Integrated Scientific Workflow Management for the Emulab Network Testbed Eric Eide Eide , Leigh Eric , Leigh Stoller Stoller, , Tim Stack, Juliana Freire Freire, , Tim Stack, Juliana and Jay Lepreau Lepreau and Jay University of Utah, University of Utah, School of Computing School of Computing USENIX 2006 / June 3, 2006 USENIX 2006 / June 3, 2006
This Talk in One Slide � Current network Current network testbeds testbeds � � …manage the “laboratory” …manage the “laboratory” � � …not the experimentation process. …not the experimentation process. � � → → A big problem for large A big problem for large- -scale activities! scale activities! � � Evolve Emulab for experiments based on Evolve Emulab for experiments based on � scientific workflows scientific workflows � Big mutual benefits: Big mutual benefits: testbed testbed ↔ ↔ workflow workflow � � Work in progress Work in progress � 2
Example: UAV Simulation � A distributed, real A distributed, real- -time time � images → → images application application UAV UAV � Evaluate improvements Evaluate improvements � to real- -time middleware time middleware to real alerts → → alerts Receiver Receiver � vs. CPU load vs. CPU load � � vs. network load vs. network load � � 4 research groups 4 research groups � ATR ATR � x 19 experiments x 19 experiments � ← images images ← � x 56 metrics x 56 metrics � 3
Use Emulab write “ns” file write “ns” file “swap in” swap in” “ Concept Experiment Emulate Concept Experiment Emulate 4
Problems Solved � I get machines! I get machines! � � 328 PCs, and more 328 PCs, and more � � Time Time- - & space & space- -shared shared � � Loads OS and software Loads OS and software � � I get network! I get network! � � Config Config. topology & quality . topology & quality � � I get to collaborate! I get to collaborate! � � Available to researchers Available to researchers � and educators worldwide and educators worldwide � File storage, email, … File storage, email, … � 5
Problems Not Solved � “ “Now what?” Now what?” � � Getting off the ground Getting off the ground � � Run all my software Run all my software � � Add instrumentation Add instrumentation � � Collect all my data Collect all my data � � Analyze it Analyze it � � Scaling up Scaling up � � 19 configurations 19 configurations � � Automation Automation � 6
More Problems Not Solved � “ “How did I get here?” How did I get here?” � � Over the short term… Over the short term… � � “Where are the results “Where are the results � I got last week?” I got last week?” � “How did I get those “How did I get those � results anyway?” results anyway?” � “What if…?” “What if…?” � � …and the long term …and the long term � � Reproducing results Reproducing results � � Reusing artifacts Reusing artifacts � 7
Idea: Scientific Workflow � Managing activities, inputs, and outputs is the Managing activities, inputs, and outputs is the � job of a scientific workflow system scientific workflow system job of a � Our approach: Our approach: evolve Emulab with evolve Emulab with � integrated support for scientific workflows integrated support for scientific workflows � Build on existing abstractions & mechanisms Build on existing abstractions & mechanisms � � Resource focus Resource focus → → user & task focus user & task focus � � Users work “within” and “across” experiments Users work “within” and “across” experiments � 8
Contributions � Address demand + opportunity Address demand + opportunity � � Users need to manage large Users need to manage large- -scale complexity scale complexity � � A symbiotic combination: A symbiotic combination: leverage and impact leverage and impact � � Advance the applicability of Advance the applicability of testbeds testbeds � � Not just Emulab Not just Emulab — — e.g., e.g., PlanetLab PlanetLab and DETER and DETER � � Advance scientific workflow systems Advance scientific workflow systems � � Exploit Exploit testbed testbed capabilities capabilities — — e.g., “total control” e.g., “total control” � � Address Address testbed testbed requirements requirements — — e.g., flexible use e.g., flexible use � 9
Issue: Encapsulation � Current “experiment” model Current “experiment” model � is not fully encapsulating is not fully encapsulating Topology + static events Topology + static events � � Need everything else! Need everything else! � � ns file OSes packages ns file OSes packages � Challenge: specification Challenge: specification � Complete and precise… Complete and precise… � � …w/o huge user burden …w/o huge user burden � � my software inputs outputs my software inputs outputs � Approach: be automatic Approach: be automatic � NFS monitors Subversion repo. E.g., track files used E.g., track files used � � packet monitors datapository (DB) Snapshot, archive, restore Snapshot, archive, restore � � AJAX GUI research filesystems User can refine “extent” User can refine “extent” � � 10
Issue: Definition vs. Execution � Current “experiment” has Current “experiment” has � multiple roles multiple roles Definition Definition � � The thing that you run The thing that you run � � � Challenge: representing Challenge: representing � relationships relationships Multiple runs of one setup Multiple runs of one setup � � Similar configurations Similar configurations � � � Approach: a new model of Approach: a new model of � experimentation experimentation Separate the roles Separate the roles � � Evolve the new abstractions Evolve the new abstractions � � 11
New Model n = 4 n = 2 n = 4 n = 2 � Template Template � � Swapin Swapin � � Experiment Experiment � � Activity Activity � � Record Record � 12
Issue: History � Research and educational Research and educational � plans are dynamic plans are dynamic � By design & by discovery By design & by discovery � � Challenge: safe exploration Challenge: safe exploration � rev 1.1 rev 1.1 � Fork Fork � oops: need new oops: need new � Back up Back up � measurements measurements bigger nets bigger nets � Approach: keep history & Approach: keep history & � what about support temporal navigation what about support temporal navigation loss? loss? � Keep template revisions Keep template revisions � � Track provenance Track provenance � add params params add � Locate, repeat, and reuse Locate, repeat, and reuse � 13
Implementation in Progress Definition Definition Data Analysis Data Analysis Execution Execution & History & History 14
Conclusion � Large and powerful Large and powerful testbeds testbeds � � …enable complex and large …enable complex and large- -scale activities scale activities � � …lead to complex and large …lead to complex and large- -scale workflow scale workflow � management problems management problems � Integrated workflow management can Integrated workflow management can � leverage the strengths of testbeds testbeds leverage the strengths of � Systems approach Systems approach — — and systems challenges and systems challenges � � → → Better Better testbeds testbeds and workflow systems and workflow systems � 15
Thanks! http://www.emulab.net/ Thanks!
17 Extra Slides After This Point
Recommend
More recommend
