science data and the ndn paradigm · 2016. 9. 13. · program-mable network extreme data science...

17
Science Data and the NDN paradigm Inder Monga CTO, ESnet Division Deputy of Technology, Scien?fic Networking Division Lawrence Berkeley Na?onal Lab NDN Comm 2015

Upload: others

Post on 07-Oct-2020

0 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Science Data and the NDN paradigm · 2016. 9. 13. · Program-mable network Extreme Data Science Facility (XDSF) MS-DESI producing ALS LHC JGI APS LCLS ... 5 × 1015 1 × 1016 1.5

ScienceDataandtheNDNparadigmInderMongaCTO,ESnetDivisionDeputyofTechnology,Scien?ficNetworkingDivisionLawrenceBerkeleyNa?onalLab NDNComm2015

Page 2: Science Data and the NDN paradigm · 2016. 9. 13. · Program-mable network Extreme Data Science Facility (XDSF) MS-DESI producing ALS LHC JGI APS LCLS ... 5 × 1015 1 × 1016 1.5

Experimental and observational science deals with big and small

instruments, and a lot of data!

2

●  DatavolumesareincreasingfasterthanMoore’sLaw

●  Newalgorithmsandmethodsforanalyzingdata

●  Infeasibletoputasupercompu>ngcenterateveryexperimentalfacility

Compu?ngSciencesArea

Page 3: Science Data and the NDN paradigm · 2016. 9. 13. · Program-mable network Extreme Data Science Facility (XDSF) MS-DESI producing ALS LHC JGI APS LCLS ... 5 × 1015 1 × 1016 1.5

Alltoocommonprocessofdiscovery

3

Page 4: Science Data and the NDN paradigm · 2016. 9. 13. · Program-mable network Extreme Data Science Facility (XDSF) MS-DESI producing ALS LHC JGI APS LCLS ... 5 × 1015 1 × 1016 1.5

ESnet

NewMath

Real->meanalysis

HighperformanceSoFware

Novelcompute/dataplaHorms

Datamgmt.andsharing

Program-mablenetwork

Extreme Data Science Facility

(XDSF)

MS-DESI

ALS

LHC

JGI

APS

LCLS

Other data-producing sources

-4-

‘Superfacility’Vision:Anetworkofconnectedfacili>es,soFwareandexper>setoenablenewmodesofdiscovery

Page 5: Science Data and the NDN paradigm · 2016. 9. 13. · Program-mable network Extreme Data Science Facility (XDSF) MS-DESI producing ALS LHC JGI APS LCLS ... 5 × 1015 1 × 1016 1.5

ESnetisadedicatedmissionnetworkengineeredtoaccelerateabroadrangeofscienceoutcomes.

Wedothisbyofferinguniquecapabili?es,andop?mizingthenetworkfordataacquisi?on,data

placement,datasharing,datamobility.

Page 6: Science Data and the NDN paradigm · 2016. 9. 13. · Program-mable network Extreme Data Science Facility (XDSF) MS-DESI producing ALS LHC JGI APS LCLS ... 5 × 1015 1 × 1016 1.5

ESnetisdesignedfordifferentgoalsthan

generalInternet.

Page 7: Science Data and the NDN paradigm · 2016. 9. 13. · Program-mable network Extreme Data Science Facility (XDSF) MS-DESI producing ALS LHC JGI APS LCLS ... 5 × 1015 1 × 1016 1.5

Feb-94 Apr-98 Jun-02 Aug-06 Oct-10 Dec-14

1017

1010

1011

1012

1013

1014

1015

1016

Month

Tota

l

Bytes of Science Data Transferred Each Monthby the Energy Sciences Network

August2015:29.13PB

Lotsofdatatomovearound

Page 8: Science Data and the NDN paradigm · 2016. 9. 13. · Program-mable network Extreme Data Science Facility (XDSF) MS-DESI producing ALS LHC JGI APS LCLS ... 5 × 1015 1 × 1016 1.5

Feb-94 Apr-98 Jun-02 Aug-06 Oct-10 Dec-14

3 × 1016

0

5 × 1015

1 × 1016

1.5 × 1016

2 × 1016

2.5 × 1016

Month

Tota

l

Bytes of Science Data Transferred Each Month

by the Energy Sciences Network

August2015:29.13PB

Lotsofdatatomovearound(contd.)

Page 9: Science Data and the NDN paradigm · 2016. 9. 13. · Program-mable network Extreme Data Science Facility (XDSF) MS-DESI producing ALS LHC JGI APS LCLS ... 5 × 1015 1 × 1016 1.5

High-levelobjec>vesforscien>ficdata:alignmentwithNDNapproach

10/25/159

•  Radicallysimplifyhowscien?ficusersmanage,moveandmanipulatelarge,distributed,sciencedatarepositories,butwithhigh-throughputend2end

•  Abstractthestorageandnetworkcapabilityandloca?ondependencefromtheuser-datainterac?on

•  Enabletheabilityforuserstospecifyandretrievepor?onsofdatatheworkflowneeds

•  Createasecure,scalableframeworkbasedonintegrateddatamanagementandnetworktransport

Page 10: Science Data and the NDN paradigm · 2016. 9. 13. · Program-mable network Extreme Data Science Facility (XDSF) MS-DESI producing ALS LHC JGI APS LCLS ... 5 × 1015 1 × 1016 1.5

UseCase#1

10/25/1510

ResearchersfromBerkeleyLabandSLACconductedproteincrystallographyexperimentsatLCLStoinves>gatephotoexcited

statesofPSII,withnear-real->mecomputa>onalanalysisatNERSC.

“Takingsnapshotsofphotosynthe?cwateroxida?onusingfemtosecondX-raydiffrac?onandspectroscopy,”NatureCommunica.ons5,4371(9July2014)

50TBmovedanight

Page 11: Science Data and the NDN paradigm · 2016. 9. 13. · Program-mable network Extreme Data Science Facility (XDSF) MS-DESI producing ALS LHC JGI APS LCLS ... 5 × 1015 1 × 1016 1.5

UseCase#2:LHCONEdata–mul>plereplicas,globalreach

Page 12: Science Data and the NDN paradigm · 2016. 9. 13. · Program-mable network Extreme Data Science Facility (XDSF) MS-DESI producing ALS LHC JGI APS LCLS ... 5 × 1015 1 × 1016 1.5

UseCase#3:Interna>onalClimateData

10/25/1512

Page 13: Science Data and the NDN paradigm · 2016. 9. 13. · Program-mable network Extreme Data Science Facility (XDSF) MS-DESI producing ALS LHC JGI APS LCLS ... 5 × 1015 1 × 1016 1.5

Percep>onoflimita>onsofNDNmo>va>ngresearchques>ons

1.  IfIammoving50TBofdatathroughasinglepath,fromanexperimenttoastoragefacility,IreallydonotwanttocacheitateveryintermediateNDNnode

–  Whatistherightstrategyforalloca?ngdiskresourcestocaching?Whatifonedatatransferconsumesallcacheresourcesorthereisnotenoughspace?

2.  Whatistheperformanceoftheend-to-enddatatransfer?HowcanIgetlineratethroughput?

3.  HowdoIleveragetheknowledgeofnetworkcapabilityinchoosingthetransferpath?HowdoIbuildintheknowledgeofunderlayintotheNDNoverlay?

4.  HowdoIleveragenetworkprogrammabilitytodotheabove?

5.  Andmanyotherques?ons….

10/25/1513

Page 14: Science Data and the NDN paradigm · 2016. 9. 13. · Program-mable network Extreme Data Science Facility (XDSF) MS-DESI producing ALS LHC JGI APS LCLS ... 5 × 1015 1 × 1016 1.5

Whereareweat?

•  Collabora?onwithChristosandColoradoState–high-poweredNDNdevicesbetweenthreerepresenta?veclimatesitesasatestbed–  Susmitworkingonansweringsomeofthehigh-levelobjec?vesasdescribed

•  HEPandASCRinterestinNDNfromaresearchperspec?ve–paperearlierthisyear@CHEP,andPhilwilltalkaboutnext-stepsrightaher

•  Interestinexpandingafedera?onofhigh-poweredNDNdeviceswiththerightstrategyforcachinganddatamanagement

•  CombiningNDNwithSDN–wehaveanext-genSDNtestbedacrossUSandEurope–canwecombinethattoprovidetherightprimi?vesforhigh-performanceNDN?–  Letsdoitera?veexperimenta?onandimprovement!!!!!!!

10/25/1514

Page 15: Science Data and the NDN paradigm · 2016. 9. 13. · Program-mable network Extreme Data Science Facility (XDSF) MS-DESI producing ALS LHC JGI APS LCLS ... 5 × 1015 1 × 1016 1.5

ALBQ

AMST

ANL

AOFA

ATLA

BNL

BOIS

BOST

CERN

CHICDENV

ELPA

FNAL

HOUS

KANS

LANL

LBL LLNL

LOND

NASHNERSC

NEWY

ORNL

PNNLPNWG

SACR

SAND

SLAC

STAR

SUNN

WASH

ESnet PE Router

(2+)x10GE

(n)x10GE

Testbed Host

Deployed SDN Testbed node locations Deployed SDN Testbed connectivity overlay (using OSCARS circuits)

ESnetSDNTestbed

AMST

CERN

AOFA

WASH

STAR

ATLA

DENV

LBL

August2015 iDiscovery2020InderMonga15

StatusUpdate:•  Testbeddeployedatallloca?ons•  QoSsupportverified,pressrelease

nextweek•  ENOSdemoonTestbed@SC

Page 16: Science Data and the NDN paradigm · 2016. 9. 13. · Program-mable network Extreme Data Science Facility (XDSF) MS-DESI producing ALS LHC JGI APS LCLS ... 5 × 1015 1 × 1016 1.5

Thankyou!

•  Pleasefeelfreetoemailmewithques?ons,commentsorarrowsat

imongaatesdotnet

10/25/1516

Page 17: Science Data and the NDN paradigm · 2016. 9. 13. · Program-mable network Extreme Data Science Facility (XDSF) MS-DESI producing ALS LHC JGI APS LCLS ... 5 × 1015 1 × 1016 1.5

10/25/1517