Module 2 Fundamental Architectures and Models of Distributed Systems

Transcription

1 Module 2 Fundamental Architectures and Models of Distributed Systems Raouf Boutaba 1

2 Module Goal To gain a good understanding of the challenges related to heterogeneity, openness, security, scalability, failure handling, concurrency and transparency as they apply to distributed systems To place distributed systems in a realistic context through examples: the Internet, an intranet and mobile networking/computing To provide the conceptual models to support and motivate the study of many of the design problems and solutions in distributed systems Raouf Boutaba 2

3 Module Overview Fundamental Concepts Advantages/Disadvantages Taxonomy of DS Architectures Challenges Examples of Distributed Systems Computational Models Raouf Boutaba 3

4 The Tim Allen View of Computing Faster!!! Bigger Problems I want 7 days of weather not 2 I want 1024x1024x16-bit color Most modern applications such as weather prediction, aerodynamics and artificial intelligence are very computationally intensive Raouf Boutaba 4

5 Why High Performance? To calculate a 24 hour forecast for the UK requires about operations to be performed Takes about 2.7 hours on a machine capable of 10 8 operations per second What about a 7 day forecast? Okay so buy a faster computer The speed of light is 3x10 8 m/s. Consider two electronic devices (each capable of performing operations/second) 0.5mm apart. It takes longer for the signal to travel between them than it takes for either of them to process it. Raouf Boutaba 5

6 Why Are Things Getting Better? Half past century: Unprecedented improvement in computer industry 10 million dollars machines executed 1 instruction per second Price/perf gain dollars machines executing 10 millions instruction per second If car industry improved at same rate in same period of time, a Rolls Royce would now cost 10 dollars and get a billion miles per gallon Raouf Boutaba 6

7 Raouf Boutaba 7

8 What Happened Exactly? Two advances in technology allowed such improvement: Development of Powerful microprocessors (8-, 16-, 32-, & 64-bit CPUs) Invention of high-speed computer networks that allow a large number of machines to exchange information at high rates Local Area Networks (LANs) (10/100 s machines; data transfer in less than a ms) Wide Area Network (WANs) (10/100 millions machines; from 64Kbps to Gbps) Raouf Boutaba 8

9 The Bottom Line It is getting harder to extract the performance modern applications require out of a single processor machine Given practical constraints, such as the speed of light, or latency issues, machines with multiple processing elements seem to be the only way to go Sharing resources was an important use of early LAN Technology (File servers; Printers) Result: Distributed systems (Large numbers of CPUs connected by a high-speed networks) Contrast: Centralized systems (Single CPU, its memory, peripherals, & some terminals) Raouf Boutaba 9

10 What s a Distributed System? A collection of autonomous computers linked by a network, with software designed to produce an integrated computing facility A distributed system is a collection of independent computers that appear to the users of the system as a single computer A system in which hardware or software components located at networked computers communicate and coordinate their actions only by passing messages Raouf Boutaba 10

11 Working Definition A distributed system is several computers doing something together Three primary features of a distributed system Multiple computers Communications Virtual Computer A well designed distributed system should be perceived as a single, integrated computing facility World Wide Web Automatic Teller Machines Cell Phones Raouf Boutaba 11

12 Advantages Economics Speed Inherent distribution Reliability Incremental growth Microprocessors offer a better price/ performance than mainframes A distributed system may have more total computing power than a mainframe Some applications involve spatially separated machines If one machine crashes, the system as a whole can still survive Computing power can be added in small increments Raouf Boutaba 12

13 Economics Grosch s law (Quarter of a century ago) The computing power of CPU is proportional to the square of its price. By paying twice as much, you could get 4 times the performance With microprocessor technology, no longer hold! Few $100 s get a CPU chip faster than the largest 1980s mainframes By paying twice as much, you get same CPU, but running at a somewhat higher clock speed Raouf Boutaba 13

14 Economics (Cont) More cost-effective to harness a large number of cheap CPUs together in a system First & Leading motivation for distributed systems: Better price/performance ratio than a single large centralized system Raouf Boutaba 14

15 Advantages Economics Speed Inherent distribution Reliability Incremental growth Microprocessors offer a better price/ performance than mainframes A distributed system may have more total computing power than a mainframe Some applications involve spatially separated machines If one machine crashes, the system as a whole can still survive Computing power can be added in small increments Raouf Boutaba 15

16 Speed A collection of microprocessors may yield an absolute performance that no mainframe can achieve at any price! Example: Can build a system from 10,000 CPU chips If each CPU chip runs at 50 MIPS, total performance: 50,000 MIPS To achieve same performance a single CPU CPU has to execute an instruction in nsec (2 picosec) Unlikely to happen from theoretical & engineering points of view Einstein s theory of relativity: Nothing can travel faster than light, which can cover only 0.6 mm in 2 picoseconds Raouf Boutaba 16

17 Advantages Economics Speed Inherent distribution Reliability Incremental growth Microprocessors offer a better price/ performance than mainframes A distributed system may have more total computing power than a mainframe Some applications involve spatially separated machines If one machine crashes, the system as a whole can still survive Computing power can be added in small increments Raouf Boutaba 17

18 Inherent Distribution Some applications are inherently distributed and involve spatially separated machines Examples: Computer-supported collaborative work Computer-supported cooperative games Commercial applications, banking, etc. Raouf Boutaba 18

19 Advantages Economics Speed Inherent distribution Reliability Incremental growth Microprocessors offer a better price/ performance than mainframes A distributed system may have more total computing power than a mainframe Some applications involve spatially separated machines If one machine crashes, the system as a whole can still survive Computing power can be added in small increments Raouf Boutaba 19

20 Reliability Ideally, if 5% of the machines are down at any moment, the system should be able to continue to work with a 5% loss in performance Very important feature for critical applications Examples: Control of nuclear reactors or aircrafts Military control and command systems Raouf Boutaba 20

21 Advantages Economics Speed Inherent distribution Reliability Incremental growth Microprocessors offer a better price/ performance than mainframes A distributed system may have more total computing power than a mainframe Some applications involve spatially separated machines If one machine crashes, the system as a whole can still survive Computing power can be added in small increments Raouf Boutaba 21

22 Incremental Growth A big plus Upgrading a system based on a mainframe means: Replace the mainframe with a larger one (if it exist)! Add a second mainframe! With a distributed system: gradual expansion as the need arises Add more processors to the system Raouf Boutaba 22

23 Driving Force in the Long Term Existence of large numbers of personal computers Need for people to work together and share information in a convenient way without being bothered by geography or distribution of people, data, and machines Distributed Systems Versus Personal Computers Data sharing Device sharing Communications Flexibility Allow many users access to a common data base, e.g., airline reservation systems Allow many users to share expensive peripherals, e.g., color printers and optical jukeboxes Make human-to-human communication easier, for example, by electronic mail Spread the work over the available machines in the most cost effective way. Raouf Boutaba 23

24 Disadvantages Software Networking Security Little software exist at present for distributed systems The network can saturate or cause other problems Easy access also applies to secret data Still advantages of distributed systems outweigh the disadvantages and it is expected that distributed systems will become increasingly important in the coming years Raouf Boutaba 24

25 Consequences Concurrency Concurrency is the norm instead of the exception Synchronization is critical No Global Clock There is a limit as to how accurate a global clock can be Independent Failures The more stuff you add the more likely something will break Single system view says independent failures should not affect users Communication Issues Unreliable, Insecure, Costly Raouf Boutaba 25

26 Hardware A large collection of processing elements that can communicate and cooperate to solve large problems quickly Parallelism SEQUENTIAL If several operations can be performed simultaneously, the total computation time is reduced Example s parallel version has the potential of being 3 times faster INPUT PARALLEL INPUTS OUTPUT OUTPUTS Raouf Boutaba 26

27 Parallel Architectures Unlike traditional von Neumann machines, there is no single standard architecture used on parallel machines In fact dozens of different parallel architectures have been built and are being used Most important is how the hardware is organized, i.e. how the multiple CPUs are interconnected? Several people have tried to classify the different types of parallel machines The taxonomy proposed by Flynn is the most commonly used Raouf Boutaba 27

28 Flynn s Model of Computation Any computer, whether sequential or parallel, operates by executing instructions on data. A stream of instructions (the algorithm) tells the computer what to do. A stream of data (the input) is affected by these instructions. Depending on whether there is one or several of these streams, Flynn s taxonomy (1972) defines four classes of computers Instruction Streams Single Multiple Single Single Instruction Single Data Multiple Instruction Single Data Data Streams Multiple Single Instruction Multiple Data Multiple Instruction Multiple Data Raouf Boutaba 28

29 The 4 Classes Potential of the 4 Classes + A B SISD SISD A + B + * A B MISD MISD A + B C * D + A B C D SIMD SIMD A + B C + D + * A B C D MIMD MIMD A + B C * D Examples: SISD (all traditional uniprocessor computers) SIMD (repetitive computations, some supercomputers) MISD (no known computer fits this model) MIMD (group of independent computers, each with its own instruction unit and data) Raouf Boutaba 29

30 All Existing Systems Are MIMD MIMD Tightly Coupled Multiprocessors (shared memory) Parallel and Distributed Systems Loosely Coupled Multicomputers (private memory) Bus Switched Bus Switched Sequent, Encore Ultracomputer, RP3 Workstations on a LAN Hypercube, Transputer Raouf Boutaba 30

31 Multiprocessors Bus-Based Multiprocessors CPU Cache CPU Cache CPU Cache Memory Switched Multiprocessors Memories M M M M C C CPUs C C C C C Bus 2) Omega network 2*2 switches M M M M C 1) Crossbar switch 3) NUMA machine Raouf Boutaba 31

32 Multicomputers Bus-Based Multicomputers Workstation Workstation Local Memory Local Memory CPU CPU Switched Multicomputers Workstation Local Memory CPU Network Grid Hypercube Raouf Boutaba 32

33 Software Software is more important than hardware! Operating system software determines the image that a system presents to its users Software is more difficult to categorize, still 2 kinds: Loosely-coupled software Allows machines and users to be fundamentally independent of one another, but still interact to a limited degree where that is necessary Group of personal computers sharing a printer on a LAN Tightly-coupled software Allows the CPUs in a multiprocessor machine to work together in solving a single problem To run a chess program in parallel Raouf Boutaba 33

34 Software/Hardware Combinations Software Loosely Tightly Coupled Coupled Hardware Loosely Coupled Tightly Coupled Network Operating Systems True Distributed Systems Multiptocessor Timesharing Systems Raouf Boutaba 34

35 Network Operating Systems Typical example: A network of workstations on a LAN Primitive forms of communication: Remote login: rlogin host Remote copy: rcp host1:file1 host2:file2 More convenient forms of information sharing Clients File server Request LAN Reply Raouf Boutaba 35

36 True Distributed Systems Create illusion of single timesharing system, referred to as single system image or virtual uniprocessor No current system fulfills this requirement entirely, i.e., hide to users the existence of multiple CPUs Some characteristics of a distributed system: Single global interprocess communication mechanism Same process management everywhere Same system call interface everywhere File system must look the same everywhere Raouf Boutaba 36

37 Multiprocessor Timesharing Systems Tightly-coupled software on tightly-coupled hardware Used for special purpose machines, e.g., dedicated data base machines Most common general purpose are multiprocessors operated as a UNIX timesharing system, but with multiple CPUs (e.g., MIPS CPUs providing a single-system image, i.e., a 960 CPU) Key characteristic: A single run queue CPU 1 Process A (running) Cache CPU 2 Process B (running) Cache CPU 3 Process C (running) Cache Disk Bus Memory E (ready) D (ready) C (running) B (running) A (running) Run Queue: D, E Operating system Raouf Boutaba 37

38 Comparison Network True Multiprocessor Operating Distributed Timesharing System System System Does it look like a virtual uniprocessor? No Yes Yes Do all have to run the same OS? No Yes Yes How many copies of the OS are there? N N 1 How is communication achieved? Shared files Messages Shared memory Are agreed upon net. protocols required? Yes Yes No Is there a single run queue? No No Yes Does file sharing have well-defined semantics? Usually No Yes Yes Raouf Boutaba 38

39 Examples of Distributed Systems The Internet Intranets Mobile Cellular Networking Mobile Ubiquitous Computing World Wide Web Other Distributed Systems Raouf Boutaba 39

40 The Internet Heterogeneous network of computers and applications Implemented through the Internet protocol stack A typical portion of the Internet ISP Intranet Backbone Satellite Link Desktop Computer: Server: Network Link: Raouf Boutaba 40

41 Intranets Locally administered network Usually proprietary (e. g., the University campus network) Interfaces with the Internet through Firewalls Provides services internally and externally Print and other Servers server Desktop Computers Web server Local Area Network server File server the rest of the Internet router/firewall Print other servers Raouf Boutaba 41

42 Mobile Cellular Networking Cellular phone systems (e. g., GSM, UMTS) Resources being shared Radio frequencies Transmission times on one frequency (UMTS: multiplexing) The mobile on the move Global System for Mobile Communications (GSM) Forms the basis for what is called digital cellular system 1800 (DCS1800). GSM/DCS is designed to include a wide variety of services including: Speech Fax Emergency call Raouf Boutaba 42

43 DCS1800/GSM900 BCS BCS M S C VLR HLR AC BCS EIR Raouf Boutaba 43

44 Description The MSC is the heart of GSM sets up, manages, and clear connections routes calls to the proper cells interfaces to the telephone system Databases Home Location Register (subscriber information) Visitor Location Register (users in a particular area) Authentication Center Equipment Identity Register Raouf Boutaba 44

45 Call Routing Telephone Network Gateway MSC Terminating MSC BSC 9 HLR 4 VLR 1 = Call made to mobile unit 2 = Telephone network recognizes number and gives to gateway MSC 3 = MSC can t route further, interrogates user s HLR 4 = HLR interrogates VLR currently serving user 5 = Routing number returned to HLR and then gateway MSC 6 = Call routed to terminating MSC 7 = MSC asks VLR to correlate call to the subscriber 8 = VLR complies 9 = Mobile unit is pages 10 = Mobile unit responds; information conveyed back to telephone Raouf Boutaba 45

46 Location Updating Previous MSC 2 4 New MSC BSC 1 4 New VLR 5 HLR 3 New VLR 1 = Location update request 2 = Location update message 3 = Subscription data return 4 = Location update acknowledgement 5 = Location cancellation message Raouf Boutaba 46

47 Mobile Computing Wireless LANs (university campus WLAN soon to come here?) Laptop Computers Handheld devices, PDAs etc. Wearable devices Internet Host intranet Wireless LAN WAP Gateway Home intranet PDA Printer Camera Laptop Host site Raouf Boutaba 47

48 The World Wide Wait! Open client- server architecture implemented on top of the Internet Shared resources Information, uniquely identified through a Uniform Resource Locator (URL) Variants: Intranet- based Webs PC running Explorer Mac running Navigator Browsers Internet WWW Server Server running BBCR Web server Cs454.html File system of bbcr.uwaterloo.ca Raouf Boutaba 48

49 Other Distributed Systems Distributed Multimedia- Systems Often use Internet infrastructure Characteristics Heterogeneous data sources and sinks (Video, Audio, Text) that need to be synchronized in real time Video Audio Text Often: Distribution services (Multicast) Examples Teleteaching tools (mbone- based, etc.) Video- conferencing Video and audio on demand Raouf Boutaba 49

50 Other Distributed Systems Embedded systems Avionics control systems Flight management systems in aircraft Automotive control systems Mercedes S- Class automobiles these days are equipped with 50+ autonomous embedded processors Connected through proprietary bus- like LANs Network File Systems Architecture to access file systems across a network Famous example Network File System (NFS) originally developed by SUN Microsystems for remote access support in a UNIX context Raouf Boutaba 50

51 DSs vs. Computer Networks Computer Network the autonomous computers are explicitly visible (have to be explicitly addressed) Distributed System Existence of multiple autonomous computers is transparent However Many problems in common, In some sense networks (or parts of them, e. g., name services) are also distributed systems, and Normally, every distributed system relies on services provided by a computer network. Raouf Boutaba 51

52 Challenges in Designing DSs There are a number of challenges found in building distributed systems Heterogeneity Openness Security Scalability Reliability Concurrency Performance Flexibility Transparency Raouf Boutaba 52

53 Heterogeneity Applies to: Networks Computer Hardware Operating Systems (Compare UNIX Socket & Winsock calls) Programming Languages Some Approaches: Middleware (e.g., CORBA): transparency of hard- and software and programming languages heterogeneity Mobile code (e.g., Java): transparency of hard- and software and programming languages heterogeneity through virtual machine concept Raouf Boutaba 53

54 Openness Historically, computer systems were largely closed UNIX broke the mold for OS IBM PC broke the mold for hardware Openness: The characteristic that a system can be extended in various ways Hardware extensions Software extensions Approach: Adherence to standard interfaces Raouf Boutaba 54

55 Security Security is a huge issue in computing in general, but even more so in distributed computing Communication Distributed Resources Infrastructure Attacks Files/resources must be protected from unauthorized usage Approach Privacy Authentication Availability Raouf Boutaba 55

56 Scalability Distributed systems operate at many different scales Two workstations and a file server The CS computers Most current distributed systems designed to work with few hundred CPUs. Future systems will be orders of magnitude larger Computers in the Internet Date Computers 1979, Dec , July 130, , July 56,218,000 Computers Vs. Web servers in the Internet Date Computers Web servers Percentage 1993, July 1,776, , July 6,642,000 23, , July 19,540,000 1,203, , July 56,218,000 6,598, Raouf Boutaba 56

57 Scalability Distributed systems operate at many different scales Two workstations and a file server The CS computers Often the more important question is not can you scale, but can you scale well Consider the Internet Most current distributed systems designed to work with few hundred CPUs. Future systems will be orders of magnitude larger. Solutions for 200 machines will fail miserably for 2,000,000. Raouf Boutaba 57

58 Guiding Principles Often the more important question is not can you scale, but can you scale well Solutions for 200 machines will fail miserably for 2,000,000 Designers should try to avoid the potential bottlenecks: Centralized components (e.g., A single mail server for all users) Centralized tables (e.g., A single Directory Name Server) Centralized algorithms (e.g., Routing based on complete information) Designers should provide correct resources dimensioning E.g., IP addresses: from 32 to 128 bits Raouf Boutaba 58

59 Decentralized Algorithms The following characteristics distinguish decentralized algorithms from centralized ones: No machine has complete information about the system state (no Global state) Machines make decisions based only on local information Failures on one machine does not ruin the algorithm There is no implicit assumption that a global clock exists Raouf Boutaba 59

60 Reliability Take a distributed system with 4 file servers, each with a 0.95 chance of being up at any instant, the probability of all 4 being down simultaneously is = , so the probability of at least one being available is , far better than that of any individual server Leslie Lamport once defined a distributed system as One on which I cannot get any work done because some machine I have never heard of has crashed Raouf Boutaba 60

61 Aspects of Reliability Two Aspects: Availability: Fraction of time that the system is usable Fault tolerance: The ability of the distributed system to mask failures, i.e. hide from the users Can be enhanced by: A design that does not require simultaneous functioning of a substantial number of critical components Redundancy, by replicating the key pieces of hardware and software. Raouf Boutaba 61

62 Hardware Redundancy Two computers are employed for a single application, one acting as a standby Very costly, but often very effective, solution Redundancy can be planned at a finer grain Individual servers can be replicated Redundant hardware can be used for non-critical activities when no faults are present Redundant routes in network Raouf Boutaba 62

63 Software Redundancy Software must be designed so that the state of permanent data can be recovered or rolled back when a fault is detected Transaction processing Dilemma: The more copies are kept, the better the availability, but the greater the chance that they will be inconsistent, especially if updates are frequent Raouf Boutaba 63

64 Fault Handling Detection May be impossible Masking Retransmission Redundancy of data storage Tolerance Exception handling (e.g., timeout when waiting for a Web resource) Raouf Boutaba 64

65 Concurrency Concurrency in a distributed system does not necessarily mean concurrency within a single program Many users invoke similar commands Many different server processes may be running Consistent scheduling of concurrent threads (so that dependencies are preserved, e. g., in concurrent transactions) Avoidance of deadlock problems Synchronization, of course, is a problem Raouf Boutaba 65

66 Synchronization It is impossible to get all the clocks exactly synchronized! Any algorithm that starts out with: At precisely 12:00:00 all machines shall note the size of their output queue will fail On a single LAN, with considerable effort it may be possible to get all clocks synchronized down to a few milliseconds But the larger the system, the larger the uncertainty Raouf Boutaba 66

67 Performance Building a distributed system will not win you any prizes if it is as slow as molasses Performance metrics Response time Throughput (number of jobs per hour) System utilization Amount of network capacity consumed Performance Problem Exchange of messages is typically slow (about 1 msec RTT on a LAN, mainly due to protocol handling) Raouf Boutaba 67

68 Way Out Reduce number of messages to optimize the performance! Dilemma: Running many activities in parallel on different processors allows a performance gain, but requires sending many messages Pay considerable attention to the size of computations Fine-grained parallelism: large # of small computations, interact highly with one another -> Cause trouble Coarse-grained parallelism: large computations, low interaction rates, and little data -> A better fit Raouf Boutaba 68

69 Flexibility Structure of distributed systems: 2 schools of thought Monolithic Kernel Microkernel User Monolithic Kernel User Microkernel File Server Microkernel Directory Server Microkernel Process Server Microkernel Includes: files system Directory Process mgmt Net Microkernel provides 4 minimal services: An interprocess communication mechanism, Some memory management, A small amount of low-level process mgmt Low-level input/output Raouf Boutaba 69

70 Monolithic Kernel vs. Microkernel Advantages of Monolithic Kernel: Performance: trapping to the kernel and doing most work locally is faster than sending messages The reigning champion Advantages of Microkernel: Highly modular: well-defined interface to each service, equally accessible to every client, independent of location More flexible: easy to implement, install, and debug new services; users are free to write their own services A simple example: a distributed system with multiple file servers (MS-DOS, UNIX, ); users can use either or both The up-and-coming challenger Raouf Boutaba 70

71 Transparency Access Location Concurrency Replication Failure Mobility Parallelism Scaling Local & remote resources accessed in the same way Remote resources accessed without knowledge of their location Concurrent use without interference Multiple instances are used without knowledge of the replicas by the users Enables the concealment of faults Resources & clients can be moved within a system without affecting the operation of users/programs Activities can happen in // without user knowing Allows system and applications to expand in scale without change to the system structure Raouf Boutaba 71

72 DS Software Architecture Distributed Systems are foremost highly complex software systems Nortel Networks DMS- 100 switch: million lines of code, 3000 software developers, 20 years life cycle to date. Motorola: 20% of engineers produce hardware, 80% produce software Subject to all kinds of software engineering problems Architectural paradigms pertinent to distributed systems Layers Client- Server Raouf Boutaba 72

73 Layers Breaking up the complexity of systems by designing them through layers and services layer: group of closely related and highly coherent functionality service: functionality provided to a superior layer Examples Layer Layer n+1 n n service Layer n (n-1) service Layer Layer n-1 n Operating System (Kernel, other services) Computer network protocol architecture Raouf Boutaba 73

74 Typical Layering in DSs Applications, services Middleware Operating system Platform Computer and network hardware Raouf Boutaba 74

75 DS Layers Platform Windows NT / Pentium processor Solaris / SPARC processor Middleware: Achieve transparency of heterogeneity at platform level Achieve communication and resource sharing " e. g., remote method invocation Examples: CORBA (OMG) DCOM (Microsoft) ODP (ITU- T/ ISO) Java Remote Method Invocation (Sun) Note: Not all communication related functions can be abstracted Raouf Boutaba 75

76 Computational Models Client-Server Client invocation invocation Server result Server result Client Key: Process: Computer: Client: Process wishing to access data, use resources or perform operations on a different computer Server: Process managing data and all other shared resources amongst servers and clients, allows clients access to resource and performs computation Interaction: invocation / result message pairs Raouf Boutaba 76

77 Variant of Client-Server Model A service provided by multiple servers Service Client Server Server Client Server Examples: Many commercial web services are implemented through different physical servers Motivation Performance (e. g., cnn. com, download servers, etc.) Reliability Raouf Boutaba 77

78 Variant of Client-Server Model Web Proxy Server Client Server Proxy Server Client Server Render replication/ distributedness transparent Caching Proxy server maintains cache store of recently requested resources Raouf Boutaba 78

79 Variant of Client-Server Model Mobile Code Code that is sent to a client process to carry out a specific task Examples Applets Active Packets (containing communications protocol code) a) Client request results in downloading of applet code Client Applet code Web Server b) Client interacts with the applet Client Applet Web Server Raouf Boutaba 79

80 Variant of Client-Server Model Mobile Agents Executing program (code + data), migrating amongst processes, carrying out of an autonomous task, usually on behalf of some other process Advantages: flexibility, savings in communications cost Virtual markets, worm programs Network Computers Downloads its operating system and any application software needed by the user from a remote file server Applications are run locally but the files are managed by a remote file server User can migrate from one network computer to another since all application code and data is stored by a file server Raouf Boutaba 80

81 Variant of Client-Server Model Thin Clients and Compute Servers Executing windows- based user interface on local computer while application executes on compute server Example: X11 server (run on the application client side) In reality: Palm Pilots, Mobile phones Network Computer or PC Compute Server Thin Client Network Web Server Raouf Boutaba 81

82 Variant of Client-Server Model Spontaneous Networking Internet Gateway Music service Alarm service Discovery service Hotel wireless network Camera TV/PC Discovery services Laptop PDA Guest s Devices Services available in the network Their properties, and how to access them (including devicespecific driver information) Raouf Boutaba 82

83 Peer-to-Peer Model Applications based on peer processes Not Client-Server processes that have largely identical functionality Application Coordination Code Application Coordination Code Application Coordination Code Raouf Boutaba 83

84 Summary Distributed systems consist of autonomous CPUs that work together to make the complete system look like a single computer Selling points: Good price/performance ratios Better match distributed applications High reliability Incremental growth as the workload grows Disadvantages: More complex software Potential communication bottlenecks Weak security Raouf Boutaba 84

85 Summary (Cont) Multiple CPUs can be organized as: Multiprocessors Multicomputers Three rough classes of software for multiple CPUs: Network operating systems Distributed operating systems Shared memory multiprocessors Key issues when designing distributed systems: Transparency, flexibility, reliability, performance, scalability,... Raouf Boutaba 85

86 Summary (Cont) Computational Models Client-Server: Most widely used Variants of Client-Server Model Service provided by multiple servers Proxy servers Mobile code Mobile agents Network Computers Thin Clients Spontaneous networking Peer-to-peer Model Raouf Boutaba 86