Code Performance

It seems to me that some developers often give a shrug of the shoulders response to performance problems. The suggestion these developers make is that the slowness is just a price the users have to pay for the functionality. It is true that solving performance problems is often a tedious task without any obvious solutions.  For starters, the code works, so why muck with it? There are two very good reasons.  They usually are not that hard to fix and everyone thinks you're a genius when you do fix them.

The hardest part of fixing performance problems is identifying them.  This is where profiling and heap analysis tools come into play.  The rule of thumb is that 5% of the code will be 95% of the problem.  I'm sure people are making up the percentages, but the point is simply that fixing one or two small pieces of code will make the whole application faster.  Finding the cause is job one.  That seems like a simple statement, but more often than not developers will be quick to throw out solutions firsts.  "Let's get new servers", "We need to make it multi-threaded", "The network is slow", etc...

Plenty of open source profiling and heap analysis tools exist to help track down bottlenecks.  Not to mention simply dumping timestamps to the console can go a long way.  Many areas of the code are also natural places where a slow down could happen.  These include communication across the network, loops, object creation, and database access.  When you find these areas, try to avoid using gimmicks that likely will only moderately improve performance while making the code harder to read.

Instead, try one of the following top 7 ways to speed up your code.

1. Increase memory
2. Add caching
3. Write efficient SQL with joins and indexed columns
4. Do more in each step
1. Optimize loops (attributes, unrolling)
2. Chunks across the wire, database, etc
5. Choose the right collection
6. Add more threads, with care
7. Reduce object creation