research in an open cloud exchange - systor · pdf fileresearch in an open cloud exchange....
TRANSCRIPT
CLOUD COMPUTING IS HAVING A DRAMATICIMPACT
• On-demand access
• Economies of scale
All compute/storage will move to the cloud?
Today’s IaaS clouds
• One company responsible for implementing and operating the cloud
• Typically highly secretive about operational practices
• Exposes limited information to enable optimizations
What’stheproblem
• LotsofinnovationabovetheIaaSlevel…but• considerEnterpriseDB,orAkamai
• Lotsofdifferentproviders…but• bandwidthbetweenproviderslimited• offeringsincompatible;switchingaproblem• pricechallengestomoving
• Novisibility/auditinginternalprocesses• Priceisterribleforcomputersrun24x7x365
Morechallenges
• Providerincentivenotalignedwithefficientmarketplace:• stickinessinprice,indifferentiation• advantageotherservices• homogeneityforefficiency
• Hardforlargeprovidertoefficientlysupportnichemarkets,radicallydifferenteconomicmodels…
• Nicheprovidersprobablycan’tsupportrichecosystem
We are in the equivalent of the pre-Internet world, where AOL and CompuServe dominated on-
line access
Whyisthisimportant• Anyonecanaddanewserviceandcompeteinalevelplayingfield• Historytellsustheopeninguptorichcommunity/marketplacecompetitionresultsininnovation/efficiency:• “TheCathedralandtheBazaar”byEricStevenRaymond• “TheMasterSwitch:TheRiseandFallofInformationEmpires”byTimWu
• Thiscouldfundamentallychangesystemsresearch:• accesstorealdata• accesstorealusers• accesstoscale
Thisisn’tcrazy…really
• Currentcloudsareincrediblyexpensive…• Muchofindustrylockedoutofcurrentclouds• lotsofgreatopensourcesoftware• lotsofgreatnichemarkets;marketsimportanttous…• lotsofusersconcernedbyvendorlockin…• thisdoesn’tneedtobeAWSscaletobeworthit
• “Pastacertainscale;littleadvantagetoeconomyofscale”— JohnGoodhue
OperatingSystems,Power,Security,Marketplace…
CloudTechnology
UniversityResearchITPartnersBU,HU,NU,UMass,MIT,MGHPCC
PartnersBrocade,CISCO,Intel,Lenovo,RedHat,TwoSigma,USAF,Dell,Fujitsu,Mellanox,CambridgeComputer…
Users/applicationsBigData,HPC,LifeSciences,…
CoreTeam&StudentsOCXmodel,HIL,Billing,Intermediaries…
DataBU,HU,NU,MIT,UMass,Foundations,Govt…
EducationandWorkforceStudents,industry
15
MOC Ecosystem
Keystone
Neutron GlanceNova Cinder
Keystone
OPENSTACK FOR AN OCX
• OpenStack is a natural starting point
• Mix & Match federation
Keystone
Neutron GlanceNova Cinder
Mix and Match (Resource Federation)
• Solution• ProxybetweenOpenStackservices
• Statusoftheproject• HostedupstreambytheOpenStackinfrastructure
• https://github.com/openstack/mixmatch• ProductiondeploymentplannedforQ12017
• Team:• CoreTeam:KristiNikolla,EricJuma,JeremyFreudberg
• Contributors:AdamYoung(RedHat),GeorgeSilvis,Wjdan Alharthi,Minying Lu,KyleLiberti
• Moreinformation:• https://info.massopencloud.org/blog/mixmatch-federation/
BostonUniversity
NortheasternUniversity
mixmatch
NovaKeystone
CinderKeystone
It’sreal…• Availablenow:ProductionOpenStackservices…
• Smallscale,butgrowing(coupleofhundredservers,550TBstorage),200+users
• VMs,on-demandBigData(Hadoop,SPARK...),
• What’scoming:– SimpleGUIforendusers– OpenShift – RedHat– Federationacrossuniversities– Rapid/secureHardwareasaService– 20+PBDataLake– CloudDataverse
• PlatformforenormousrangeofresearchprojectsacrossBU,NEU,MIT&Harvard
Researchchallenges
• Marketplacemechanisms• HostingDatasets• Multi-providercloudlet• Softwaredefinedstorage• HPContheCloud• SecureHardwareMultiplexing
Researchchallenges
• Marketplacemechanisms• HostingDatasets• Multi-providercloudlet• Softwaredefinedstorage• HPContheCloud• SecureHardwareMultiplexing
Researchchallenges
• Marketplacemechanisms• HostingDatasets,Mercè Crosas Harvard• Multi-providercloudlet• Softwaredefinedstorage• HPContheCloud• SecureHardwareMultiplexing
AWS Public Datasets
“WhendataismadepubliclyavailableonAWS,anyonecananalyzeanyvolumeofdatawithoutneedingtodownloadorstoreitthemselves.”
But, AWS public datasets miss key aspects needed in data repositories
• Incentivestosharedata• Citationtoeachversionofthedata• MetadataforDiscoverability• Tieredaccesstonon-publicdata• Commitmenttodataarchival& preservation
Today’s repositories incentivize data sharing by giving credit to data authors through formal citation
Persistentcitationstodatasetspublishedindatarepositories
Bibliography
The Dataverse open-source platform enables building any type of data repository
Agriculturedata
RepositoryinFudan,China
Datafrom20Universities
Publicdatarepository
ScienceConsortium
Researchchallenges
• Marketplacemechanisms• HostingDatasets• Multi-providercloudlet• Softwaredefinedstorage• HPContheCloud• SecureHardwareMultiplexing
Researchchallenges
• Marketplacemechanisms• HostingDatasets• Multi-providercloudlet• Softwaredefinedstorage,PeterDesnoyersNU• HPContheCloud• SecureHardwareMultiplexing
Researchchallenges
• Marketplacemechanisms• HostingDatasets• Multi-providercloudlet• Softwaredefinedstorage• HPContheCloud:ChrisHillMIT• SecureHardwareMultiplexing
Researchchallenges
• Marketplacemechanisms• HostingDatasets• Multi-providercloudlet• Softwaredefinedstorage• HPContheCloud• SecureHardwareMultiplexing:PeterDesnoyers NU,GeneCoopermanNU,NabilSchear MITLL,LarryRudolph&TrammellHudsonTwoSigma,JasonHennesseyBU,…
Hardware Isolation Layer (HIL):CONVERGING HPC, BIG DATA & CLOUD
SLURM,PBS
OpenStack
Custom
OS
(Neu
roDe
bian?)
SLURM,PBS
OpenStack
What about security?
SLURM,PBS
OpenStack
SLURM,PBS
OpenStack
Custom
OS
(Neu
roDe
bian?)
Secure Cloud Project
• Shared project with Two Sigma, MIT LL, USAF, Lenovo, Intel
• Integrating attestation infrastructure & secure FW
How fast can we do this?
Bare Metal Imaging Service
iSCSI-basedAbletoprovision+bootin<5min
Turk,A.,Gudimetla,R.S.,Kaynar,E.U.,Hennessey,J.,Tikale,S.,Desnoyers,P.,&Krieger,O.(2016).AnExperimentonBare-MetalBigData Provisioning.In8thUSENIXWorkshoponHot
TopicsinCloudComputing(HotCloud 16).
39
RapidBare-MetalProvisioningandImageManagement,RavisantoshGudimetlaandApoorve Mohan
Researchchallenges• Canweexposerichinformationaboutserviceswhilenotviolatingcustomerprivacy
• Howcanwecorrelatebetweentheinformationbetweenthedifferentlayers?
• Howcanweidentifysourceoffailures?• HowcanwecreateaNetworkingMarketplace?
Researchchallenges• Canweexposerichinformationaboutserviceswhilenotviolatingcustomerprivacy
• Howcanwecorrelatebetweentheinformationbetweenthedifferentlayers?
• Howcanweidentifysourceoffailures?• Networking Marketplace:RodrigoFonsecaBrown
Common view: Networking is like air conditioning, or power
Part of the infrastructure, provided by the datacenter
Researchenabled• Newhardwareinfrastructure;e.g.FPGAs,newprocessors
• CachingstoragefromDataLakes• Cloudsecurityandcomposabilityofsecurityproperties;e.g.,MACSproject
• Smartcities• Analysisofcloudinternalinformation(logs,metrics)forsecurity,foroptimization…
• Highlyelasticenvironments;e.g.,1000serversforaminute:
Researchenabled• Newhardwareinfrastructure;e.g.FPGAs,newprocessors:MartinHerbordt(BU)
• CachingstoragefromDataLakes• Cloudsecurityandcomposabilityofsecurityproperties;e.g.,MACSproject
• Smartcities• Analysisofcloudinternalinformation(logs,metrics)forsecurity,foroptimization…
• Highlyelasticenvironments;e.g.,1000serversforaminute:
Researchenabled• Newhardwareinfrastructure;e.g.FPGAs,newprocessors
• CachingstoragefromDataLakes:Desnoyers NU,KriegerBU
• Cloudsecurityandcomposabilityofsecurityproperties;e.g.,MACSproject
• Smartcities• Analysisofcloudinternalinformation(logs,metrics)forsecurity,foroptimization…
• Highlyelasticenvironments;e.g.,1000serversforaminute:
Simple deployment:• Cache Node per rack • L1 : Rack Local
– reduce inter rack traffic• L2 : Cluster Local
– reduce clusters and back-end storage traffic
• Implemented by modifying CEPH Rados Gateway
Node
Rack1
NodeNodeNode
L1 CACHE
CACHENODE1
Node
Rack2
NodeNodeNode
L1 CACHE
CACHENODE2
Node
RackN
NodeNodeNode
L1 CACHE
CACHENODEN
L2 CACHE
ComputeCluster
Data Lake
DatacenterscaleDataDeliveryNetwork(D3N)
D3NResults
1 2 3 4 5 6 7 8Number of Hadoop Nodes
00.5
11.5
22.5
33.5
44.5
55.5
Ag
gre
gat
e T
hro
ug
hp
ut
(GB
/s)
RGWD3N L1 Hit
Maximum SSD Bandwidth
1 2 3 4 5 6 7 8Number of Curl Nodes
0.51
1.52
2.53
3.54
4.55
Aggre
gat
e T
hro
ughput
(GB
/s)
RGWD3N L1 Hit
Maximum SSD Bandwidth
• Exceeds maximum bandwidth Hadoop
• Demonstrates makes sense to share expensive SSDs –faster than local disk
• With extreme benchmark can saturate SSD & 40 Gb NIC
• Will be of enormous value with NESE data lake
Researchenabled• Newhardwareinfrastructure;e.g.FPGAs,newprocessors
• CachingstoragefromDataLakes:Desnoyers NU,KriegerBU
• Cloudsecurityandcomposabilityofsecurityproperties;e.g.,MACSproject
• Smartcities• Analysisofcloudinternalinformation(logs,metrics)forsecurity,foroptimization…
• Highlyelasticenvironments;e.g.,1000serversforaminute:
ModularApproachtoCloudSecurity
Insecurity,thesumofthepartsisoftenahole.
– DaveSafford,circa2000
Ourgoalistobuildsecuritysystemssothatthesumofthepartsisaholisticsecurityguarantee.
– RanCanetti,2016
SynergybetweenMACSandMOC
Typesofconnections• People:researcherscancontributetowardbothprojects
– SizeofMACS:13faculty,11postdocs,25+graduatestudents• Techtransition:deployMACStechinMOCmarketplace• Problemcreation:MOC’sproblemsfeedMACSresearch• Funding:jointcloudresearchhasmultipliereffect
ValuethatMOCprovidestoMACS• Access:data,meta-data,scale,problems,andusers• Uniquetrustrelationships:federateddatacenter Hardware
CloudIaaSmanagement
Operatingsystem
Applications&platforms
Algorithms&techniques
MACS→MOC MOC→MACS
InterplaybetweenMACSandMOC
Federated Monitoring
MOCMonitoringInfrastructure
Private MonitoringEbbRT
SecureHardware HI
LBMI Secure
Cloud
Secure Dataverse
UC Analysis
Legend:• Yellow =MACS• Blue =MOC• Green =Joint
Researchenabled• Newhardwareinfrastructure;e.g.FPGAs,newprocessors
• CachingstoragefromDataLakes:Desnoyers NU,KriegerBU
• Cloudsecurityandcomposabilityofsecurityproperties;e.g.,MACSproject
• Smartcities:AzerBestavrosBU• Analysisofcloudinternalinformation(logs,metrics)forsecurity,foroptimization…
• Highlyelasticenvironments;e.g.,1000serversforaminute:
Researchenabled• Newhardwareinfrastructure;e.g.FPGAs,newprocessors
• CachingstoragefromDataLakes:Desnoyers NU,KriegerBU
• Cloudsecurityandcomposabilityofsecurityproperties;e.g.,MACSproject
• Smartcities• Analysisofcloudinternalinformation(logs,metrics)forsecurity,foroptimization…:…:AlinaOpreaNU
• Highlyelasticenvironments;e.g.,1000serversforaminute:
Analytics-baseddefenses
• Goals– Correlatedatasourcesfrommultiplecloudlayers
• Builduser,VMandapplicationprofiles– Machinelearningtechniquestodetectwiderangeofthreats
• Protectionofcloudinfrastructure• Enableclouduserstoprotecttheirresources
– ProvidedatacollectionandanalyticsAPIstousers58
Behavior-basedauthentication
• Detect credential compromise–Developers leak their AWS passwords in GitHub
• Build user profiles based on historical data–Login information (IP address, time)–VM usage (CPU, memory, disk)
• Anomaly detection–Detect unusual activities
59
Suspiciousaccounts
Networktrafficanalysis
60
Usecases• DetectsuspiciouscommunicationwithexternalIPaddresses• Detectdataexfiltrationattempts• Preventcloudabuse
– Malwareinfection,applicationexploits,illegaluseofcloud
sFlowcollecto
r
sFlowcollecto
r
MongoDB
Researchenabled• Newhardwareinfrastructure;e.g.FPGAs,newprocessors
• CachingstoragefromDataLakes:Desnoyers NU,KriegerBU
• Cloudsecurityandcomposabilityofsecurityproperties;e.g.,MACSproject
• Smartcities• Analysisofcloudinternalinformation(logs,metrics)forsecurity,foroptimization…:…:AlinaOpreaNU
• Highlyelasticenvironments;e.g.,1000serversforaminute: JonathanAppavooBU
ExampleSupportingInteractive,BurstyHPCApplications:OSDI2016
EbbRTdistributedlibraryOS[AppavooBU]:• Front-endLinuxallocatesbare-metalback-endnodesondemand
• Back-endnodeslibraryOScustomizedtosingleapplicationneeds
[Appavoo]
B
LinuxFront-End
LibraryOSBack-Ends
BWebInterface
ElasticSoftware
ElasticInfrastructure
BInfrastructureasElasticResourcePool
XSPcomputeservicebasedonKittyhawk[AppavooIBM]• Fastprovisioningbasedonbroadcast• HardwarelevelbasedonHaaS• IaaSlevelbypre-allocatingVMsoutofOpenStack
RequestResponse
Exemplar
0
2250
4500
6750
9000
11250
PC BG1K BG4k BG16k
seco
nds
synthetic1024x1024200 slices
Fetal Image Reconstruction
~2.4hrs
~24s
resized,cropped96x9650 slices
APP
IRTK24hrs
RedHatCollaboratory
• MonitoringandAnalytics• OpenShift ontheMOC• DatacenterscaleDataDeliveryNetwork(D3N)• HIL&QUADS• AcceleratorTestbed• BigDataAnalyticsandCloudDataverse
End-to-endPOC:RadiologyinthecloudtargetingOpenShift withaccelerators
Concludingremarks• MOCafunctioningsmallscalecloudforregiontoday:
–http://info.massopencloud.org• KeydriveristheOCXModel:
–KeyenablersgoingoninOpenStack (beenachallenge)–couldbecomeimportantcomponentofclouds–Majorresearchchallenge &opportunities–Enablingresearchtoco-existswithproduction:
• realdata,realusers,realscale• Getinvolved:useit,internships,exposeresearch• Startreplicatedthemodelelsewhere