A memory based caching layer will be added to the frontends so the hot files won't really be a problem. I just mentioned it because the memory layer might fail and result in increased use of the data source.
However, I am still not sure if it is possible to use MapReduce to get a ordered list of the last accessed objects within a bucket. And the vclock consistency check shouldn't be performed for certain requests. I don't update any items and if so it wouldn't be a problem if an outdated version would be served for a short period of time. It simply consumes too much ressources if you are dealing with larger objects. Am 4. März 2012 17:08 schrieb Paul Armstrong <p...@otoh.org>: > On 04/03/2012, at 2:40, Philip <flip...@googlemail.com> wrote: > >> I am looking for a software to host millions of files in the order of >> 2-20MB for web serving. A few (~5%) files are "hot" and accessed >> heavily but most of the files are cold. The system is going to work >> like a big cache and therefore I need to make sure that the cluster >> won't go out of space while new files are beeing added. I'd like to >> store the last time a file was accessed and delete the least accessed >> files to make sure there is always 20-30% free space in the cluster. >> This system is currently spread across alot of different standalone >> servers and serves a few gbit of bandwidth. > > Couchbase is probably a better solution for your needs (let it eject things > based on LRU). Another option is to have memcache in front of Riak so your > hot objects are fast and kicked by LRU and you have a large, replicable store > behind it. _______________________________________________ riak-users mailing list riak-users@lists.basho.com http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com
