Most workflow scheduling algorithms are restricted to the domain of single workflow. There are few researches for scheduling online workflows. In the thesis, we propose OWM approach for scheduling online workflows in a Grid environment. Our experiments show that OWM outperforms RANK_HYBD [21] and Fairness_Dynamic for average makesapn, average SLR and win (%) in different
experimental workloads.
However, RANK_HYBD and Fairness_Dynamic do not work with workflows composed of data-parallel tasks. There are few studies focused on workflow scheduling for data-parallel tasks. The thesis takes this issue into account. We incorporate well-known approaches, e.g., first fit, easy backfilling [22] and conservative backfilling [22] into OWM to deal with workflows composed of data-parallel tasks. The experiments show that OWM(FCFS) is almost equal to OWM(conservative), and outperforms OWM(easy) and OWM(first fit) for average SLR, and outperforms the other three approaches for win (%).
In the future, we will investigate the trade-off between the QoS (Quality of Service) and the performance, i.e., the relation between the fairness and average makespan for workflows. In addition, we will implement OWM to real Grid environments to validate our simulation results.
58
Appendix
The remainder experimental results for workflows composed of data-parallel tasks are shown below. OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
0 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
Figure A-2 Results of different computation intensity for average SLR with (uniform, max, uniform)
Figure A-1 Results of different computation intensity for average makespan with (uniform, max, uniform)
59 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
0 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
Figure A-3 Results of different computation intensity for win (%) with (uniform, max, uniform)
Figure A-4 Results of different computation intensity for average makespan with (uniform, max, exponential)
60 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
0 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
Figure A-5 Results of different computation intensity for average SLR with (uniform, max, exponential)
Figure A-6 Results of different computation intensity for win (%) with (uniform, max, exponential)
61 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
0 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
Figure A-8 Results of different computation intensity for average SLR with (uniform, max, normal)
Figure A-7 Results of different computation intensity for average makespan with (uniform, max, normal)
62 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
0 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
Figure A-9 Results of different computation intensity for win (%) with (uniform, max, normal)
Figure A-10 Results of different computation intensity for average makespan with (uniform, half, uniform)
63 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
0 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
Figure A-11 Results of different computation intensity for average SLR with (uniform, half, uniform)
Figure A-12 Results of different computation intensity for win (%) with (uniform, half, uniform)
64 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
0 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
Figure A-14 Results of different computation intensity for average SLR with (uniform, half, exponential)
Figure A-13 Results of different computation intensity for average makespan with (uniform, half, exponential)
65 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
0 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
Figure A-15 Results of different computation intensity for win (%) with (uniform, half, exponential)
Figure A-16 Results of different computation intensity for average makespan with (uniform, half, normal)
66 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
0 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
Figure A-17 Results of different computation intensity for average SLR with (uniform, half, normal)
Figure A-18 Results of different computation intensity for win (%) with (uniform, half, normal)
67 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
0 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
Figure A-20 Results of different computation intensity for average SLR with (uniform, min, exponential)
Figure A-19 Results of different computation intensity for average makespan with (uniform, min, exponential)
68 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
0 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
Figure A-21 Results of different computation intensity for win (%) with (uniform, min, exponential)
Figure A-22 Results of different computation intensity for average makespan with (uniform, min, normal)
69 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
0 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
Figure A-23 Results of different computation intensity for average SLR with (uniform, min, normal)
Figure A-24 Results of different computation intensity for win (%) with (uniform, min, normal)
70 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
0 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
Figure A-26 Results of different computation intensity for average SLR with (exponential, max, uniform)
Figure A-25 Results of different computation intensity for average makespan with (exponential, max, uniform)
71 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
0 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
Figure A-27 Results of different computation intensity for win (%) with (exponential, max, uniform)
Figure A-28 Results of different computation intensity for average makespan with (exponential, max, exponential)
72 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
0 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
Figure A-29 Results of different computation intensity for average SLR with (exponential, max, exponential)
Figure A-30 Results of different computation intensity for win (%) with (exponential, max, exponential)
73 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
0 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
Figure A-32 Results of different computation intensity for average SLR with (exponential, max, normal)
Figure A-31 Results of different computation intensity for average makespan with (exponential, max, normal)
74 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
0 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
Figure A-33 Results of different computation intensity for win (%) with (exponential, max, normal)
Figure A-34 Results of different computation intensity for average makespan with (exponential, half, uniform)
75 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
0 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
Figure A-35 Results of different computation intensity for average SLR with (exponential, half, uniform)
Figure A-36 Results of different computation intensity for win (%) with (exponential, half, uniform)
76 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
0 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
Figure A-38 Results of different computation intensity for average SLR with (exponential, half, exponential)
Figure A-37 Results of different computation intensity for average makespan with (exponential, half, exponential)
77 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
0 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
Figure A-39 Results of different computation intensity for win (%) with (exponential, half, exponential)
Figure A-40 Results of different computation intensity for average makespan with (exponential, half, normal)
78 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
0 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
Figure A-41 Results of different computation intensity for average SLR with (exponential, half, normal)
Figure A-42 Results of different computation intensity for win (%) with (exponential, half, normal)
79 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
0 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
Figure A-44 Results of different computation intensity for average SLR with (exponential, min, uniform)
Figure A-43 Results of different computation intensity for average makespan with (exponential, min, uniform)
80 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
0 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
Figure A-45 Results of different computation intensity for win (%) with (exponential, min, uniform)
Figure A-46 Results of different computation intensity for average makespan with (exponential, min, exponential)
81 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
0 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
Figure A-47 Results of different computation intensity for average SLR with (exponential, min, exponential)
Figure A-48 Results of different computation intensity for win (%) with (exponential, min, exponential)
82 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
0 OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
Figure A-50 Results of different computation intensity for average SLR with (exponential, min, normal)
Figure A-49 Results of different computation intensity for average makespan with (exponential, min, normal)
83 0
5 10 15 20 25 30 35 40
general computation communication
win (%)
computation intensity
Wi_DisType=exponential, maxTaskNP=min, NP_DisType=normal OWM (FCFS) OWM (conservative) OWM (easy) OWM (first fit)
Figure A-51 Results of different computation intensity for win (%) with (exponential, min, normal)
84 Scheduling and Rescheduling Methods in the GrADS Project”, in Proc. of the 18th International Parallel and Distributed Processing Symposium (IPDPS'04), pp.199--206, Santa Fe, New Mexico USA, April 2004.
[2] T.D. Braun, H.J. Siegel, N. Beck, L.L. Bölöni, M. Maheswaran, A.I. Reuther, J.P. Robertson, M.D. Theys, B. Yao, D. Hensgen, R.F. Freund, “A Comparison of Eleven Static Heuristics for Mapping a Class of Independent Tasks onto Heterogeneous Distributed Computing Systems”, Journal of Parallel and Distributed Computing, Volume 61, Number 6, June 2001 , pp. 810-837(28).
[3] M. Wu and D. Gajski. “ Hypertool: A Programming Aid for Message Passing Systems”. IEEE Transactions on Parallel and Distributed Systems, vol. 1, pp.
330-343, July 1990.
[4] Y. Kwok and I. Ahmad. “Dynamic Critical-Path Scheduling: An Effective Technique for Allocation Task Graphs to Multi-processors”. IEEE Transactions on Parallel and Distributed Systems, vol. 7, no. 5, pp. 506-521, May 1996.
[5] G.C. Sih and E.A. Lee. “ A Compile-Time Scheduling Heuristic for Interconnection-Constrained Heterogeneous Processor Architectures”. IEEE Transactions on Parallel and Distributed Systems, vol. 4, no. 2, pp. 175-186, Feb 1993.
[6] H. EI-Rewini and T.G. Lewis. “Scheduling Parallel Program Tasks onto Arbitrary Target Machines”. J. Parallel and Distributed Computing, vol. 9, pp.
138-153, 1990.
[7] H. Topcuoglu, S. Hariri, and M.-Y. Wu. “Performance-Effective and Low-Complexity Task Scheduling for Heterogeneous Computing”. IEEE Transactions on Parallel and Distributed Systems, 2(13):260-247, 2002.
[8] T. Yang and A. Gerasoulis. “DSC: Scheduling Parallel Tasks on an Unbounded Number of Processors”. IEEE Transactions on Parallel and Distributed Systems, vol. 5, no. 9, pp. 951-967, Sept. 1994.
[9] G. Park, B. Shirazi, and J. Marquis. “DFRN: A New Approach for Duplication Based Scheduling for Distributed Memory Multi-processor Systems”. Proc.
85
Int'l Conf. Parallel Processing, pp. 157-166, 1997.
[10] E. Deelman et al. “Pegasus: Mapping Scientific Workflows onto the Grid”. In European Across Grids Conference, pp. 11-20, 2004.
[11] R. Sakellariou and H. Zhao. “A Low-Cost Rescheduling Policy for Efficient Mapping of Workflows on Grid Systems”. Scientific Programming, 12(4), pp.
253-262, December 2004.
[12] M. Resende and C. Ribeiro. “Greedy Randomized Adaptive Search Procedures, State-of-the-art Handbook in MetaHeuristics”. Glover and Kochenberger, eds., Kluwer Academic Publishers, 2002.
[13] H. Singh and A. Youssef. “Mapping and Scheduling Heterogeneous Task Graphs Using Genetic Algorithms”. Proc. Heterogeneous Computing Workshop, pp. 86-97, 1996.
[14] A. YarKhan and J.J. Dongarra. “Experiments with Scheduling Using Simulated Annealing in a Grid Environment”. In Grid 2002, November 2002.
[15] Y. Kwok and I. Ahmad, “Benchmarking and comparison of the task graph scheduling algorithms”, Journal of Parallel and Distributed Computing, 59(3):381–422, 1999
[16] J. Blythe, S. Jain, E. Deelman, Y. Gil, K. Vahi, A. Mandal, K. Kennedy, “Task Scheduling Strategies for Workflow-based Applications in Grid”. Cluster Computing and the Grid, 2005. CCGrid 2005. IEEE International Symposium, pp. 759-767 Vol. 2, 9-12 May 2005.
[17] J. Yu, R, Buyya, “Workflow Scheduling Algorithms for Grid Computing”.
Technical Report, Grid-TR-2007-10.
[18] R. Sakellariou and H. Zhao. “A hybrid heuristic for DAG scheduling on heterogeneous systems”.In 18th International Parallel and Distributed Processing Symposium (IPDPS’04), page 111. IEEE Computer Society, 2004.
[19] H. Casanova, F. Desprez, and F. Suter. “From Heterogeneous Task Scheduling to Heterogeneous Mixed Parallel Scheduling”. In 10th Int. Euro-Par Conference, volume 3149 of LNCS, pages 230-237, Aug. 2004.
[20] H. Zhao and R. Sakellarious. “Scheduling Multiple DAGs onto Heterogeneous Systems”. In Proceedings of the 15th Heterogeneous Computing Workshop (HCW), Rhodes Island, Greece, April 2006.
[21] Z. Yu and W. Shi. “A Planner-Guided Scheduling Strategy for Multiple Workflow Applications”. On Parallel Processing Workshops, ICPP-W 08, 8-12 Sept. 2008.
[22] A.W. Mu'alem and D.G. Feitelson. “Utilization, Predictability, Workloads, and User Runtime Estimates in Scheduling the IBM SP2 with Backfilling”. IEEE Transactions on Parallel and Distributed Systems, vol. 12, no. 6, June 2001.
86
[23] T. N'takpe' and F. Suter. “Concurrent Scheduling of Parallel Task Graphs on Multi-Clusters Using Constrained Resource Allocations”. Rapport de recherché n° 6774, December 2008.
[24] Y. Zhang, C. Koelbel, and K. Kennedy. “Relative Performance of Scheduling Algorithms in Grid Environments”. In 7th IEEE International Symposium on Cluster Computing and the Grid (CCGrid 2007).
[25] A. Mandal, K. Kennedy, C. Koelbel, G. Marin, J. Mellor-Crummey, B. Liu, and L. Johnsson. “Scheduling Strategies for Mapping Application workflows onto the Gird”. In 14th IEEE Symposium on High Performance Distributed Computing (HPDC 14), pp. 125-134, 2005.
[26] F. Dong and S.G. Akl. “Scheduling Algorithms for Grid Computing: State of the Art and Open Problems”. Technical Report No. 2006-504.
[27] E. Deelman, G. Singh, and C. Kesselman. “Optimizing Grid-based Workflow Execution”. Journal of Grid Computing, 3(3): 201-219, 2005.
[28] T. N'takpe' and F. Suter. “A Comparison of Scheduling Approaches for Mixed-Parallel Applications on Heterogeneous Platforms”. In 6th International Symposium on Parallel and Distributed Computing (IS-PDC), pp. 250-257, Hagenberg, Austria, July 2007.
[29] G. Singh, E. Deelman, and G. Bruce Berriman et al. “Montage: a Grid Enabled Image Mosaic Service for the National Virtual Observatory”.
Astronomical Data Analysis Software and Systems (ADASS), (13), 2003.
[30] S. Ludtke, P. Baldwin, and W. Chiu. “EMAN: Semiautomated Software for High Resolution Single-Particle Reconstructions”. J. Struct. Biol, (128): 82-97, 1999.
[31] M.R. Gary and D.S. Johnson. “Computers and Intractability: A Guide to the Theory of NP-Completeness”. W.H. Freeman and Co., 1979.
[32] J.D. Ullman. “NP-Complete Scheduling Problems”. J. Computer and Systems Sciences, vol. 10, pp. 384-393, 1975.
[33] Miguel L. Bote-Lorenzo, Yannis A. Dimitraidis, and Eduardo Gomez-Sanchez.
“Grid Characteristics and uses: a Grid Definition”.