On Nov 3, 2008, at 4:34 AM, Jan Lehnardt wrote:
On Nov 3, 2008, at 10:26, Ask Bjørn Hansen wrote:
On Nov 2, 2008, at 21:19, Paul Davis wrote:
View updates are admittedly slower than we'd like. There's planned
work on making this sort of thing parallelized to automagically fill
out multi-node clusters. Last I remember this is a 'probably 1.0'
feature though.
I wonder if some of the slowness is from running it through JS?
Perl has an extremely fast JSON parser (JSON::XS), would there be
any potential optimization in using an alternate view language like
that - or is most of the time spent elsewhere?
Another part is the Erlang to JSON conversion that is known to be
slow-ish. I don't know
which side of the conversion weighs more though.
I have to assume our Erlang json parser is much slower than the
spidermonkey one, which is just a call to eval(). Based on what I've
seen when profiling Erlang code, a built-in parser will give us a big
speed-up in view indexing.
-Damien
We are working with the Erlang
community and developers to get a C-based JSON encoder into Erlang.
Since building
a view server is relatively straightforward, giving a Perl version a
shot is probably worth it.
Which Erlang version are you running. R11B-x is known for causing
view-build slowness.
R12B-3 and R12B-4 are good.
I don't expect more than a few tens of thousands of new documents a
day though, so other than on index changes this isn't such a big
deal.
That sounds about correct.
Cheers
Jan
--
