Thread: Lockfree hashtables
The other day I bumped into some ideas about lockfree hashtables. Are these of any use in PostgreSQL? http://blogs.azulsystems.com/cliff/2007/03/a_nonblocking_h.html http://video.google.com/videoplay?docid=2139967204534450862 -- Sincerely, Stephen R. van den Berg. "I hate spinach, and I'm glad that I hate it, because if I wouldn't hate it,I would have to eat it, and I hate it!"
On Wed, Dec 31, 2008 at 7:33 AM, Stephen R. van den Berg <srb@cuci.nl> wrote:
Lock-free and wait-free algorithms have been used in various databases, but most people tend to shy away from them because of their complexity, difficulty to debug, and low-level portability issues.
I've used them in the past (lock-free hash tables and skip lists), and they're pretty awesome if used properly, but the majority of PG's current performance problems aren't generally found as part of our hash table implementation (which I'm quite fond of actually). FWIS, I think we'll look more into this sometime in the future.
--
Jonah H. Harris, Senior DBA
myYearbook.com
The other day I bumped into some ideas about lockfree hashtables.
Are these of any use in PostgreSQL?
Lock-free and wait-free algorithms have been used in various databases, but most people tend to shy away from them because of their complexity, difficulty to debug, and low-level portability issues.
I've used them in the past (lock-free hash tables and skip lists), and they're pretty awesome if used properly, but the majority of PG's current performance problems aren't generally found as part of our hash table implementation (which I'm quite fond of actually). FWIS, I think we'll look more into this sometime in the future.
--
Jonah H. Harris, Senior DBA
myYearbook.com