[development] Why do we avoid auto-incrementing db columns?
gerhard at killesreiter.de
Wed Apr 25 17:11:36 UTC 2007
-----BEGIN PGP SIGNED MESSAGE-----
Earnest Berry III schrieb:
> Hi Sammy,
> There's only on thing I disagree on; the performance hit of the mutex lock.
> If you do not have a relatively large number of insertions, than this is not
> a large performance hit; however, if you do, this does become a performance
> hit. I admit, I'm on the extreme end of the spectrum as I was inserting some
> 20 million nodes (which, expanded means 20 million nodes, 20 million
> node_revisions, and 7-14 million other insertions in diff. tables for the
> After implementing random-id generation with out locking the table ever at
> all, the performance increase was very noticeable, on the order to 70-100%
> more throughput of node creation.
You'll find this interesting.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.6 (GNU/Linux)
-----END PGP SIGNATURE-----
More information about the development