Hi, we use the ceph object store: http://ceph.newdream.net/ to store virtual machine images and disks. It is similar to Amazon Dynamo or Google Bigtable and its architecture spreads objects with one ore more copies to many simple servers. This gives you cheap high availabilty and high performance, as you gain more performance with every added node. We have a 4 node setup and can write to this object store with 1GByte/s sustained! It is written in C++ but uses a own messaging implementation. I already pointed the developers to zeromq but they didnt catch it up. This project is very promising and is getting to a stable state soon.
A other java based solution we use is cassandra (http://cassandra.apache.org) , which is used at facebook for example. It uses its own java based messaging called thrift/avro. We use jzmq to store logs in cassandra. And we are very happy with this solution, as it is rock solid and "just works". Greetings Stefan On Wed, Jun 1, 2011 at 7:40 AM, Andrew Hume <[email protected]> wrote: > yes. we have an existing thing we built that distributes a > shell-variable-like > key-value pairs across a cluster. > On May 31, 2011, at 2:20 PM, Pieter Hintjens wrote: > > Hi, > > Has anyone experience with scalable data stores, DHTs, etc. that would > help us compare existing (open source) products with a new > architecture built on top of 0MQ? > > -Pieter > _______________________________________________ > zeromq-dev mailing list > [email protected] > http://lists.zeromq.org/mailman/listinfo/zeromq-dev > > > ------------------ > Andrew Hume (best -> Telework) +1 623-551-2845 > [email protected] (Work) +1 973-236-2014 > AT&T Labs - Research; member of USENIX and LOPSA > > > > > _______________________________________________ > zeromq-dev mailing list > [email protected] > http://lists.zeromq.org/mailman/listinfo/zeromq-dev > > -- Stefan Majer _______________________________________________ zeromq-dev mailing list [email protected] http://lists.zeromq.org/mailman/listinfo/zeromq-dev
