Engineering a Scalable High Quality Graph...

31
1 April 29, 2010 Manuel Holtgrewe, Peter Sanders, Christian Schulz – Institute for Theoretical Computer Science, Algorithmics II Scalable High Quality Graph Partitioner KIT – die Kooperation von Forschungszentrum Karlsruhe GmbH und Universit¨ at Karlsruhe (TH) Institute for Theoretical Computer Science, Algorithmics II Engineering a Scalable High Quality Graph Partitioner Manuel Holtgrewe, Peter Sanders, Christian Schulz KIT – die Kooperation von Forschungszentrum Karlsruhe GmbH und Universit¨ at Karlsruhe (TH)

Transcript of Engineering a Scalable High Quality Graph...

Page 1: Engineering a Scalable High Quality Graph Partitioneralgo2.iti.kit.edu/documents/talks_cs/ipdps_atlanta.pdf · Scalable High Quality Graph Partitioner KIT ... Successful in existing

1 April 29, 2010 Manuel Holtgrewe, Peter Sanders, Christian Schulz – Institute for Theoretical Computer Science, Algorithmics IIScalable High Quality Graph Partitioner

KIT – die Kooperation von Forschungszentrum Karlsruhe GmbH und Universitat Karlsruhe (TH)

Institute for Theoretical Computer Science, Algorithmics II

Engineering a Scalable High QualityGraph PartitionerManuel Holtgrewe, Peter Sanders,Christian Schulz

KIT – die Kooperation von Forschungszentrum Karlsruhe GmbH und Universitat Karlsruhe (TH)

Page 2: Engineering a Scalable High Quality Graph Partitioneralgo2.iti.kit.edu/documents/talks_cs/ipdps_atlanta.pdf · Scalable High Quality Graph Partitioner KIT ... Successful in existing

Balanced Graph Partitioning

2 April 29, 2010 Manuel Holtgrewe, Peter Sanders, Christian Schulz – Institute for Theoretical Computer Science, Algorithmics IIScalable High Quality Graph Partitioner

KIT – die Kooperation von Forschungszentrum Karlsruhe GmbH und Universitat Karlsruhe (TH)

Partition graph G = (V , E) into k disjoint blocks s.t.

total node weight of each block ≤ 1 + ε

ktotal node weight

total weight of cut edges as small as possible

Applications:finite element simulations, VLSI design, route planning, . . .

Page 3: Engineering a Scalable High Quality Graph Partitioneralgo2.iti.kit.edu/documents/talks_cs/ipdps_atlanta.pdf · Scalable High Quality Graph Partitioner KIT ... Successful in existing

Multi-Level Graph Partitioning

3 April 29, 2010 Manuel Holtgrewe, Peter Sanders, Christian Schulz – Institute for Theoretical Computer Science, Algorithmics IIScalable High Quality Graph Partitioner

KIT – die Kooperation von Forschungszentrum Karlsruhe GmbH und Universitat Karlsruhe (TH)

Successful in existing systems:DiBaP, Chaco, Jostle, Metis, Scotch,. . .

Page 4: Engineering a Scalable High Quality Graph Partitioneralgo2.iti.kit.edu/documents/talks_cs/ipdps_atlanta.pdf · Scalable High Quality Graph Partitioner KIT ... Successful in existing

Why Yet Another MGP Algorithm?Karlsruhe Parallel Partitioner

4 April 29, 2010 Manuel Holtgrewe, Peter Sanders, Christian Schulz – Institute for Theoretical Computer Science, Algorithmics IIScalable High Quality Graph Partitioner

KIT – die Kooperation von Forschungszentrum Karlsruhe GmbH und Universitat Karlsruhe (TH)

scalable parallel algorithm for the case k =# of processorslarge inputshigh qualityunderstand / engineer multi level methodbridge gaps theory↔ practice

Page 5: Engineering a Scalable High Quality Graph Partitioneralgo2.iti.kit.edu/documents/talks_cs/ipdps_atlanta.pdf · Scalable High Quality Graph Partitioner KIT ... Successful in existing

Matching SelectionGoals:

5 April 29, 2010 Manuel Holtgrewe, Peter Sanders, Christian Schulz – Institute for Theoretical Computer Science, Algorithmics IIScalable High Quality Graph Partitioner

KIT – die Kooperation von Forschungszentrum Karlsruhe GmbH und Universitat Karlsruhe (TH)

1. large edge weights sparsify2. large #edges few levels3. uniform node weights “represent” input4. small node degrees “represent” input

unclear objective gap to approx. weighted matchingwhich only considers 1.,2.

Our Solution:Apply approx. weighted matching to general edge rating function

input graph

...

contract

match

Page 6: Engineering a Scalable High Quality Graph Partitioneralgo2.iti.kit.edu/documents/talks_cs/ipdps_atlanta.pdf · Scalable High Quality Graph Partitioner KIT ... Successful in existing

Edge Ratings Tried

6 April 29, 2010 Manuel Holtgrewe, Peter Sanders, Christian Schulz – Institute for Theoretical Computer Science, Algorithmics IIScalable High Quality Graph Partitioner

KIT – die Kooperation von Forschungszentrum Karlsruhe GmbH und Universitat Karlsruhe (TH)

ω({u, v})

expansion({u, v}) :=ω({u, v})

c(u) + c(v)

expansion∗({u, v}) :=ω({u, v})c(u)c(v)

expansion∗2({u, v}) :=ω({u, v})2

c(u)c(v)

innerOuter({u, v}) :=ω({u, v})

Out(v) + Out(u)− 2ω(u, v)where c = node weight, ω =edge weight,Out(u) := ∑{u,v}∈E ω({u, v})

Page 7: Engineering a Scalable High Quality Graph Partitioneralgo2.iti.kit.edu/documents/talks_cs/ipdps_atlanta.pdf · Scalable High Quality Graph Partitioner KIT ... Successful in existing

Approx. Weighted Matching

7 April 29, 2010 Manuel Holtgrewe, Peter Sanders, Christian Schulz – Institute for Theoretical Computer Science, Algorithmics IIScalable High Quality Graph Partitioner

KIT – die Kooperation von Forschungszentrum Karlsruhe GmbH und Universitat Karlsruhe (TH)

Use parallel prepartitioner (optional)currently geometric recursive bipartitioning when coords.availableGlobal Path Algorithm [MaueSanders 2007] locallyparallel [ManneBisseling 2007] for border nodes

Page 8: Engineering a Scalable High Quality Graph Partitioneralgo2.iti.kit.edu/documents/talks_cs/ipdps_atlanta.pdf · Scalable High Quality Graph Partitioner KIT ... Successful in existing

Initial Partitioning

8 April 29, 2010 Manuel Holtgrewe, Peter Sanders, Christian Schulz – Institute for Theoretical Computer Science, Algorithmics IIScalable High Quality Graph Partitioner

KIT – die Kooperation von Forschungszentrum Karlsruhe GmbH und Universitat Karlsruhe (TH)

Every PE can perform initial partitioning using different seedsCompared PARTY, PMETIS, and SCOTCH

SCOTCH yielded best results

match

... ...local improvement

uncontractcontract

outputpartition

input graph

partitioning

initial

Page 9: Engineering a Scalable High Quality Graph Partitioneralgo2.iti.kit.edu/documents/talks_cs/ipdps_atlanta.pdf · Scalable High Quality Graph Partitioner KIT ... Successful in existing

Local Improvement – Basic Approach

9 April 29, 2010 Manuel Holtgrewe, Peter Sanders, Christian Schulz – Institute for Theoretical Computer Science, Algorithmics IIScalable High Quality Graph Partitioner

KIT – die Kooperation von Forschungszentrum Karlsruhe GmbH und Universitat Karlsruhe (TH)

Use linear time local search [Fiduccia Mattheyses 82]on multiple pairs of blocks in parallel

input graph

...

initial

...

outputpartition

local improvement

partitioning

match

contract uncontract

Page 10: Engineering a Scalable High Quality Graph Partitioneralgo2.iti.kit.edu/documents/talks_cs/ipdps_atlanta.pdf · Scalable High Quality Graph Partitioner KIT ... Successful in existing

Finding Pairs of Blocks

10 April 29, 2010 Manuel Holtgrewe, Peter Sanders, Christian Schulz – Institute for Theoretical Computer Science, Algorithmics IIScalable High Quality Graph Partitioner

KIT – die Kooperation von Forschungszentrum Karlsruhe GmbH und Universitat Karlsruhe (TH)

Color edges of quotient graph Qseveral parallel algorithms triedEach color→ matching in Q→ independent block pairs

Page 11: Engineering a Scalable High Quality Graph Partitioneralgo2.iti.kit.edu/documents/talks_cs/ipdps_atlanta.pdf · Scalable High Quality Graph Partitioner KIT ... Successful in existing

Pairwise Local Search

11 April 29, 2010 Manuel Holtgrewe, Peter Sanders, Christian Schulz – Institute for Theoretical Computer Science, Algorithmics IIScalable High Quality Graph Partitioner

KIT – die Kooperation von Forschungszentrum Karlsruhe GmbH und Universitat Karlsruhe (TH)

two local searches

adopt best

exchange boundaries

New queue selection strategy topGain

Page 12: Engineering a Scalable High Quality Graph Partitioneralgo2.iti.kit.edu/documents/talks_cs/ipdps_atlanta.pdf · Scalable High Quality Graph Partitioner KIT ... Successful in existing

Results – Example

12 April 29, 2010 Manuel Holtgrewe, Peter Sanders, Christian Schulz – Institute for Theoretical Computer Science, Algorithmics IIScalable High Quality Graph Partitioner

KIT – die Kooperation von Forschungszentrum Karlsruhe GmbH und Universitat Karlsruhe (TH)

Streetnetwork Europe |V | = 18M, |E | = 44M, k = 16,KaPPa↔ParMetis

Page 13: Engineering a Scalable High Quality Graph Partitioneralgo2.iti.kit.edu/documents/talks_cs/ipdps_atlanta.pdf · Scalable High Quality Graph Partitioner KIT ... Successful in existing

Scalablity

13 April 29, 2010 Manuel Holtgrewe, Peter Sanders, Christian Schulz – Institute for Theoretical Computer Science, Algorithmics IIScalable High Quality Graph Partitioner

KIT – die Kooperation von Forschungszentrum Karlsruhe GmbH und Universitat Karlsruhe (TH)

k

tota

l ti

me

[s]

510

2050

100

200

500

4 8 16 32 64 128 256 512 1024

510

2050

100

200

500

4 8 16 32 64 128 256 512 1024

510

2050

100

200

500

4 8 16 32 64 128 256 512 1024

510

2050

100

200

500

4 8 16 32 64 128 256 512 1024

510

2050

100

200

500

4 8 16 32 64 128 256 512 1024

510

2050

100

200

500

4 8 16 32 64 128 256 512 1024

KaPPa−StrongKaPPa−FastKaPPa−Minimalscotchkmetisparmetis

Streetnetwork Europe (|V | = 18M, |E | = 44M)

Page 14: Engineering a Scalable High Quality Graph Partitioneralgo2.iti.kit.edu/documents/talks_cs/ipdps_atlanta.pdf · Scalable High Quality Graph Partitioner KIT ... Successful in existing

QualityComparison with Other Systems

14 April 29, 2010 Manuel Holtgrewe, Peter Sanders, Christian Schulz – Institute for Theoretical Computer Science, Algorithmics IIScalable High Quality Graph Partitioner

KIT – die Kooperation von Forschungszentrum Karlsruhe GmbH und Universitat Karlsruhe (TH)

Geometric mean, imbalance ε = 0.03:14 graphs (78 136–18M nodes) ×k ∈ {2, 4, 8, 16, 64}

Variant avg. cut balance avg. time [s]KaPPa-Strong 24 227 1.028 36.93KaPPa-Fast +2% 1.028 21.40KaPPa-Minimal +3% 1.028 5.94seq. scotch +7% 1.027 5.95kmetis +18% 1.026 0.79parmetis +30% 1.041 0.59

Walshaw instances, road networks, Florida Sparse MatrixCollection, random Delaunay triangulations, random geometricgraphs, social networks.

Page 15: Engineering a Scalable High Quality Graph Partitioneralgo2.iti.kit.edu/documents/talks_cs/ipdps_atlanta.pdf · Scalable High Quality Graph Partitioner KIT ... Successful in existing

Sources of Quality Improvement

15 April 29, 2010 Manuel Holtgrewe, Peter Sanders, Christian Schulz – Institute for Theoretical Computer Science, Algorithmics IIScalable High Quality Graph Partitioner

KIT – die Kooperation von Forschungszentrum Karlsruhe GmbH und Universitat Karlsruhe (TH)

1. edge ratings largest effect by far2. more localized refinement.

Surprise! previous parallelizations caused worse quality3. better queue selection4. better matching algorithms5. more time invested

Page 16: Engineering a Scalable High Quality Graph Partitioneralgo2.iti.kit.edu/documents/talks_cs/ipdps_atlanta.pdf · Scalable High Quality Graph Partitioner KIT ... Successful in existing

Quality – Walshaw BenchmarkZero percent case

16 April 29, 2010 Manuel Holtgrewe, Peter Sanders, Christian Schulz – Institute for Theoretical Computer Science, Algorithmics IIScalable High Quality Graph Partitioner

KIT – die Kooperation von Forschungszentrum Karlsruhe GmbH und Universitat Karlsruhe (TH)

Graph 2 4 8 16 32 64data - 383 - - - -uk - 43 - - - -add32 - - - - - 493bcsstk29 - 8159 - - - -bcsstk32 - 9492 - - - -fetooth - 7142 11935 18382 - -ferotor - 7480 13292 - - -598a - 8154 - 26427 - -feocean - 1902 4299 - - -144 - 15804 26175 - 58462 -wave - 17475 30511 - - -m14b - 13391 - - - -auto - 28048 - - - -

Page 17: Engineering a Scalable High Quality Graph Partitioneralgo2.iti.kit.edu/documents/talks_cs/ipdps_atlanta.pdf · Scalable High Quality Graph Partitioner KIT ... Successful in existing

Quality – Walshaw BenchmarkOne percent case

17 April 29, 2010 Manuel Holtgrewe, Peter Sanders, Christian Schulz – Institute for Theoretical Computer Science, Algorithmics IIScalable High Quality Graph Partitioner

KIT – die Kooperation von Forschungszentrum Karlsruhe GmbH und Universitat Karlsruhe (TH)

Graph 2 4 8 16 32 64data - 383 - - - -add32 - - - - - 493bcsstk29 - 8159 - - - -memplus - 9559 - - - -fepwt - 704 - - 5612 -bcsstk32 - 9492 - - - -febody - 636 - 1931 - -fetooth - - 11935 - - -ferotor - 7480 - - - -598a - 8154 16467 26427 40674 -feocean - - - 8432 13660 21548wave - 17475 30511 44611 64551 -m14b - 13391 26666 43975 - -auto - - 47650 79847 - -

Page 18: Engineering a Scalable High Quality Graph Partitioneralgo2.iti.kit.edu/documents/talks_cs/ipdps_atlanta.pdf · Scalable High Quality Graph Partitioner KIT ... Successful in existing

Quality – Walshaw BenchmarkThree percent case

18 April 29, 2010 Manuel Holtgrewe, Peter Sanders, Christian Schulz – Institute for Theoretical Computer Science, Algorithmics IIScalable High Quality Graph Partitioner

KIT – die Kooperation von Forschungszentrum Karlsruhe GmbH und Universitat Karlsruhe (TH)

Graph 2 4 8 16 32 64add32 - - - - - 493wingnodal - 3561 - - - -bcsstk30 - 16537 34513 - - -bcsstk31 - 2676 7181 - - -fepwt - 704 1419 - - -bcsstk32 - 8799 21023 36613 - -febody - 601 1054 1800 2947 -brack2 - 2839 - - - -fetooth - 6946 11662 17760 - -ferotor 1963 7222 12852 - - -598a - 7955 - 25966 39829 -feocean - 1698 3974 - - 21033144 - - 25491 - - -wave - 16936 28839 43063 62743 87325m14b - 13136 26057 42783 67326 100286auto - 26379 45525 77611 122902 -

Page 19: Engineering a Scalable High Quality Graph Partitioneralgo2.iti.kit.edu/documents/talks_cs/ipdps_atlanta.pdf · Scalable High Quality Graph Partitioner KIT ... Successful in existing

Quality – Walshaw BenchmarkFive percent case

19 April 29, 2010 Manuel Holtgrewe, Peter Sanders, Christian Schulz – Institute for Theoretical Computer Science, Algorithmics IIScalable High Quality Graph Partitioner

KIT – die Kooperation von Forschungszentrum Karlsruhe GmbH und Universitat Karlsruhe (TH)

Graph 2 4 8 16 32 64add32 - - - - - 493cti - 890 - - - -bcsstk30 - 16332 34350 - - -bcsstk31 2676 7152 - - - -fepwt - 701 1409 2777 5546 -bcsstk32 - 4644 8481 35965 - -febody - 601 1054 1784 2887 4888brack2 - 2739 6781 - - -fetooth - 6864 11662 17603 - -ferotor 1955 7045 12566 20132 - -598a - 7851 15721 25808 39369 58031feocean - 1697 3941 7722 - 20667144 - - 25259 38225 56926 -wave - 16820 28700 42800 62520 86663m14b - 13136 25679 42608 66793 99063auto - 25883 45039 76488 122378 173968

Page 20: Engineering a Scalable High Quality Graph Partitioneralgo2.iti.kit.edu/documents/talks_cs/ipdps_atlanta.pdf · Scalable High Quality Graph Partitioner KIT ... Successful in existing

Conclusion – KaPPa

20 April 29, 2010 Manuel Holtgrewe, Peter Sanders, Christian Schulz – Institute for Theoretical Computer Science, Algorithmics IIScalable High Quality Graph Partitioner

KIT – die Kooperation von Forschungszentrum Karlsruhe GmbH und Universitat Karlsruhe (TH)

Record setting quality for large graphsScalable parallelismConcept of edge ratingeasy to integrate into other systems?Make results for weighted matchings usableParallelization and quality improvement by quotient graph edgecoloringImproved queue selection strategiesMore information: http://arxiv.org/pdf/0910.2004includes some errata

Page 21: Engineering a Scalable High Quality Graph Partitioneralgo2.iti.kit.edu/documents/talks_cs/ipdps_atlanta.pdf · Scalable High Quality Graph Partitioner KIT ... Successful in existing

Future Work

21 April 29, 2010 Manuel Holtgrewe, Peter Sanders, Christian Schulz – Institute for Theoretical Computer Science, Algorithmics IIScalable High Quality Graph Partitioner

KIT – die Kooperation von Forschungszentrum Karlsruhe GmbH und Universitat Karlsruhe (TH)

Getting fasterReconsider every aspect of MGP.Example: n-level GP based on observations from routeplanning. More: edge ratings, stopping rules, initial partitioning,refinement algorithms 6=local search, overall strategy,. . .Parallel external partitioning of Tera-Node-NetworksSoftware releaseGeneralizations (clustering, objective functions, hypergraphs,. . . )

Page 22: Engineering a Scalable High Quality Graph Partitioneralgo2.iti.kit.edu/documents/talks_cs/ipdps_atlanta.pdf · Scalable High Quality Graph Partitioner KIT ... Successful in existing

Recursive Coordinate Bisection

22 April 29, 2010 Manuel Holtgrewe, Peter Sanders, Christian Schulz – Institute for Theoretical Computer Science, Algorithmics IIScalable High Quality Graph Partitioner

KIT – die Kooperation von Forschungszentrum Karlsruhe GmbH und Universitat Karlsruhe (TH)

Similar to Bentley’s kd-treesParallel QUICK-SELECT

Could be improved using Parallel Randomized Selectionalgorithm by Blume et al.

Page 23: Engineering a Scalable High Quality Graph Partitioneralgo2.iti.kit.edu/documents/talks_cs/ipdps_atlanta.pdf · Scalable High Quality Graph Partitioner KIT ... Successful in existing

Uncontract and Local Improvement

23 April 29, 2010 Manuel Holtgrewe, Peter Sanders, Christian Schulz – Institute for Theoretical Computer Science, Algorithmics IIScalable High Quality Graph Partitioner

KIT – die Kooperation von Forschungszentrum Karlsruhe GmbH und Universitat Karlsruhe (TH)

BFS to reduce communication volumeQueue Selection Strategies of local searchimproved quality1. use queue promising larger gain (TopGain)2. use queue of heavier block (MaxLoad) if one block is

overloaded

Page 24: Engineering a Scalable High Quality Graph Partitioneralgo2.iti.kit.edu/documents/talks_cs/ipdps_atlanta.pdf · Scalable High Quality Graph Partitioner KIT ... Successful in existing

References I

24 April 29, 2010 Manuel Holtgrewe, Peter Sanders, Christian Schulz – Institute for Theoretical Computer Science, Algorithmics IIScalable High Quality Graph Partitioner

KIT – die Kooperation von Forschungszentrum Karlsruhe GmbH und Universitat Karlsruhe (TH)

B. Hendrickson and R. W. LelandA Multi-Level Algorithm For Partitioning Graphsin Supercomputing 1995

G. Karypis and V. KumarMultilevel Graph Partitioning SchemesICCP (3) 1995, p. 113-122

Page 25: Engineering a Scalable High Quality Graph Partitioneralgo2.iti.kit.edu/documents/talks_cs/ipdps_atlanta.pdf · Scalable High Quality Graph Partitioner KIT ... Successful in existing

References II

25 April 29, 2010 Manuel Holtgrewe, Peter Sanders, Christian Schulz – Institute for Theoretical Computer Science, Algorithmics IIScalable High Quality Graph Partitioner

KIT – die Kooperation von Forschungszentrum Karlsruhe GmbH und Universitat Karlsruhe (TH)

C. Walshaw, M. Cross, M. G. EverettDynamic mesh partitioning: a unified optimisation andload-balancing algorithmTech. Rep. 95/IM/06, Univ. Greenwich, London

M. J. Berger and S. H. BokhariA Partitioning Strategy For Nonuniform Problems OnMultiprocessorsIEEE Trans. Comput., 36(5):570580, 1987

Page 26: Engineering a Scalable High Quality Graph Partitioneralgo2.iti.kit.edu/documents/talks_cs/ipdps_atlanta.pdf · Scalable High Quality Graph Partitioner KIT ... Successful in existing

References III

26 April 29, 2010 Manuel Holtgrewe, Peter Sanders, Christian Schulz – Institute for Theoretical Computer Science, Algorithmics IIScalable High Quality Graph Partitioner

KIT – die Kooperation von Forschungszentrum Karlsruhe GmbH und Universitat Karlsruhe (TH)

H. D. SimonPartitioning of Unstructured Problems for Parallel ProcessingComputing Systems in Engineering

A. Pothen, H. D. Simon, K. P. LiuPartitioning Sparse Matrices with Eigenvectors of Graphs.SIAM J. on Matrix Analysis and Applications, 11(3):430-452,1990.

Page 27: Engineering a Scalable High Quality Graph Partitioneralgo2.iti.kit.edu/documents/talks_cs/ipdps_atlanta.pdf · Scalable High Quality Graph Partitioner KIT ... Successful in existing

References IV

27 April 29, 2010 Manuel Holtgrewe, Peter Sanders, Christian Schulz – Institute for Theoretical Computer Science, Algorithmics IIScalable High Quality Graph Partitioner

KIT – die Kooperation von Forschungszentrum Karlsruhe GmbH und Universitat Karlsruhe (TH)

J. Maue and P. SandersEngineering Algorithms for Approximate Weighted MatchingExperimental Algorithms, p. 242-255

B. Monien, R. PreisQuality Matching and Local Improvement for MultilevelGraph-PartitioningParallel Computing, 26(12):1609-1634, 2000

Page 28: Engineering a Scalable High Quality Graph Partitioneralgo2.iti.kit.edu/documents/talks_cs/ipdps_atlanta.pdf · Scalable High Quality Graph Partitioner KIT ... Successful in existing

References V

28 April 29, 2010 Manuel Holtgrewe, Peter Sanders, Christian Schulz – Institute for Theoretical Computer Science, Algorithmics IIScalable High Quality Graph Partitioner

KIT – die Kooperation von Forschungszentrum Karlsruhe GmbH und Universitat Karlsruhe (TH)

B. W. Kernighan, S. LinAn Efficient Heuristic Procedure for Partitioning GraphsBell Systems Technical Journal 14:291-307

C. SchulzScalable Parallel Refinement of Graph PartitionsDiplomarbeit, Universitt Karlsruhe (TH)

Page 29: Engineering a Scalable High Quality Graph Partitioneralgo2.iti.kit.edu/documents/talks_cs/ipdps_atlanta.pdf · Scalable High Quality Graph Partitioner KIT ... Successful in existing

References VI

29 April 29, 2010 Manuel Holtgrewe, Peter Sanders, Christian Schulz – Institute for Theoretical Computer Science, Algorithmics IIScalable High Quality Graph Partitioner

KIT – die Kooperation von Forschungszentrum Karlsruhe GmbH und Universitat Karlsruhe (TH)

C. Fiduccia, R. MattheysesA Linear-Time Heuristic for Improving Network PartitionsDesign Automation, 1982

M. Halappanavar, F. Dobrian, A. PothenMatchings in Massive Graphs of Terrascale Computers viaApproximationTo appear, 2009

Page 30: Engineering a Scalable High Quality Graph Partitioneralgo2.iti.kit.edu/documents/talks_cs/ipdps_atlanta.pdf · Scalable High Quality Graph Partitioner KIT ... Successful in existing

References VII

30 April 29, 2010 Manuel Holtgrewe, Peter Sanders, Christian Schulz – Institute for Theoretical Computer Science, Algorithmics IIScalable High Quality Graph Partitioner

KIT – die Kooperation von Forschungszentrum Karlsruhe GmbH und Universitat Karlsruhe (TH)

F. Pellegrini, J. RomanScotch: A Software Package for Static Mapping by DualRecursive Bipartitioning of Process and Architecture GraphsHigh-Performance Computing and Networking, p. 493-498

F. Manne and R. H. BisselingA Parallel Approximation Algorithm for the Weighted MaximumMatching Problem7th International Conference on Parallel Processing andApplied Mathematics (PPAM), 2007,

Page 31: Engineering a Scalable High Quality Graph Partitioneralgo2.iti.kit.edu/documents/talks_cs/ipdps_atlanta.pdf · Scalable High Quality Graph Partitioner KIT ... Successful in existing

References VIII

31 April 29, 2010 Manuel Holtgrewe, Peter Sanders, Christian Schulz – Institute for Theoretical Computer Science, Algorithmics IIScalable High Quality Graph Partitioner

KIT – die Kooperation von Forschungszentrum Karlsruhe GmbH und Universitat Karlsruhe (TH)

Fiduccia, C. M. and Mattheyses, R. M.A Linear-Time Heuristic for Improving Network Partitions19th Conference on Design Automation, 1982