Starting Workflow Tasks Before They re Ready

Size: px

Start display at page:

Download "Starting Workflow Tasks Before They re Ready"

Dylan Stewart
6 years ago
Views:

1 Starting Workflow Tasks Before They re Ready Wladislaw Gusew, Björn Scheuermann Computer Engineering Group, Humboldt University of Berlin

2 Agenda Introduction Execution semantics Methods and tools Simulation results Experimental results Conclusion 1 / 21

3 Big data in research 2 / 21

4 Scientific workflow example Directed Acyclic Graph (DAG) Executed on distributed systems Aggregation and broadcast types of tasks Demanding for network resources 3 / 21

5 Execution semantics 4 / 21

6 Execution semantics 4 / 21

7 Execution semantics But in reality resources are limited Execute only a subset of parent tasks concurrently (insufficient number of workers) Congestion of network (all parent tasks have the same priority) 4 / 21

8 Example execution 5 / 21

9 Example execution 5 / 21

10 Example execution 5 / 21

11 Example execution Network congestion can slow down processing even further (effects of data losses at the transport protocol layer) High delay to the start of the aggregation task Low performance and high execution costs (e.g., in computation clouds) 5 / 21

12 What can we do to improve this? 6 / 21

13 What can we do to improve this? 6 / 21

14 What can we do to improve this? 6 / 21

15 What can we do to improve this? 6 / 21

16 What can we do to improve this? 6 / 21

17 What can we do to improve this? 6 / 21

18 What can we do to improve this? List of actions: 1. Obtain information on task s input characteristics 2. Refine the workflow and inform the execution engine 3. Let the aggregation task feel comfortable in changed setting 6 / 21

19 What can we do to improve this? List of actions: 1. Obtain information on task s input characteristics 2. Refine the workflow and inform the execution engine 3. Let the aggregation task feel comfortable in changed setting 6 / 21

20 Obtaining input characteristics 1. Annotations to workflows 2. Manual code review 3. Automated profiling 7 / 21

21 Automated profiling Operating system instrumentation tool Enables interception of system calls (file open, read/write, file close) Record and evaluate logfiles with traces of conducted file accesses. 8 / 21

22 Automated profiling Operating system instrumentation tool Enables interception of system calls (file open, read/write, file close) Record and evaluate logfiles with traces of conducted file accesses. Read accesses [MB] Reads by madd in a small workflow Execution progress [10 8 CPU cycles] Read accesses [MB] Reads by madd in a medium sized workflow Execution progress [10 8 CPU cycles] 8 / 21

23 Refining workflow by transforming DAG 9 / 21

24 Refining workflow by transforming DAG 9 / 21

25 Refining workflow by transforming DAG 9 / 21

26 Refining workflow by transforming DAG 9 / 21

27 Realizing virtual task split Real task is transparently wrapped FUSE enables the setup of a virtual File system in USEr space Access to input files is performed through our wrapper Wrapper is responsible for maintaining the correct execution logic 10 / 21

28 Evaluation with the Montage workflow 11 / 21

Simulating workflow execution Java-based simulation framework for scientific workflows Simulates an execution on a Pegasus/HTCondor stack Use provided Montage workflows with 25, 50, 100, 1000 tasks

29 Simulating workflow execution Java-based simulation framework for scientific workflows Simulates an execution on a Pegasus/HTCondor stack Use provided Montage workflows with 25, 50, 100, 1000 tasks Python script conducted DAG transformation of DAX files Network configured as bottleneck (by bandwidth limitation) W. Chen and E. Deelman, WorkflowSim: A toolkit for simulating scientific workflows in distributed environments, in escience / 21

30 Simulation results 13 / 21

31 Simulation results 13 / 21

32 Variation of number of tasks Simulation results for 50 workers and max-min Total workflow runtime (log.) [s] Normal Split 31% 25% 15% 19% Number of tasks 14 / 21

33 Variation of workers 15 / 21

34 Variation of workers Total workflow runtime [s] Simulation results for Montage 100 and min-min Normal Split 10% 14% 26% 25% Number of workers 16 / 21

35 Variation of scheduling algorithms 17 / 21

36 Variation of scheduling algorithms Simulation results for Montage 100 on 100 workers Total workflow runtime [s] Normal Split 25% 27% 28% 25% HEFT Round-robin Max-min Min-min 17% 34% Random DHEFT Scheduling algorithm 18 / 21

37 Evaluation in a computing cluster Small cluster of up to 10 compute nodes Intel i7 CPU@ 2.5GHz, 8GB RAM, connected to common network switch with 1Gbit/s Execute Montage 133 workflow in Pegasus/HTCondor Network bandwidth was limited on application layer to 10Mbit/s 10 repetitions, mean values with 95% confidence intervals 19 / 21

38 Measurement results Total workflow runtime [s] Computing cluster results for workers Original Montage 133 Transformed Montage Number of computing nodes 20 / 21

39 Conclusion Many legacy workflows exist which are executed with classic semantics Our approach is applicable to aggregation tasks that are often the most time intensive tasks in a workflow By using DAG transformation, no changes to task implementations and execution engines are required 21 / 21

40 Conclusion Many legacy workflows exist which are executed with classic semantics Our approach is applicable to aggregation tasks that are often the most time intensive tasks in a workflow By using DAG transformation, no changes to task implementations and execution engines are required Simulation and real experiment show that performance can be improved by up to 15% Potential of outperforming the original workflow grows with increasing #workers and #tasks 21 / 21

DynamicCloudSim: Simulating Heterogeneity in Computational Clouds

DynamicCloudSim: Simulating Heterogeneity in Computational Clouds Marc Bux, Ulf Leser Department of Computer Science, Humboldt-Universität zu Berlin, Unter den Linden 6, 10099 Berlin, Germany {buxmarcn,leser}@informatik.hu-berlin.de