View
0
Download
0
Category
Preview:
Citation preview
www. chameleoncloud.org
Kate Keahey
Mathematics and CS Division, Argonne National Laboratory
CASE, University of Chicago
keahey@anl.gov
Im IMPROVINGREPRODUCIBILITYWITHCLOUDSANDNOTEBOOKS
November 8, 2019 GEFI Workshop, Coimbra, Portugal
www. chameleoncloud.org
REPRODUCIBILITYDILEMMA
� Challenges� Actionabledigitalartifacts:configurations,scientificpracticesandprocesses� Contextinwhichtheycanbeshared:instruments,resources,etc.� Publication,discovery,indexing,etc.
� Towardsintentionalupfrontshareableresearch
?
Should I invest in more new research instead?
Should I invest in making my experiments repeatable?
www. chameleoncloud.org
CHAMELEONASASCIENTIFICINSTRUMENT� Weliketochange:testbedthatadaptsitselftoyourexperimentalneeds
� Deepreconfigurability(baremetal)andisolation(CHI)–butalsoeaseofuse(KVM)� CHI:poweron/off,reboot,customkernel,serialconsoleaccess,etc.
� Wewanttobeallthingstoallpeople:balancinglarge-scaleanddiverse� Large-scale:~largehomogenouspartition(~15,000cores),5PBofstoragedistributedover
2sites(now+1!)connectedwith100Gnetwork…� …anddiverse:ARMs,Atoms,FPGAs,GPUs,Corsaswitches,etc.
� Cloud++:leveragingmainstreamcloudtechnologies� PoweredbyOpenStackwithbaremetalreconfiguration(Ironic)+“specialsauce”� ChameleonteamcontributionrecognizedasofficialOpenStackcomponent
� Welivetoserve:open,productiontestbedforComputerScienceResearch� Startedin10/2014,testbedavailablesince07/2015,renewedin10/2017� Currently3,500+users,500+projects,100+institutions
www. chameleoncloud.org
BEYONDTHEINSTRUMENT:ANECOSYSTEMFORREPEATABILTYANDSHARING� Clouds/testbedsgenerateawealthofshareableartifacts
� Images,orchestrationtemplates,tools,etc.� Clouds/testbedsas“players”forcommonartifacts
� Repeatability/replicabilityfeatures� Testbedversioning(>50versionsofthetestbed)� Appliance/digitalartifactversioning� ExperimentPrecis:ahistorycommandanalogue:usestestbedloggingdatato
reconstructasetofactions
� Documentingascientificprocess� Imperative,non-transactional,versioncontrolled,etc.� Orchestrationversusnotebooks
www. chameleoncloud.org
CHAMELEONJUPYTERINTEGRATION� Combiningtheeaseofnotebooksandthepowerofasharedplatform
� StorytellingwithJupyter:ideas/text,process/code,results� Chameleon:sophisticatedexperimentalcontainersinneedof“storytelling”
� JupyterLabserverforourusers
� Gotojupyter.chameleoncloud.organdusewithyourChameleoncredentials
� Chameleon/Jupyterintegration� Python/bashinterfacestothetestbed,
storingandsharing,Chameleoncredentials� Namedcontainers
� Templatesofexistingexperiments
Screencastofacomplexexperiment:https://vimeo.com/297210055“ACaseforIntegratingExperimentalContainerswithNotebooks”,CloudCom2019
www. chameleoncloud.org
SHARING,PUBLISHING,LEVERAGING� Wenowhaveeverythingweneedtoshareexperiments
� Waystoestablishanexperimentalenvironment+player
� Waystodocumentanexperimentalprocess
� Butwait…howdoIactuallysharethem?� Sendmail,Chameleonobjectstore,github…
� PublishingviaZenodo:storeyourexperimentsandmakethemcitableviaDOIs
� Creatingbridges,integration� Import/Exportfrom/toZenodo
� Makingresearchfindable:thesharingplatform
SC19Poster:SharingandReplicabilityofNotebook-BasedResearchonOpenTestbeds
www. chameleoncloud.org
?
Well-documented process
Executable code
Accessible, consistent code environment
Easy to find experiment
?
Notebooks
Open testbeds Sharing
services
?
Well-documented process
Executable code
Accessible, consistent code environment
Easy to find experiment
?
Notebooks
Open testbeds Sharing
services
?
Well-documented process
Executable code
Accessible, consistent code environment
Easy to find experiment
?
Notebooks
Open testbeds Sharing
services
?
Well-documented process
Executable code
Accessible, consistent code environment
Easy to find experiment
?
Notebooks
Open testbeds Sharing
services
Well-documented process
Experiment actions
Accessible, consistent experimental environment
Publically shared experiment
? Integration
Publishing platform
Open testbeds
Notebooks
Sharing platform
www. chameleoncloud.org
PARTINGTHOUGHTS� Logisticalbarriersstuntcreativityandambition� TowardsaDigitalResearchEcosystem:ameetingplaceofusersand
providerssharingresourcesandresearch� Clouds/testbedsaremorethanjustexperimentalplatforms;theycreatea“common
denominator”thatcaneliminatemuchcomplexitythatgoesintosystematicexperimentation,sharing,andreproducibility
� Notebooks+testbedsprovideboththesharingunderpinnings(commonartifacts)andtheabilitytodocumenttheprocess
� Digitalerapublishingtoolsfacilitatesharing
� Leveragingnewdigitalartifactandsharingpatternstowardsupfrontshareableresearch
Recommended