Zdenek Kotala wrote:
> Does we need random_bool to spread workload? It seems to me a useless,
> because it also invokes one backend to use more pages instead of using
> one which is already in buffer cache.I think that it should generate a
> lot of extra i/o. Do not forget WAL full page write for firstime
> modified page.
random_bool() is gone in the latest version of the patch, in favor of a
"next pointer". You must be looking at an old version, and I must've
forgotten to update the link in the Wiki. That change was discussed in
the "New FSM allocation policy" thread.
Anyway, here's is the new version for your convenience, and I also added
a paragraph to the README, mentioning that the tree is degenerated from
the right.
--
Heikki Linnakangas
EnterpriseDB http://www.enterprisedb.com