cloud-papi

this project implements a Mesos Executor that adds
papi (performance counter) support to Python, JVM,
R, Ruby, and C/C++ applications running on a Mesos
cluster/cloud.

the implementation uses a fork-exec technique to
attach the papi library to the user's process in
order to collect performance counter data

for python, jvm, r, ruby users, each system's
runtime (interpreter or virtual machine) is embedded
into the executor - this was a requirement in order
to collect accurate performance counter data using
papi's 3rd party, process attachment feature set

papi only supports collection of two counters
worth of data

data is stored using google's log library

users will be able to run the papi mesos executor
by adding a label to a task called 'PAPI' and then
add a space delimited string which enumerates up
to 2 papi performance counters to enable.

cloudpapi.cpp, line 239 provides an enumerative
listing of available performance counters.

-----

container users won't necessarily be able to use
the MesosExecutor. papi will only work when it's
attached to the interpreter's runtime executing
the interpreted program.

to that end, embedded versions of python, jvm, r,
and ruby have been integrated with papi's third
party attach capabilities.

containers desiring papi support will need to add
a papi-enabled version of their respective
interpreter to the container.

-----

Users will need to modify the project Makefile
in order to compile the codes. The following
Makefile variables will need to be modified.

* MESOS_HOME = path to the mesos source tree

* RTSUPPORT = used to select which interpreter(s)
              support to enable

-----

dependencies

  * PAPI http://icl.cs.utk.edu/papi/
  * Python, R, Ruby, Jvm and their respective dependencies