Intro duction 1 Objectives of this Lecture Course - PowerPoint PPT Presentation

Intro duction 1 Objectives of this Lecture � Course Objectives � Course Contents � Course fo rmat, grading etc. � Why distributed systems ? � Puzzles � Vija c y K. Ga rg Distributed Systems F all 94

Intro duction 2 Course Objective � Cover fundamental concepts � Details of my resea rch interests � Ho w to do resea rch in dist. systems ? � Mathematica l to ols � Mak e friends � Instructo r: Vija y Ga rg, ENS 527, 471-9424 � email: ga rg@ece.utexas.ed u � O�ce Hourse: TTh 3:30 - 5:00 � Have fun - solve puzzles � Vija c y K. Ga rg Distributed Systems F all 94

Intro duction 3 Course Contents � Global Time � Global State and Kno wledge � Algo rithms: Mutual Exclusion, Causal Ordering � F ault-tolerance � Distributed Op erating Systems � Securit y � Vija c y K. Ga rg Distributed Systems F all 94

Intro duction 4 Course F o rmat � Grading � 25 % Assignments � 20 % Mid-T erm Exam � 25 % T erm P ap er � 30 % Final � Lectures � Questions � F eedback � Vija c y K. Ga rg Distributed Systems F all 94

Intro duction 5 Distributed Systems � Essential cha racteristics of distributed systems � no sha red clo ck � no sha red memo ry - @R I@ � ? 6 San F rancisco Time 12:01 MEM PE Austin Time 12:04 MEM PE Comm uni cation Net w ork Time 11:58 New Y ork MEM PE � Vija c y K. Ga rg Distributed Systems F all 94

Intro duction 6 P a rallel Systems � Distributed systems vs P a rallel systems � Physical vs logical mo del P arallel System Distributed System Soft w are La y er Soft w are La y er e.g. D ASH e.g. PVM Distributed System P arallel System c � Vija y K. Ga rg Distributed Systems F all 94

Intro duction 7 Advantages of Distributed Systems � Scalabilit y � Sha ring of resources � F ault-tolerance � Ease in p rogramming � Puzzle-lik e qualit y � Vija c y K. Ga rg Distributed Systems F all 94

Intro duction 8 Disadvantages of Distributed Systems ? ? ? Overhead Communication Synchronization Programming � Vija c y K. Ga rg Distributed Systems F all 94

Intro duction 9 Existing Machines System F eatures In tel P aragon XP/S nCUBE/2 6480 P arsys Ltd. J Mac hine/MIT Sup erNo de1000 No de T yp e and 50 MHz i860 XP CISC 64-bit CPU m ultiple T-800 Message-Driv en Memory no des with 16-128 with FPU, 14 DMA T ransputers/no de Pro cessor Mb ytes/no de 1-64 Mb ytes/no de Net w ork and 2-D Mesh with SCSI 13-dimensional Recon�gurable 8 � 8 � 8 Mesh I/O HIPPI, VME, Ethernet Hyp ercub e of 8192 in terconnect custom I/O no des OS 4.3 BSD V ertex/OS or UNIX IDRIS/OS w ormhole routing UNIX compatible Application Sparse Matrices Scien ti�c and Scien ti�c and Academic database academic P erformance 5-300 G�ops p eak 27 G�op, 36 200 MIPS to 13 64-bit results Gb ytes/s I/O GIPS p eak Source: Adv anced Computer Arc h. b y Kai Hw ang and 1993 IEEE � Vija c y K. Ga rg Distributed Systems F all 94

Intro duction 10 Time I disco v ered I disco v ered cure to AIDS cure to AIDS Time Barrier Martian : Prof T o o cold Earthling : Prof Ac heb e � clo cks not synchronized � Ho w to de�ne happ ed b efo re Time considered dangerous : replaced b y causalit y Reference: Lamp ort 79 c � Vija y K. Ga rg Distributed Systems F all 94

Intro duction 11 State � T aking picture of sky � T aking census in a country � Counting numb er of tok ens in a distributed system � � � � � � � � � � . . . . . . . . . . . . . . . . . . . . . . . . � � . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . � � . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . � � . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . � � � � � � � � � � . . . . . . . . . . x . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . � � . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Notion of physical state replaced b y consistent state Reference: Chandy and Lamp ort 85 � Vija c y K. Ga rg Distributed Systems F all 94

Intro duction 12 Secrets � Alice calls up Bob. � @R � @I � Alice and Bob do not sha re any p rivate info rmation. � Ho w can Alice transmit a secret ? � Ho w to sign y our email messages ? ? Alice Bob Eve Reference: Riv est, Shamir and Adleman 78 � Vija c y K. Ga rg Distributed Systems F all 94

Intro duction 13 Kno wledge � F ather: at least one of y ou have mud on y our fo rhead � He rep eatedly asks the question: Do y ou kno w if y ou have mud on y our fo rhead ? � What happ ens ? u t F ather n children, k have mud Kno wing is di�erent from kno wing what is kno wn Reference: Hailp ern and Moses 84 � Vija c y K. Ga rg Distributed Systems F all 94

Intro duction 14 Ho w to check violation of Mutual Exclusion � A company buys a single cop y of y our p rogram. � can run the p rogram on multiple machines 6 � - I@ @R ? � so long as there is a single cop y of the p rogram at any time. � Ho w will y ou detect any violation of the agreement � allo w ed access to a single computer at any time ? San F rancisco v Austin v Comm uni cation Net w ork New Y ork Reference: Garg and W aldec k er 94 � Vija c y K. Ga rg Distributed Systems F all 94

Intro duction 15 This course will not deal with.. � Ha rdw a re issues � Net w o rking issues � P a rallel Algo rithms � Numerical Metho ds c � Vija y K. Ga rg Distributed Systems F all 94