Description
-------------------------------------------------------------------------------

This is a benchmark suite for different implementations of the SAGA Advert 
Service (http://www.ogf.org/OGF27/materials/1767/saga_adverts.pdf) and their 
backend implementations. It takes three parameters:

(1) Number of components: This describes the number of individual (potentially
                          distributed) processes that access the advert service
                          concurrently.
                          
(2) Number of attributes: The number of attributes that "belong" to each 
                          component. The attributes are stored in the 
                          component's advert server entry and are accessible by
                          other components.

(3) Number of iterations: Number of iterations the main benchmark loop is
                          executed and measured. More iterations usually mean
                          a more accurate average. 
                          
The Main Benchmark Loop

  The main benchmark loop is controlled by the parameters above. The control 
  flow is quite simple:
  
  For each Iteration

    (1) A component CX opens the advert workspace of another, randomly assigned 
        component CY.
  
    (2) CX lists all of CY's attributes

    (3) CX reads each of CY's attributes 
        (# 'read' calls = Number of attributes)

    (4) CX writes its own attributes into CY's "inbox"
        (# 'write' calls = Number of attributes)
    
    (5) CX records the duration that it took to execute the above steps
                        

Example Usage
-------------------------------------------------------------------------------

(1) Using the default saga advert service with 10 components (workers) 10 
    attributes per component and 5 iterations. The components are running 
    locally (using the default job adaptor):

    $> python advert-benchmark.py \
       advert://SAGA:SAGA_client@advert.cct.lsu.edu:8080/benchmark \ 
       -p -c 10 -a 10 -i5
    
    The results are human-readable and accessible in the advert database:
    
    $> saga-advert list_attributes \
       advert://SAGA:SAGA_client@advert.cct.lsu.edu:8080/benchmark/jobstats
    
    An entry for an indivdual component may look like this:
        (vector) comp0 	:
            STATUS:DONE, 
            STIME:Wed,  07 Sep 2011 14:59:56 +0000, 
            ETIME:Wed,  07 Sep 2011 15:00:13 +0000, 
            ITERTIME0:1.00112605095, 
            ITERTIME1:1.00112605095, 
            ITERTIME2:1.0011241436

    The keys have the following meaning: 
      STATUS:     Can be OFFLINE (not yet started), ONLINE, DONE or FAILED
      STIME:      The component startup time
      ETIME:      The component shutdown time
      ITERTIME(X) The duration of the Xth iteration
      
    The log-files of the individual components can be found in /tmp/. The 
    location can be changed with the --tmpfile option. 
    
      
(2) The same works also with the new 'node.js/mongoDB' advert service, just by 
    changing the advert url:
    
    $> python advert-benchmark.py \
       sqlasyncadvert://faust.cct.lsu.edu:8080/benchmark \
       -p -c 10 -a 10 -i5

    $> saga-advert list_attributes \
       sqlasyncadvert://faust.cct.lsu.edu:8080/benchmark/jobstats