Middleware and Distributed Systems. System Models. Dr. Martin v. Löwis. Freitag, 14. Oktober 11

Size: px

Start display at page:

Download "Middleware and Distributed Systems. System Models. Dr. Martin v. Löwis. Freitag, 14. Oktober 11"

Angelica Price
8 years ago
Views:

1 Middleware and Distributed Systems System Models Dr. Martin v. Löwis

2 System Models (Coulouris et al.) Architectural models of distributed systems placement of parts and relationships between them e.g. client-server, peer-to-peer Fundamental models formal description of properties common to all architectural models addresses correctness, reliability, and security Selected drawings taken from Coulouris, Dollimore and Kindberg Distributed Systems: Concepts and Design, Edition 4, Pearson Education 2005 System Models Middleware and Distributed Systems 2

client-server, peer-to-peer Fundamental models formal description of properties common to all architectural models

3 Architectural Models placement of components across a network of computers define useful patterns for the distribution of data and workload interrelationships between components functional roles, patterns of communication abstraction: server processes, client processes, peer processes variation of models, e.g. for client-server architecture define mobile code to have some part of the application run on the client support mobile clients to allow matching of clients and servers dynamically System Models Middleware and Distributed Systems 3

processes, peer processes variation of models, e.g.

4 Software Layers Pearson Education 2005 Application Services Middleware Operating System Computer and network hardware }Platform System Models Middleware and Distributed Systems 4

5 5 Protocol Layers: OSI-RM Application Presentation Session Transport Network Data Link Physical Application Protocol Presentation Protocol Session Protocol Transport Protocol Network Protocol Data Link Protocol Physical Protocol Application Presentation Session Transport Network Data Link Physical System Models Middleware and Distributed Systems

Protocol Network Protocol Data Link Protocol Physical Protocol Application

6 Client - Server Model Client: consumer process, uses remote services / information Server: provider process, offers a service / information may in turn be client of another server Pearson Education 2005 Client Server Server Client Node Process System Models Middleware and Distributed Systems 6

in turn be client of another server Pearson Education 2005 Client Server

7 potentially large number of participants often, home users communication pattern varies over time Sharable objects replication necessary to provide resilience in the event of disconnection Peer 1 Application Peer 2 Application Pearson Education 2005 Peer 3 Application Peer 4 Application Peers 5... N System Models Middleware and Distributed Systems 7

disconnection Peer 1 Application Peer 2 Application Pearson Education 2005 Peer 3

8 Variation: Services provided by multiple servers Services provided by multiple servers Service Pearson Education 2005 Server Client Server Client Server System Models Middleware and Distributed Systems 8

Pearson Education 2005 Server Client Server Client

9 Variation: Proxy Servers and Caches Pearson Education 2005 Client Proxy server Web server Client Web server System Models Middleware and Distributed Systems 9

10 Variation: Mobile Code Pearson Education 2005 a) client request results in the downloading of applet code Client Applet code Web server b) client interacts with the applet Client Applet Web server System Models Middleware and Distributed Systems 10

Applet code Web server b) client interacts with the applet

11 Variation: Mobile Agent running program is moved from node to node (with both code and data) security challenge to the server: client gets hold of the entire state of the agent only have non-secret data in the agent security challenge to the client: client node runs arbitrary server-defined code need to establish trust in agent code need to restrict agent's access to local resources (sandboxing) System Models Middleware and Distributed Systems 11

challenge to the client: client node runs arbitrary server-defined code need to establish trust in agent code

12 Variation: Thin Clients Client performs just I/O, no computation Pearson Education 2005 Network computer or PC Compute server Thin Client network Application Process System Models Middleware and Distributed Systems 12

PC Compute server Thin Client network Application

13 s and Objects Set of operations offered by a process is defined by its interface more precisely: set of messages it is able to send and receive often formally specified in interface definitions object-oriented middleware applications: interfaces get implemented by classes System Models Middleware and Distributed Systems 13

specified in interface definitions object-oriented middleware applications:

14 Design Requirements for Distributed Architectures Various objectives for creation of distributed systems sharing of computational resources (e.g. cluster computing) sharing of data sharing of services Performance issues: responsiveness, throughput, load balancing Quality of Service (QoS): reliability, security, performance, adaptability, timecritical data Dependability: correctness, security, fault tolerance, (maintainability) Caching and Replication System Models Middleware and Distributed Systems 14

cluster computing) sharing of data sharing of services Performance issues: responsiveness, throughput, load balancing Quality

15 Fundamental Models Model: abstraction of essential properties of a natural phenomenon, for the purpose of understanding and analysis make explicit all relevant assumptions make generalizations concerning what is possible or impossible Models of distributed systems: Reasoning about Interaction (e.g. communication involves delays) Failure (node and network failure threatens correct operation of system) Security (consider attacks by both internal and external agents) System Models Middleware and Distributed Systems 15

16 Interaction Notion of distributed algorithm: each node has a set of data, and runs a program state of each node not accessible to any other node nodes interchange messages assume that all activity in the system is driven by message reception (may need to consider system boundaries specially) relative speed of nodes should assumed to be unknown performance of network: latency, bandwidth, jitter each node has its own internal clock with specific drift rate - clock synchronization is necessary System Models Middleware and Distributed Systems 16

boundaries specially) relative speed of nodes should assumed to be unknown performance of network: latency, bandwidth, jitter each node

17 Interaction (cntd.) two variants of interaction: synchronous and asynchronous synchronous systems: known lower and upper time bound for each execution step, for each message transmission, and for the clock drift consequence: can introduce a pulsed execution system practically difficult to build, may help in simplifying analysis asynchronous systems: messages can arrive and be sent at any time event ordering: can usually assume no relative order of reception wrt. sending of messages exception: messages sent on an order-preserving channel System Models Middleware and Distributed Systems 17

message transmission, and for the clock drift consequence: can introduce a pulsed execution system practically difficult to build, may help in

18 Interaction: Lamport's "Logical Time" (1978) happens-before relation: e1 happens before e2, iff e1 is executed by the same process before e2, or e1 is a send operation, and e2 is the corresponding receive operation, or there is an e3 such that e1 happens before e3, and e3 happens before e2 logical time: Assign a number L to each event, such that L(e1) < L(e2) if e1 happens before e2 System Models Middleware and Distributed Systems 18

operation, or there is an e3 such that e1 happens before e3, and e3 happens before e2 logical time: Assign a

19 Logical Time (cntd.) send receive Pearson Education 2005 receive X 1 m 1 4 Y 2 receive send 3 m 2 receive Physical time Z receive receive send A m 3 m 1 m 2 receive receive receive t 1 t 2 t 3 System Models Middleware and Distributed Systems 19

20 Failure Model Omission Failures: process or channels fails to perform an operation process omission failure (e.g. crash failure) fail-stop: other processes can detect crash (requires guaranteed delivery of messages) communication omission failures (message drop) Arbitrary Failures (Byzantine failure): anything may happen Timing Failures: synchronous system: activities not completed within pulse real-time systems: activity not completed within promised time System Models Middleware and Distributed Systems 20

failures (message drop) Arbitrary Failures (Byzantine failure): anything may happen Timing Failures: synchronous system:

21 Failure Model (cntd.) Masking failures: reconstruct reliable services on top of unreliable ones through retries, error correction,... Reliability of one-to-one communication: validity (messages are eventually delivered to the receiver) integrity (received message identical to sent one, and no message is delivered twice) System Models Middleware and Distributed Systems 21

22 Security Model securing processes and channels against unauthorized access protecting objects: access rights given to a principal assumption of an enemy (aka adversary), capable of (threat model) sending messages to any process reading and copying any message between a pair of processs enemy may operate either legitimitely-connected node, or illegal node System Models Middleware and Distributed Systems 22

23 Security Model (cntd.) threat to processes: may receive messages sent by enemy may not be able to reliably determine identity of sender server: may not be able to identify principal client: may fall to "spoofing" Threats to communication channels: enemy may copy, alter, inject, or delete messages gain information only intended for the communication partner Other threats: denial of service, trojan horses,... Defeating security threats: cryptography, authentication, secure channels System Models Middleware and Distributed Systems 23

Distributed Systems Lecture 1 1

Distributed Systems Lecture 1 1 Distributed Systems Lecturer: Therese Berg therese.berg@it.uu.se. Recommended text book: Distributed Systems Concepts and Design, Coulouris, Dollimore and Kindberg. Addison