#musicbrainz

bah, too much python lately... keep forgetting adding ";" in other language

s

alright, bed time

natta!

right... so "/etc/init.d/apache-perl restart" won't do the trick, i actually have to explicity stop and start for it to reload changes

canidae: Actually, it's worse than that, stop && start won't necessarily do it, I find a sleep in the middle helps. Make sure it prints the Preloading 230 components line. :p

yes, i noticed it didn't do that on restart, it does on stop & start, though

anyways, i think i can nail this now as i actually can debug stuff now

i'm using an old computer for this, duron 1200, no need for sleep :p

although, raid5 kicks butt

it's disturbingly fast on i/o, compared to chewing numbers

yeapz, READWRITE probably shouldn't be undef when you set it up as RT_SLAVE

*READONLY

since i just copied from READWRITE to READONLY it does have write access, but i'm the sole user, and i didn't set it up for changing data anyways

seemingly the documents doesn't tell you what to do if you want to set up a slave, but i might just've missed it too

aww, come on.... ~125 rows/second? creating this index' gonna take a week

ok, fine, only ~24 hours

thats a little slow. :-(

1200 duron

how much ram?

384mb, old box

don't have newer hardware to spare, really

ok, it picked up a little, ~190 rows/sec now

what kind of ram does it take?

i've no idea... probably whatever was cool before ddr hit the market

pc-100 or pc-133

I still have some of that laying around...

well, no stress... this isn't gonna be a permanent setup, it's just for testing. also, i mostly just have this week of vacation i where i got time/strength to play, so by the time it would come it'd probably be too late

i'm not so familiar with pylucene, although perhaps it could be an idea to look at clucene and make some bindings to python? iirc pylucene is more or less a hack, gcj compiled java code or some crack like that :p

meh, this was tedious to watch. maybe it'll be more exciting after some hours sleep. *poof*

gcj compiled java lucene, C python bindings into that

last time i looked at clucene the featureset was weak

xapian looked like it had improved a lot

one of these days i'll get around to trying it again

know of anyone who has compared it to lucene?

not recently, but i haven't spent much time looking

i was just poking through their site a couple months ago and noticed that things had progressed a lot since i had last evaluated it

and while lucene is nice, java is not

ding!

xapian would be a lot easier to deploy

which is the only reason why I'm even looking at it.

one should not core infrastructure pieces like search engines in java.

*write

i don't know that c++ is much righter though

is the index creating thing ram-dependent then?

dsp: its better than java.

index creation, no. searching is both heavy on cpu and on ram

fortunately someone is donating 2 servers with 8 gigs of ram and dual quad core procs

oh, lucene indexes...

nice

i've still been buying amd for search b/c of the extra memory bw

but it looks like intel is solving that shortly

I should get around to buying more ram...

at work tomorrow i have a machine showing up with 32 GB, yay

nice

~4 yrs ago at a diff company i had itanics with 24 GB of mem :)

i can't believe the itanic still exists

my macbook has a gig and that's the most I've ever had in one computer :P

that *has* to be a political thing

for a laptop that is fine

for search you need a lot of mem if you have big indexes and want a lot of q/s

is it just me or is svn.musicbrainz.org s l o w?

its copying out the weekly data dump right now.

so you're only getting tablescaps of bandwidth right now.

ah

is there a time delay on the freedb gateway ?

up to 70 mins

e

does barcode and ASIN do anything really ?

I think they both play jazz instruments in some band, IIRC

I actually laughed

lon vs lol ?

sorry, loi

it was a little chuckle so I guess lol

I've added quite a bit of data today, hopefully someday in the future it will be come useful

I'm sure it will. :-)

just spent some time playing around with xapian again

the indexes are still huge compared to lucene

I'm not so worried about that.

the indexes end up 2x the source data

whereas with lucene they're more like 30% of source

ok, I can deal.

how is the index speed?

and the search speed?

25k emails (127 MB) in ~ 3.5 minutes

2 GHz athlon 64

ouch.

not good.

is it slow because it is written in java ? or simply because it cannot handle those types of searches ?

xapian is c++

i'm using it via the python bindings

java is actually pretty fast these days

assuming it is used well

java lucene is still quite a bit slower than pylucene with gcj

have you actually measured that?

i saw ppl say that a lot, but it was never my experience

not qualitaively.

xapian supposedly supports index updates better than lucene, fwiw

or at least not in a scientific fashion.

luks was never able to gety more than a fraction of the build speed that we get with gcj

was he writing a competing java soln?

i never paid too much attn to index speed

solr?

s/soln/solution/

sorry

ah.

a lot of times i was using indexes that were built by java stuff

to prototype newer search features

no just using lucene/solr in java vs our own pylucene indexing stuff

the search side of pylucene was never as fast as java

but i didn't spend much time looking into it

so it could have been a lot of things

ah

at pycon last weekend i learned of something called grassyknoll (http://rds.yahoo.com/_ylt=A0geu8o1hOBHW9MAsg9XN...)

grr yahoo

anyway, it is some wrapper for search crack for python

i haven't had a chance to look into it yet

interpretive languages aren't the best choice for search engines

depends on what you're trying to do

i don't think there are any hard and fast rules

wth, pylucene now uses jcc?

i don't even know what that is

(as opposed to gcj)