#python - Sun 18 Feb 2007 between 22:06 and 22:17

NY Lost Funds



marcin_antok let's forget about zope - it's design model is really not in my taste (and speed of course)
sproingiekosh: no surprise, rdbms's are designed for throughput, not latency
koshsproingie: the problem is that the rdb had an abysmal design
sproingiekosh: mysql?
marcin_antso any other proposals?
koshsproingie: the person was making about 1000 calls to the db and then doing the joins server side
sproingie: it was foxpro I think
sproingiethat's not a relational db
triplah_wi really wish psyco supported generators :(
sproingiefoxpro is xbase
koshsproingie: it was some kind of relational db that only runs on windows, it used sql, tables etc
sproingieactually foxpro had a reputation for being rather fast in its day
koshsproingie: no db is very fast if you query it thousands of times and then try to join the results manually together
finding the intersection between a thousand sets is very inefficient
arnalhey
sproingiekosh: sounds like the overall design was insane
koshit was
however I have seen the same stupid things in other sites people have done
marcin_antskunkweb? is it good or not?
koshpeople wonder why their site that uses postgres is slow and blames it on postgres when they retrieved 10K records from it and all they wanted to do is know how many records and get the first 20
arnaldoes setuptools has been added to python 2.5?
sproingiemarcin_ant: unmaintained from what i can see
koshor in zope where people load up 6000 objects just to find out how many there are
I helped someone with a plone problem fix a case like that, they had an image archive thing and on the same page it queried the db about 4 times to load up several thousand objects just to find the length
so zope was useing up FAR too much memory and the page was running abysmally slowly
sproingieyah, an interesting thing about gmail, which *is* rdbms-backed
it guesses at counts
koshby changing just those four places the zope memory usage dropped to about 1/10th and ran about 100 times faster for those pages
sproingieit only gives you an exact count when it knows exactly
koshusually in a an rdb you can ask it for a count of results instead of returning the result set
marcin_antdjango - prefferred deployment method is mod_pytjon, it can also work with fastcgi and has it's own server.. ... what do you think about this one?
sproingiei recommend fastcgi
it's a lot more portable
koshsproingie: so the basic thing is that any solution can be made to suck badly enough when in the hands of people that don't understand what they are doing
sproingie: IO usage and memory usage are the usually ways they blow their limbs off
sproingiekosh: true, but sometimes you have to blame the framework when its default behavior is the wrong one
kosh: object lesson, ASP.NET and JSF viewstates
marcin_anteeeh it's really overcomplicated
sproingiewhy do these frameworks have to put every damn thing in the viewstate and not just the stuff that registers for an event callback
who knows ... but everyone pays the price

Page: 4 11 18 25 32 39 

IrcArchive

NY Lost Funds