Comment 5 for bug 851691

We've deployed the production server, first with a solandra ring, and when that fell down under testing we threw it away and deployed a haproxy-fronted trio of solrs, and so far testing has indicated this will work. There still is room for improvement, and we will be doing some things to make it better, but as it stands it can handle over 3k concurrent requests without breaking a sweat, delivering 99% of results within the second. When under too much load it starts returning empty results rather than falling over completely.

We expect to improve the throughput by an order of magnitude in time for release.