I have a set of image files, they are generally very small, between 5k and 100k. They can be any size though, upwards of 50mb but this is very rare. When these images are put into the system they are not ever modified. There is about 50 TB of these images total. They are currently chunked and stored in BLOBs in Oracle, but we want to change this since it requires special software to extract them.

These images are access sometimes at a rate of over 100 requests per second among about 10 servers.

I'm thinking about Hadoop or Cassandra, but I really don't know which would be best or how best to index them.