RIB is a generic benchmarking framework for investigating performance related
effects due to sharing of resources such as the memory architecture or file
system among multiple software components.

The support for benchmarks includes:

- Creating and executing parametrized benchmarks -- the user can
  specify ranges of individual parameters, the framework executes the benchmark
  with all the specified combinations of the parameter values.

- Collecting precise timing information through RDTSC instruction
  and performance counter values through PAPI.

- Executing either isolated benchmarks or combinations of benchmarks
  to investigate the sharing effects.

- Plotting of results through R. Supports boxplots for
  examining dependency on one benchmark parameter and plots with multiple lines
  for different values of other benchmark parameters.

- Utility functions such as page coloring based allcoation.

=========

For installation instruction, see the INSTALL file.

=========

Usage instructions:

running the compiled 'rib' binary executes the framework. It expects rib.ini in
the current directory.
See example-configs for examples of rib.ini

rib.ini defines
- names of two workloads ('benchmark' and 'thrashing') whose execution is
  interleaved
  currently supported are "none" (use it to execute only one workload),
  "cache_artificial" (memory acccess workload with random or linear strategy),
  "cache_col" (workload accessing a single TLB or cache set)
  "cache_fft" (workload executing FFT transform using either simple FFT
  implementation or FFTW)
- "warmup_count" defines the number of samples that are discarded before
  collecting
- "samples_count" defines the number of samples that are collected in one run
- two sections [benchmark] and [thrashing] to define parameters of the defined
  workloads. currently please see comments in the source code of the
  individual workloads in src/benchmarks/ for their description
- there are two parameters common to most of the workloads if RIB is compiled
  with PAPI support - "papi_counters" gives the max. number of events to collect
  simultaneously, "papi_events" lists the event names (execute  
  papi_native_avail in thirdparty/papi/ to get alist of events supported on 
  your architecture)

You may run "rib" repeatedly to collect more samples, the number of collected
samples is stored in the rib.samples file.

Running "ribplot" binary creates plot of the collected results.
When run with no parameters, produces boxplots against every workload parameter
that changes.