>>>>> "Troy" == Troy Grady <troy@stripped> writes:
Troy> In a related note: I've written my own text index, which is essentially a
Troy> table with two columns -- a list of unique words, and a TEXT field
Troy> associated with each word containing a comma-delimited list of ids of
Troy> documents where the word can be found. Ids are added to the TEXT field by
Troy> CONCAT-ing a comma and a new id to the end of the field. So far,
Troy> performance has been great, but if what you're saying about fragmentation is
Troy> true it sounds like I can expect performance to slow as I continue to CONCAT
Troy> ids to the blobs. I can't see any way around using CONCAT here, since
Troy> keeping the ids on one row is essentially what makes it an index. If anyone
Troy> can think of another way of doing it, I'd appreciate the insight.
Troy> Otherwise, is this something that regular disk defragmentation can remedy?
Note that you can fix the fragmentation by doing OPTIMIZE TABLE once
in a while.