Does anyone have a list of 'donts' when optimizing?

[BUU]

Mostly im asking if anyone has something like 'Top 5 functions to avoid when trying to write speedy code'.. If any function takes alot of resources or is really inefficient. Or hell, anything related to optimization, particularly in a cgi context, but in general as well.

[talexb]

Don't work on speeding stuff up that has little or no effect on the overall efficiency.

Reminds me of the time I helped my friend move a clothes washer. I insisted that we remove the agitator first, and spent five minutes removing it to find -- surprise! it only weighed about a pound, and we still had to carry the rest of this 180 pound monster up a flight of stairs and onto a pickup.

He was very patient. Never said a thing, but he sure smiled a bit.

--t. alex

"Nyahhh (munch, munch) What's up, Doc?" --Bugs Bunny

[Kanji]

To echo talexb, be very wary of premature optimization; you can waste an obscene amount of time optimizing things that you know to be relatively inefficient but pale in significance against less obvious bottlenecks in your CGI.

Anyhoo, searching for 'optimization' turns up a fair amount of nodes, and I'm sure Super Searching would turn up even more, but two nodes (RFC CGI.pm refactoring and CGI::Simple vs CGI.pm - Is twice as fast good enough?) that might be of particular interest ...

    --k.

[thelenm]

My favorite quote about optimization:

Rules of Optimization:
Rule 1: Don't do it.
Rule 2 (for experts only): Don't do it yet.
- M.A. Jackson

And some others I blatantly stole from the Java Optimization page:

"More computing sins are committed in the name of efficiency (without necessarily achieving it) than for any other single reason - including blind stupidity."
- W.A. Wulf

"We should forget about small efficiencies, say about 97% of the time: premature optimization is the root of all evil."
- Donald Knuth

"The best is the enemy of the good."
- Voltaire

[Fletch]

If you can at all manage it, don't run as a CGI. Use FastCGI or mod_perl. That'll immediately cut the overhead of each request by a large amount. And see the excelent mod_perl guide for mod_perl specific tips.

If you must have real CGIs, fork off a child (or pass off the work to another persistent process) to finish any long running tasks after you've interacted with the browser. That'll let the httpd you're interacting with go on and service another request.

[dws]

Mostly im asking if anyone has something like 'Top 5 functions to avoid when trying to write speedy code'

Avoid   sleep() Seriously, though, my answer is:

"Don't worry about speed before it's clear that a) speed is important, and b) speed is a problem, and c) the problem is one you can get at."

I've seen too much time go down the drain in projects when somebody decides to rework some chunk of the system to be as fast as possible, when it really didn't matter. The result is usually code that's problematic (less readable, less supportable) than if they'd just been straightforward.

Performance problems are almost always algorithmic (where "algorithm" includes the structure of SQL queries).

[tachyon]

To optimise you need to identify where the time is spent. The most useful tool is Devel::DProf which will detail the percentage of time spent in each section of your code. You are wasting your time doubling the speed of a part of code that only takes up 1% of the total runtime. On the other hand a small improvement in bottleneck areas will reap large benefits. Assuming you have a 'normal' database driven website consider:

1. Profiling with tools, see Devel::Dprof is your friend
2. Optimising bottlenecks with Benchmark et al
3. use mod_perl and Apache::Registry or for vanilla CGI CGI::Simple or CGI.pm in order of speed to process your params.
4. Generate a cache of static documents for slow changing content rather than serving all requests from a database
5. Connect to your database once (slow) and use prepare_cached($sql) with ? placeholders in the SQL and then execute(@values)
6. Think about how much your time is worth versus the cost of new faster equipment or more ram or more processors or faster disks or a better load balancer or....

Often perlcode per say is not the bottleneck. You can't effectively optimise unless you identify where it is worth investing the effort.

cheers

tachyon

s&&rsenoyhcatreve&&&s&n.+t&"$'$`$\"$\&"&ee&&y&srve&&d&&print

[wardk]

one "DO" comes immediately to mind as I have dealt with it recently with a system that went from reasonably few transactions (departmental) to high volume (enterprise wide usage).

If using a database, create as few connections as possible. DB connections are expensive and in a high transaction CGI (as well as non-CGI) scenario, creating multiple connections where less (one is best) will do will have a significant impact on your response time, as well as anyone else you may be sharing db resources with (like listeners).

Of course this suggestion (as well as many others you may receive on this) is really valid for all languages.

[Biker]

Dont:

while(1){}





---

Everything went worng, just as foreseen.

[Molt]

Not really hints on how to optimise, more hints on how to avoid a lot of problems with it. Make sure your code works beforehand, and use unit tests such as Test::More, or Test::Harness to make sure your code remains working whilst you're torturing it. Yes, there are rather a lot more of them to chose from if you look.. this is one wheel important enough that it's worth having a few choices as to what colour you want.

Oh, and make sure you know the output from your profiler. There's nothing worse than optimising the wrong part.

<CITE>"Premature optimization is the root of all evil." - Donald Knuth</CITE>

[simon.proctor]

Don't think theres always another optimisation you can apply.

You'll end up wasting time shaving that extra microsecond off when you could be focusing on a better solution, an n-tier application for example.

Update: fixed my bad English :). Guess I'm too tired.

[danichka]

Optimizations always bust things, because all optimizations are, in the long haul, a form of cheating, and cheaters eventually get caught.

--Larry Wall

My list of don'ts is pretty short. Don't get caught! :)


use Your::Head;