--------------------------------------------------------------------------------

              Register Interest and Continuous Query PST

--------------------------------------------------------------------------------
                                 how to run
--------------------------------------------------------------------------------

Reserve as many Linux hosts as possible.  If comparing to previous runs,
either use the same hosts in the same way or rerun the baseline using the
available hardware.

Select a host for MasterController that has at least 4 CPUs for scheduling
large numbers of clients, for example, newton or a jade.

Set up a local.conf as instructed below.

Run cq.bt on the master host.  Be sure to invoke batterytest with these
properties:
        -DnukeHungTest=true
        -DmoveRemoteDirs=true
The last argument causes batterytest to move directories to the test result
directory from the remote hosts after the test concludes, or when the test
is nuked.

Analyze the results as instructed below.

--------------------------------------------------------------------------------
                             sample local.conf
--------------------------------------------------------------------------------

See tests/cacheperf/comparisons/cq/local.conf.

--------------------------------------------------------------------------------
                               how to compare
--------------------------------------------------------------------------------

In all tests, verify that the feeder put rate does not fall below 1000/second.

Check the scalability of register interest and continuous query by comparing
riReplicate* to itself for each value of edgeHosts.  Same for cqReplicate*,
riPartition*, and cqPartition*.

Compare register interest to continous query by comparing the register interest
tests riReplicate* to the continuous query tests cqReplicate*, and riPartition*
to cqPartition*, for each value of edgeHosts.  Continuous query latency should
be comparable to register interest latency.

Compare the multiple continuous query tests, cqsReplicate* and cqsPartition*,
to the single continuous query tests, cqReplicate* and cqPartition*, with the
same host-VM mapping.  Multiple continuous queries that register and deregister
should cause no more than a 20% degradation in performance.

--------------------------------------------------------------------------------
