Is there any way of compromising somehow? This feels like wishful thinking but I do not know. Anyway - somehow semiarchiving old stuff so it isn't actually affecting anything except on the rare occassion someone isn't going through old threads. Hmm, no, this doesn't sound likely. Hm.

 

having it all separate and readonly feels wrong somehow.

We could convert old stuff to static HTML files, and then use a search engine to index both the database and files together.

Lucene is an Apache open-source one of those, which can be integrated with PHP via Zend, although again it appears to be a PHP5 solution.

I'm not entirely sure if I know what I'm talking about but would it not be possible to periodically move posts/threads etc. that are older than a certain amount of time (say a year) to a seperate table (or tables) which are then only ever searched when absolutely neccessary?

 

So regular thread list 'stuff' would only look at the 'current' table(s) wheras the Search might look at both the 'current' and the 'archive' table(s) using whatever SQL magic that requires.

 

Any time a post is retrieved it would compare its number to the 'last-archived-thread' and based on that decide which table(s) to get it from. For the most part it would only require recent posts so only ever do stuff with the leaner 'current' table(s), thereby making everything faster, finding a cure for AIDS and achieving world peace.

 

Does that make sense? Am I repeating people? Can I go home yet?!

 

Hmm. But then posting in a thread from 3 years ago might cause problems, or would it? I don't know.

I think the main problem with that is it would be hard as fuck to make. And would pretty much be a one-off. So no one will make it.

 

And also the script which does the splitting would have to be run as a cron job or something like that. And would be, I reckon, slow as fuck.

 

(That's not to say it's a bad idea, it would be the ideal solution, just I don't see it happening)

 

I think we have three choices:

 

1. something like you suggested
2. carry on as we are
3. start again and archive the rest somewhere

 

I think 1 won't happen and 2 shouldn't happen so I like 3.

Or 4) Look around for new hosting.

S'pose. Seems like overkill though. We shouldn't need a dedicated server for a web forum.

It wouldn't need to be a cron job - it could be a manual admin function. This place has got, what, five years of posts? So even on busier places it'd probably not need to be run more than yearly, thus speed isn't a huge issue; people expect/accept that archive operations take a period of time.


The search page could have an additional option:
Search [current|archived|both]
Or maybe it could be automatic based on the from/to dates selected, with a message signifying whether archived messages would be searched.


As for the difficultly of implementing it, I don't agree with that either.
You could keep the existing queries as they are, and if relevant do a second SELECT on the archived bits and UNION it to the initial query.


And finally, regarding the no-one will make it bit, I will be working on integrating Beehive with my next site in (hopefully) a month or two, so if nobody else does it I may have a look at doing it.


:)

True, but this server has always been slow. It's never been as fast as some of our previous hosting has been. Granted it's mostly been faster than recently, but never blazingly quick.

Hmm. I wasn't really thinking of it as anything more than a one-off. If it would be a useful feature generally then, aye, makes it worth it.

 

It'd still take ages to execute here and that is a problem.

 

As for complexity - yeah, it could be done as you say but... why bother? That offers no advantage over just making this forum read-only and starting again. Well, little advantage.

 

If it were to be done properly it would happen in a sort of rolling way. And that would be hard. By which I mean not having to run a script to do the archiving but mechanisms put in place whereby older posts (beyond an arbitrary threshhold) aren't involved in queries. But... I have no idea how that might be done.

 

But aye, if you're up for making it then that's cool by me, of course.

I'm just looking at the message SQL with my optimizing hat on, and there's quite a lot that could be done to reduce the load on the database there.

 

The SQL includes a join to the THREAD table from every post, which is redundant. The values from the THREAD table can be retrieved in a single SELECT before the message SQL is run.

 

There are four joins to the user table, one of which (APPROVED_USER) is only relevant if posts have to be approved by an administrator, and another (EDIT_USER) only if the post has been edited by someone other than the original author. The APPROVED_USER join could be added in conditionally, and the EDIT_USER could be done using singleton SELECTs after the main select is finished.

 

This one might be a stretch: there are two joins to the USER_PEER table for the relationship and the new peer nickname thing. It might make more sense to do a single select against the USER_PEER table for the logged-in user and cache the result in an array, then get the values from that. It kind of depends on how many peers people have, on average. I've got 23, mostly ignoring sigs. Some testing might be advisable for this one.

Also, I reckon the slow performance in the threadlist could probably be improved just by deleting most of the records in the USER_THREAD table, e.g. ones for threads which haven't been updated in the last three(?) months.

Would that not make all the old threads show up as unread?

That's what Delphi used to do, I remember. Although maybe their cut-off point was closer to 6 months. Anyway.

(Pete: You just modify the query to be "SELECT [unread] AND [most-recent-post-in-thread-date > NOW() - 3months]").

Yes, yes it would. I submit that that doesn't really matter one tiny little bit.

Done all that (the joins I mean). It's not in CVS yet, but it's live here. The messages_get function now only uses POST, USER_PEER (x2) and USER (x2). The join to the thread table now uses the previous call to thread_get() in messages.php and the other joins to the USER table have been moved to separate queries to be done only if the post needs approval or the edit message needs displaying.

 

I've been running a profiler against the code on my own machine (with a copy of the database from here) and noticeable bottle necks are the emoticons code which takes up to 500ms and most noticeably the threads_any_unread() function which is taking at least 1 second to complete(!), compared to the next longest running function being the RSS feed checker which takes 7ms. The threads_any_unread function is usually the cause for the load here. I've tried fiddling with it and adding indexes to the tables and rewriting the query to better use said indexes but it makes diddly squat. It even has a LIMIT clause on it now when it never used to but even that makes no difference, so if it's not the USER_THREAD table that's causing I'm stumped as to what it could be.

The emoticons code as in running that preg_replace against every post at runtime? There any other way to do it? (besides the way it was originally set, converting the emoticons at post-time)

It's the calls to file_exists() in Emoticons->Emoticons() that is the resource hog. It only takes 0.4ms each time to call file_exists() but it's being called quite often. I'm not sure if there is a way around it, maybe initialize the emoticons class within the emoticons.inc.php itself (like the gzipenc.inc.php script does) so when another script includes it it's done automatically. Also maybe don't use file_exists() but rather suppress the error from include() when it doesn't find the definitions.php file.

On my test page, I have 20 posts all with the same number of emoticons and convert() only takes ~35ms to process all 20 of them.

That link just takes me to your portfolio page..

http://83.67.95.239/forum/index.php?webtag=DEFAULT&msg=16024.16

 

Maybe?