Skip to content

AD1024/Sager

Repository files navigation

Sager

A Demonic Graph Synthesizer for Worst Case Performance. For more information, plase see our report, slides, and demo.sh

Project Structure

  • sager.rkt contains the pipeline procedure from synthesizing to scaling and the concrete calls we make to generate data.
  • core.rkt contains procedures use incremental solving to synthesize graphs
  • scaler.rkt contains utilities for scaling a gadget to a larger graph
  • graphgen.rkt contains graph representation and symmetry breaking techniques
  • helper.rkt contains some helper function and auxiliary symbolic data structure
  • algorithms/ contains concrete implementations of target algorithms (SPFAs)

Comparison Data Generators

Dependency:

python3 -m pip install cyaron
  • shortest_path_tree_hack.py generates data that utilize Shortest Path Tree based graphs
  • random_graph.py generates graphs based on randomization; it guarantees that all nodes are connected and reachable from each other
  • dag.py generates the graph that utilize the property of a DAG (linked list approach)
  • cyaron_hack.py uses cyaron to generates graph that can hack SPFA based on empirical experiments.

Synthesize a Graph

  • Configure the gadget size and the final size of the graph, the prefix of name of data file to store.

Example:

;;; Sample Calls
;;; Generate 4 files in data directory: 
;;; - test-sager-demo-spfa-vis-10000-A
;;; - test-sager-demo-spfa-vis-10000-B
;;; - test-sager-demo-spfa-vis-100000-A
;;; - test-sager-demo-spfa-vis-100000-B
;;; where A / B stands for the Connect by Entry Nodes and Connect by Exit Nodes
;;; the target algorithm to hack is SPFA with out heuristics
;;; the scaled graph contain 10000 nodes and 100000 nodes
(let ([prefix (make-string-prefix "data/test-sager-demo-spfa-vis-")])
  (sager 
        (map prefix (list "10000" "100000"))   ;; Name of files
        spfa-vis                               ;; Target Algorithm
        (list '() '())                         ;; Manually constructed Gadgets (with human insights)
        (list 10000 100000)                    ;; Size of scaled graphs 
        4 30))                                 ;; Size of Synthesized Gadget and Searching Bound

See more example in sager.rkt

  • run racket sager.rkt

Run Tests for Algorithms

Compile Programs

Place the cpp implementation in ./algorithms the run

make <filename>

Run Tests

make test pattern=<pattern> program=<program>

The pattern is a regular expression. For all files whose name can be matched by the regex, the script will feed them to the selected program once a time.

Example:

make test pattern="*sager*spfa-vis*" program="spfa-vis"

will run spfa-vis with all the pre-generated data that target at SPFA with no heuristics.

Implemented Algorithms

  • SPFA
  • SPFA+SLF
  • SPFA+LLL
  • SPFA+SLF+LLL
  • Dijkstra