
 In this directory you find simple programs that could be used to benchmark
DIPC's performance in your system. All of them are 'worst-case' distributed 
programs: They spend all their time communicating (possibly over a network), 
with no computations being done.

 You could use them in a single computer, and then over a network. In this
way you will know how communications over the network could affect the 
performance of your programs. See the programs' document to see how you
can determine the computer(s) used for the bench marking.  

 When running these benchmarks, keep in mind that Linux is a multitasking 
operating system. Other than that, the network used to connected the test 
computers are often shared among many other machines, which may be using it 
heavily. This means that the benchmark results could vary from run to run.


 Benchmarking a single computer

 For two reasons, the results obtained by using the benchmarks in a single 
computer may not be very useful: 

 1) The operations may be executed too fast for the resolution of the
benchmark's timing mechanism. This causes the operations to have an apparent
zero execution time. See the program in shmem directory. 

 2) The results are not as reproducible as I hoped for. According to my 
tests, the big differences between runs in the same computer make drawing 
any conclusions about its speed impossible. This could partly be attributed
to the multitasking nature of Linux. See the program in syscalls directory.  


 Any suggestions for improving the benchmarks are appreciated.

