I am trying out lmdb as an inverted index storage. My usage pattern is very write-heavy
initially (by appending a lot of small bits to existing keys) and when the indexing is
done, there is only read-access needed.
 
During testing I found that write speed decreases with the amount of appends done to the db.
So I wonder if it would bring any benefit, if I write the existing key/value pairs to a second db
(after a certain amount of values appended). The purpose would be to "compact" the values
into one big memory blob per key (this time overwriting values instead of appending).
Would this gain a bit of speed?
 
-Tom