On Thu, Sep 8, 2011 at 6:46 PM, Robert Haas <robertmhaas@gmail.com> wrote:
> I'm not convinced it's anywhere near that easy. For one thing, on at
> least one big server I'm playing with, memory latency on shared memory
> is vastly higher (like >10x!) than on backend-local memory due to NUMA
> effects.
I wonder if both the shared mem and non-local memory issue can be
circumvented by using a slru like mechanism as a side channel to
publish taken snapshots and make concurrent xids available with a
sinval/hasmessages like per proc flag in shared memory to notify of
migrated snapshots.
I'll have to think through the space, locking and performance
considerations. That might take a small while though, I just managed
to contract the flu and can't really think straight.
Sorry to waste your time if this whole approach is completely untenable.
It seemed like a interesting topic to sink my teeth in, but in hindsight
seems a bit too much for a first try.
--
Ants Aasma