Echoprint Solr odd behaviour

[Daniel Cukier]

I am running around eight solr servers (version 3.5) instances behind a Load Balancer. All servers are identical and the LB is weighted by number connections. The servers have around 4M documents and receive a constant flow of queries. When the solr server starts, it works fine. But after some time running, it starts to take longer respond to queries, and the server I/O goes crazy to 100%. Look at the New Relic graphic:

If the servers behaves well in the beginning, I it starts to fail after some time? Then if I restart the server, it gets back to low I/O for same time and this repeats over and over.