplist became unusably slow
with map_nommap. Curiously, at first it seemed to run just great, but deleting
an entire user took ages. I was just thinking if it was worthwhile to try
making skiplist code use the real mmap() and have the rest of Cyrus still
using map_nommap, since I know that part not to work otherwise.

--mika