The social data are fairly small documents, but we index lots of other things as well that is larger in size. Another factor is our large replica count, especially for the hot data.

We also have quite a lot of metadata.