We haven't used any SAN's we've stayed to single box solutions (HP 165/185/385/585). On the PK I referenced earlier we needed to query individual records on a dataset of 600 million rows in under 20 ms. Due to the particulars of the query it had to come off a b-tree in memory. We tried innodb, but the load times took to long. The index itself was 14 bytes, so we didn’t want to use ndb as we were getting close to our memory limitations on the box. So we created the b-tree on a myisam structure and pinned it using the query cache. In retrospect the memory engine would most likely have been a better choice.