Author  : Edison Figueira Junior  http://efigueira.blogspot.com
Sponsor : BRconnection http://www.brc.com.br
License : Free - GPL
Revision: 1.0

* Jul 14 2008 - Edison Figueira Junior
- Make this doc 1.0

BRMABench  is  a  system  of  benchmark  per  comparison. This benchmark makes
diverse  routines  of  test  in one determined time, the final time expense is
always  the same and what it determines score it is the amount of interactions
in this same time.

It is common administrators of networks and servers  to concluirem that upgrade
of the determined hardware is necessary to make. But few have 100% of certainty
what it really improved in this possible upgrade, or  in  extreme cases to have
certainty that really something changed.

For this i developed a system of Benchmark per comparsion the BRMABench.
This benchmark makes diverse routines of test in one determined time, the final
time expense is always the same and what it determines score it is  the  amount
of interactions in this same time.

The BRMABench has 5 groups of tests:(file_io, proc_l, process_i, shell_exec and
socket_net).
*file_io, this group of test it make  stress of I/O in the machine, it is not a
DISK test, until makes a difference this test to be carried through in slow and
fast medias, but it is not total determinative. What  it  makes  is to test the
performance of transference of data between the memory/processor.

*proc_l, it make processor load tests, with  this test  is possible to perceive
the difference of machines with clock different.

*process_i, makes simple tests of mathematics basic with integers.

*shell_exec, in mixing systems, i want  to  say  in  machines  that  have  many
different routines running at the same time, the score of this test it  is very
important.

*socket_net, makes tests of bind,  connection,  transference  and  unbind. Very
important for network jobs.

The number in the side of the test (file_io10) is  the  simultaneous  number of
processes of that test. It is  important  to  observe  if  it does  not exist a
drastic fall between the concurrence.

A VERY important detail to be observed is that when score of I/O is made a test
in monoprocessed machines normally he is higher than score in machines DUAL.
This is explained because when it have more processor in the machine, the OS it
has that  to  control flow  between  BUS, MEMORY and CPUs, this cause a natural
fall. But in low cost machine this fall can be drastic. Attention  for this  if
its application is candidate the I/O load, like (EMAIL,DATABASE).

Download the package, extract in some directory and run $./brmabench.pl as root.

1) Always run the test with all services stopped.
2) It compares final score between the tested machines.
3) It compares the scores individually and it analyzes all  in  accordance with
   the points the function of the machine.
4) It is not necessary to make the test more than a time, the BRMABench already
   makes this 3x internally  and score is based on the average.