Stephen Frost <sfrost@snowman.net> writes:
> The current situation where the client-to-server latency accounts for
> multiple minutes of time is just ridiculous, however, so I feel we need
> some form of this patch, even if the server side is magically made much
> faster. The constant back-and-forth isn't cheap.
No, you're missing my point. I don't believe that client-to-server
latency, or any other O(N) cost, has anything to do with the problem
here. The problem, as Jeff has demonstrated, is the O(N^2) costs
associated with management of the local lock table. It is utterly
pointless to worry about O(N) costs until that's fixed; and it's just
wrong to claim that you've created a significant speedup by eliminating
a constant factor when all you've done is staved off occurrences of the
O(N^2) problem.
Once we've gotten rid of the local lock table problem, we can re-measure
and see what the true benefit of this patch is. I'm of the opinion
that it will be in the noise compared to the overall runtime of pg_dump.
I could be wrong, but you won't convince me of that with measurements
taken while the local lock table problem is still there.
regards, tom lane