The free webinar today might help answer some of your questions: http://blog.basho.com/2010/05/26/free-webinar:-load-testing-riak---june-3...@-2pm-eastern/
On Thu, Jun 3, 2010 at 9:46 AM, Germain Maurice < germain.maur...@linkfluence.net> wrote: > Hi > > We've started to use CouchDB earlier this year, and we go from > disappointments to failures, only to switch back to disappointment > again. We've been testing riak for a few weeks now, and we are pretty > happy with the results. But before definitely switching to riak, we have > some questions. > > We need to store a lot of documents. Each document is a web page, with a > lot of metadata (author, title, date, content extract from the article, > from the comments …). We currently have over 12 million documents, stored > in CouchDB, and the DB is currently a little bit over 1TB in size. We're > adding roughly 65k new documents a day (average document size: 220kB). > Soon, > we will also add 150k smaller documents a day (average size: 1kB), so > the size of the DB will keep growing fast. > > In our tests, we are using the following configurations: > - 1 node is a physical server (2GB RAM, XeonDualCore @1.60GHz, Ubuntu > 9.10 Amd64, riak 0.10-1, innostore 10_1) > - 2 nodes are virtual server (each node is 1.5GB RAM, XeonDualCore > @2GHz, Ubuntu 9.10 Amd64, riak 0.10-1, innostore 10_1) inside a Xen host > > We managed to crash the virtual servers on Xen, while having 15 workers > and doing about 125 writes/second (average size of the documents: > 190kB). This size was measured while writing data to the disk using > Innostore. > > What kind of configuration should we consider for this setup : what are > the potential bottlenecks we may encounter, and/or are there specific > tuning options in the Erlang VM that would help better fit our needs ? > > To start a ring in production to handle this volume, how many nodes > should we consider for good performance ? > > Finally, we are curious on how people are using riak: what do they > store, how many documents, which frequency, do they spent a lot of time > doing administration work on the nodes, what should be done after a node > crash and how should we recover in a production environment ? > > Any feedback on existing production environments would be welcome. > > Thanks! > > -- > Germain Maurice > Administrateur Système/Réseau > Tel : +33.(0)1.42.43.54.33 > > http://www.linkfluence.net > > > _______________________________________________ > riak-users mailing list > riak-users@lists.basho.com > http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com >
_______________________________________________ riak-users mailing list riak-users@lists.basho.com http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com
