Hi Dmitri, Thanks for your recommendations
On Thu, Oct 31, 2013 at 5:15 PM, Dmitri Zagidulin <dzagidu...@basho.com>wrote: > Vincent, > > How large are the objects that you're requesting? (in the 1000 objs > example). > 1KB > Also, what does your cluster configuration look like? How many nodes? Are > you load-balancing the GETs to your riak nodes (via something like > HAProxy), or are you making requests to a single riak node? > 5 nodes with haproxy in front for both http and protocol buffer > > It sounds like the first thing to investigate, in your case, is - where is > the slowdown happening, on the client side or on the server side? > To get a reference data point on the cluster-side multiget performance, > you can use an external tool like > https://github.com/basho/riak-data-migrator or > https://github.com/basho/basho_bench to get a rough idea of how many > requests/sec your Riak cluster can handle. > For example, you can download the Data Migrator (which has pretty good > multithreaded connection-pooled GETs that use the Java Riak Client) and do > an export of your cluster (or the bucket in question), and look at the > resulting objs/sec number. > 1745.6 Objs/Sec > > If the bottleneck turns out to be on the Ruby client side, you should > investigate JRuby for better multithreaded performance, or use a > concurrency library like Celluloid. > I think it's the client side, isn't it? This is why the server solution seems the most optimized (no need to instantiate any threads). A concurrency library like Celluloid is it better than the current ruby client implementation? https://github.com/basho/riak-ruby-client/blob/master/lib/riak/multiget.rb Thanks again > > > Dmitri > > > On Tue, Oct 29, 2013 at 6:24 AM, Vincent Chavelle < > vincent.chave...@gmail.com> wrote: > >> Hi, >> >> I fell in love with riak and riak-cs, I have migrated all my stack on it >> (originally from mongodb). >> But I have one big issue. I have a lot of key to request simultaneously >> and thanks for multi_get implementation (ruby) it's already optimised for >> the client side (concurrent requests). But I would like to know if any >> server implementation to come because, in my case, it is very very slow to >> request 1000 objects (unlike mongodb). >> >> You will make me the happiest man in the world. And I could take off my >> hideous memory caching solution :-) >> >> Thanks >> >> Vincent Chavelle >> >> _______________________________________________ >> 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
