Carnegie Mellon Andrew Overview Walter Wong Computing Services.
-
Upload
shannon-spencer -
Category
Documents
-
view
221 -
download
0
description
Transcript of Carnegie Mellon Andrew Overview Walter Wong Computing Services.
Carnegie Mellon
Andrew OverviewWalter WongComputing Services
Carnegie Mellon
Andrew YesterdayBuilding the next generation campus computing environmentFunded by IBM in 1982Early large scale distributed system
Carnegie Mellon
AFSAUISAMSWorkstation Configuration
Significant Contributions
Wor
ksta
tion
conf
igAFS
AUIS
AMS
Carnegie Mellon
AFSScalable Distributed FilesystemClient Side CachingGlobal NamespaceServer IndependenceUnderstandable ACLsVolumes
Carnegie Mellon
AUISAndrew User Interface SoftwareObjected orientated, customizable, multi-media environment
Carnegie Mellon
AMSAndrew Messaging SystemBuilt on AUISAuthenticatedShared folders (bboards)
Carnegie Mellon
Workstation ManagementOperating system and configurations remain consistentSynchronization of existing files on AFS for central maintenanceRemoval of files that “don’t belong”
Carnegie Mellon
Andrew/Unix Workstations
Original Andrew
Information
Backup
Mail/News
Printing
Clients ServicesInfrastructure
AFS Fileservers
Carnegie Mellon
Transition to ProductionDefault campus environment in 1988Rapid GrowthSoftware Release ManagementDelegation and Distribution of AdministrationPersonal Computers
Carnegie Mellon
Rapid GrowthNeed for monitoring and reporting Stable servicesReduce incremental administrative costs through automation and tools Otherwise, no time for development
Carnegie MellonSoftware Release Management
Identify all software componentsControlled test environmentsRollback supportImplemented under Unix with Depot & EMT http://andrew2.andrew.cmu.edu/depot http://andrew2.andrew.cmu.edu/emt
Carnegie Mellon
Distributed AdministrationProvide a default configurationProvide an infrastructure for customizationat multiple levelsAllow one to control whatthey want - ignore whatthey do not want
Vendor
Central
Departmental
Workgroup
Personal
Carnegie Mellon
Personal ComputersMore affordable than workstationsDifficult to port AFS and other applicationsMissing key features in the operating systems
Carnegie Mellon
Architectural ProblemsAFS is complicated, but necessarily soSystem built on AFS inherit its complexity, often unnecessarilyAFS has not been ported to many platforms (Mac, DOS, Windows 3.x)Performance problems in AFS are inherited by other servicesAFS is not good over slow linksAFS is going to be replaced with the DFS.… Or maybe not...
Carnegie Mellon
Personal Computers
Andrew Plus
Clients ServicesInfrastructure
Andrew/Unix Workstations
Information
Backup
Mail/News
Printing
Andrew/Unix Workstations AFS
Fileservers
Carnegie Mellon
Andrew IIInitiated in 1992 to cope with DCE/DFSNo single system or single delivery dateMultiple projects with common goals
Carnegie Mellon
Andrew II
Filesystems
OtherWorkstations
AndrewWorkstations
Information BackupMail/NewsPrinting
Carnegie Mellon
Andrew II GoalsStandards based or in wide spread useAvoid unnecessary filesystem dependenciesSimple independent scaleable servicesPC and Mac integrationDisconnected operation
Carnegie Mellon
Andrew II StrategiesLimit operating system modificationsLayer local changes on vendor solutionsSeek vendor solutions when ever possibleSeek vendor and external use of local solutionsIntegrate external solutionsAdapt local environment when market forces dictate change
Carnegie Mellon
Major Andrew II ProjectsMessaging - CyrusInformation Services - Polaris, Infoserver, Userweb
Carnegie Mellon
Project CyrusInspired by ancient Persian Royal Postal System System linked an Empire 1500 miles broad (delivery time from end to end: 3 weeks)Postal System was integral to ordered organization of societyDurable, enterprise communications -- lasted for over 1500 years
Carnegie Mellon
Project Cyrus GoalsScalable Enterprise Wide Mail SystemAccess Control ListsQuota per User/MailboxServer IndependenceMessage Service Integration:Mail, Netnews, local bulletin boards, internet mailing lists, wire services, etc.
Carnegie Mellon
Project Cyrus StatusImplemented high quality IMAP server http://andrew2.andrew.cmu.edu/cyrusWork progressing with ACAP for server independenceIMAP proxy as interim measureWaiting for client vendors to catch upSASL LibrarySMTP AUTH
Carnegie Mellon
PolarisWeb based help serviceDynamic Browse Space Can be designed and built by
documentation staff or by developers as they release help pages
Full-text searchingAUIS, man, and HTML supporthttp://polaris.andrew.cmu.edu/
Carnegie Mellon
InfoserverCMU Online Information SystemWeb based Depot managedFull-text searching http://www.cmu.edu
Carnegie Mellon
User/Course Web ServiceDistributed filesystem manages quota Immediate publishing No release managementFull-text searchinghttp://www.andrew.cmu.edu
Carnegie Mellon
Major Current InitiativesFull NT integrationDirectory ServicesSecurity Services
Carnegie Mellon
OrpheusNT provides a ‘real’ operating system on PCsMicrosoft considers interoperability as a method of conquestChallenge is fully integrating NT environment with existing infrastructure
Carnegie Mellon
Directory ServicesProvide the integration missing when filesystem is removedLDAP basedIntegration with Microsoft Active Directory
Carnegie Mellon
Security ServicesIntegration/co-existence of public and secret key systemsUnsolved public key infrastructure issues
Carnegie Mellon
Andrew Today and BeyondEducational environments differ in interesting waysRapid changesInnovative Integration with focused development
Carnegie MellonEducational Environment Differences
Security Greater (recognized) internal threats Conflicting goals of collaboration vs. privacy
Less rigid hierarchy A person may be staff, student and faculty
Nomadic behavior A person may use multiple machines, some
they do not have administrative control over Some machines may be shared among
many diverse individuals
Carnegie Mellon
Rapid ChangesLocal solutions can not keep up with the featuresOn going support of legacy systems is expensivePorting local solutions to new environments is expensiveOther groups and individuals on campus adopt commercial systems
Carnegie MellonAdopting New Technologies
Seek vendor solutionsCooperate with vendorsSeek public solutionsCooperate to enhance public solutionsCooperate to develop new technologies and get those technologies on standards tracksDo it ourselves
Carnegie Mellon
Scalability ModelsMonolithic Have a single highly
redundant system Simplifies
administration Single point of
failure
Carnegie Mellon
Scalability ModelsDistributed - server dependent Multiple servers provide
redundancy, better performance
Burden on users to know about the servers
Higher administrative cost in managing multiple machines
Carnegie Mellon
Scalability ModelsDistributed – server independent Actual server(s) hidden
by a layer of indirection Additional client/server
support
Carnegie Mellon
Integration ProblemsScale The market is still primarily producing work
group solutions Can’t scale to a single computing community
of over 10,000 users and over 6,000 networked machines
Interoperability Many solutions do not work together Many solutions do not exist across
architectures
Carnegie Mellon
Focusing DevelopmentFocus on key areas and do proof of concept development (and maybe deployment)Force issues in standards bodies (IETF, etc.)Leverage development efforts with vendors to influence directions
Carnegie Mellon
Unique Users
0
200
400
600
800
1000
1200
1400
1600
1800
Average
Peak
Carnegie Mellon
ReferencesHttp://andrew2.andrew.cmu.eduKhanna, Raman, ed. “Managing Distributed Systems” and “System Management Strategies” in Distributed Computing: Implementation and Management Strategies. Prentice Hall. 1994.Khanna, Raman, ed. “Personal Computer Integration in the Andrew System” in Integrating Personal Computers in a Distributed Client-Server Environment. Prentice Hall, 1995.