Contact: Mark Stillwell <m.stillwell@cranfield.ac.uk>

Portions of this software were also developed by Henri Casanova
<henric@hawaii.edu>.

===========================================================================

REQUIREMENTS:

GLPK should be installed.

===========================================================================

COMPILE:

From the directory you checked out of git or unpacked from the tarball:

mkdir build
cd build
cmake ..
make

should create ./src/flexsched

Sorry, there's no easy-install package at the moment.

===========================================================================

USAGE:

./flexsched <quote list of scheduling algorithms> <input file> [<output file> [<estimate file>]]

  Example:   ./flexsched "METAGREEDY METAHVP" ./input ./output

===========================================================================

INPUT FORMATS: 

problem file:

<num resources>
<num servers>
<num services>
<elem. capacity> <elem. capacity> ... <agg. capacity> <agg capacity>
...
<elem. req> ... <elem. need> ... <agg. req> ... <agg. need>

estimate file:

<elem. req> ... <elem. need> ... <agg. req> ... <agg. need>

===========================================================================

OUTPUT FORMAT: 

One or more lines of the form: a|b|c|d|e|f|g|h|i

a: algorithm name (string)
b: expected minimum yield (fraction)
c: expected average yield (fraction)
d: minimum yield using caps (fraction, only for inaccurate estimates)
e: average yield using caps (fraction, only for inaccurate estimates)
f: minimum yield using weights (fraction, only for inaccurate estimates)
g: average yield using weights (fraction, only for inaccurate estimates)
h: minimum yield using equi (fraction, only for inaccurate estimates)
i: average yield using equi (fraction, only for inaccurate estimates)
 
===========================================================================

SCHEDULING ALGORITHMS:

* GREEDY_X_Y:
   X: pre-sorting of the services, always decreasing
   Y: selection of the servers, once all infeasible serers have been removed

  X = S1 | S2 | S3 | S4 | S5 | S6 | S7
  Y = P1 | P2 | P3 | P4 | P5 | P6 | P7 | P8

  * S1: No pre-sorting
  * S2: by max of needs
  * S3: by sum of needs
  * S4: by max of requirements
  * S5: by sum of requirements
  * S6: by max of sum of requirements and needs
  * S7: by sum of all requirements and needs

  * P1: most available resource in dimension of max fluid need 
  * P2: minimum sumload / sumresource after placement
  * P3: best fit in terms of largest requirement
  * P4: best fit in terms of total available resource
  * P5: worst fit in terms of largest requirement
  * P6: worst fit in terms of total available resource
  * P7: first fit (random fit)

* METAGREEDY
    Try all greedy algorithms and pick the best.

* MILP
	Compute the optimal solution

* LPBOUND 
	Compute the rational bound on the optimal solution

* RRND
	Solve the rational LP and do a probabilistic round-off of
        rational values into integer values. 

* RRNZ 
	Solve the rational LP and do a probabilistic round-off of
        rational values into integer values. But first, set all 
        zero-valued variables to some epsilon (0.001)

* VP_X
   X: vector packing heuristic

   X = FF | BF | PP | CP 
   * FF: First Fit
   * BF: Best Fit
   * CP: Choose-Pack
   * PP: Permutation Pack

   CP/PP can also take an additional parameter, W, for the window size.

* METAVP
  Try all VP heuristics at each step of the binary search to maximize the
minimum yield.

* HVP
  Like VP, but heterogeneous-bin aware alternatives.
  additonal parameters:
       R: re-sort bins after item placement
       S: scale bin capacities
       E: equalize dimensions


* METAHVP
  Try all HVP heuristics at each step of the binary search to maximize the
minimum yield.

There may be additional algorithms that are not listed here.