We use now our own storage, called DiskMap. There will be soon a
couple of articles on our blog explaining how we use Voldemort
(modified hashing) and why we introduced this new storage.
But quickly, everything came from the problems running BdB on
commodity hardware (so motherboards with limited RAM) and plenty of
disk space. If the Bdb RAM formula is not matched, this is terribly
slow.
We also wanted to have a partition aware storage, with rebuilds
working on a disk matching basis (so no RAID required, and local copy
possible via a spare slot).
But this is a very specific way of using it, it would have some
drawbacks for other usages ( optimized for last access date caching,
no local range queries so eventually consistent BTrees on top of the
ring are a must, per store configuration requiring to guess the data
volume distribution between stores ).
François
On Jun 27, 2:49 am, Vinoth Chandar <
mail.vinoth.chan...@gmail.com>
wrote: