[Date Prev][Date Next]
In-Memory database backend for slapd ?
has anyone thought of using/implmenting an in-memory database backend
My setup is a 250,000 object database with 50,000 people and 200,000
groups. As an ldif file this is aprox. 50 Mb. My machine has 1gig of
memeory of wich 600Mb is currenlty free. I have caches set to 100x
standard so effectively my whole database is in memeory, the slapd
process is currently using 160Mb or ram.
I wonder why bother with "disk" based indexing schemes when all my data
and all of the index information can easily fit in memory. It must be
more efficient, faster to do things "memeory based".
Naturaly one whould have to consider startup, shutdown and period
dumping of the current data from the memeory based system ( ldbmcat )
but even if it took 2 to 5 minutes to start, my service operates for
weeks and months at a time without reboots so a longish startup is not a
problem to me.
My usage is for authentication and role based access control so there is
very little updates of the information but a need for very fast reads. I
am also using alias's so am a bit concerned as to the performance
ramificiations when having to do another lookup but the use of alias's
solves a realy huge data usage problem I have.
I recall the hub-bub around in-memory databases some time ago but the
details have not stuck. Besides its friday morning, my workstation has
crashed ( hard disk ) and I figured that a guru ( I don't qualify )
could knock up the new backend over the weekend :)