Hello Julien, JR>
JR> I can give you a few timings I have to give you an idea: JR> JR> Initialisation: connects to DB, pro-compile some queries, load JR> structured data (SELECT * FROM Objects; i.e. sequentially, no blobs) JR> Load Blobs: load 1/4 of all Blobs (each ~23k of size, I just checked), JR> "randomly" JR> JR> -Uncached JR> -14s to initialise JR> -31s to load blobs JR> JR> JR> -Uncached, but VACUUMed JR> -3s to initialise JR> -16s to load blobs JR> JR> -Pre cached, no VACUUM JR> -3s to cache and initialise JR> -1s to load blobs JR> JR> Notes: JR> - VACUUM took (as expected) a very long time to do it's task (few JR> minutes). JR> - This test is for a ~100MB file. From other tests I made it looks like JR> the timings depend linearly on size of file, and number of blobs loaded JR> JR> Pre-caching is clearly a winner here. This agrees with my own experiments. Precaching either by sequential reading or creating a file mapping and reading the data before you open the DB gives a significant boost in performance when the DB is opened the first time. C ----------------------------------------------------------------------------- To unsubscribe, send email to [EMAIL PROTECTED] -----------------------------------------------------------------------------
