cse 555 protocol engineering

CSE 555Protocol Engineering

Dr. Mohammed H. SqalliComputer Engineering Department

King Fahd University of Petroleum & Minerals

Credits: Dr. Abdul Waheed (KFUPM)

Spring 2004 (Term 032)

Protocol Engineering

Reference: “Communication Protocol Specification and Verification” by Richard Lai and Ajin Jirachiefpattana, Kluwer Academic

Publishers, 1998.

Term 032 1-2-3 CSE555-Sqalli

Protocol Engineering

Development of a systematic methodology to progress from an initial protocol requirement for a system to the realization of an implementation

Concerned with activities involved with design and implementation of a protocol using FDTs during its stages of development

Subset of software engineering


A Protocol Development Methodology(Waterfall model)

User Requirements

Informal Specification

Formal Specification

Protocol Verification

Implementation Development

Conformance Testing

Interoperability Testing

Maintenance

Errors detected

Errors detected


A Protocol Development Methodology (Cont.)

Specification: definition of an object or a class of objects, by way of an FDT.

Informal Specification: user requirements are studied and desired communication services are then specified in plain language. This includes: protocol model, services provided, reliability and handling of errors.

Formal Specification: specification using a formal language. This enables the specifications to be analyzed.



Protocol Verification: process of examining this specification for the presence of various errors that could lead to improper system operation.

Attempt to demonstrate the correctness and consistency of a protocol design.

Correctness: protocol performs according to specification Consistency: protocol’s behavior can be determined Automated protocol verification: the use of computer tools

to verify a communication protocol based on its formal specification (Billington et al. 1985)

If errors are detected, the specification is iterated.



Implementation Development: realization of protocol specifications on a physical system.

Development of a communication software Can be derived manually or automatically from formal

specifications using implementation tools.



Protocol Testing Conformance testing: validation of protocol

implementation by testing. Whether protocol implementation conforms to protocol

specification rules Which defined options are supported by the implementation Protocol is fed with selected set of test input Output generated are observed and checked against

specifications Formal specification may be used as the basis for deriving test

sequences Interoperability testing: validation of the protocol

implementation using implementations from different manufacturers.

Ensures the protocol achieves the purpose of open communication.


Protocol Properties

Usually defined informally Definitions of the properties of a protocol (Sajkowski,

1985): Deadlock Freeness = a protocol never enters a state in

which no further progress is possible Safety = a protocol never enters an unacceptable state Liveness = every state of the system is reachable Boundedness = during the operation of a protocol, the

number of messages in each channel between protocol entities can never exceed the capacity of the channel.

Termination or Progress = completion of required services of a protocol within a finite amount of time (reaches desired final states in terminating protocols or initial state in cyclic protocols)


Protocol Properties (Cont.) Definitions of the properties of a protocol (Sajkowski,

1985): Livelock Freeness = a system will never get into a situation

where non-productive cycles are possible and the exchange of the same messages are performed

Mutual Exclusion = certain actions in both users of a protocol cannot be executed simultaneously (e.g., access to same resource)

Partial Correctness = a protocol is guaranteed to provide a required service if it reaches its desired final state

Absence of Overspecification = lack of redundant parts in protocol specification. E.g., absence of unexercised message receptions.

Completeness = descriptions of the reactions to all possible inputs are included in protocol specifications (i.e., specification if free of unspecified message receptions)


Protocol Properties (Cont.) Definitions of the properties of a protocol (Sajkowski, 1985):

Recovery from failures = after any abnormal situation, a protocol will return to a normal state within a finite amount of time

Fairness = every protocol entity gets a chance to make progress, regardless of what others do

Total Correctness = a protocol is partially correct and additionally reaches its desired final state (in the case of a terminating protocol) or its initial state (in the case of a cyclic protocol), for all allowed sequences of inputs (messages or user commands)

These protocol properties are not necessarily independent from each other

The possession of these properties do not ensure that a communication protocol will be working as expected, but it ensures that the protocol does not enter into an undesirable state

Design and Analysis of Communication Software

Credits: David L. Dill (Stanford University), Patrice Godefroid (Bell Labs), Joost-Pieter Katoen (University of Twaente), and High Integrity

Embedded Systems Group (University of Northumbria)


Peculiarities of Communication Software

Communication software coordinates the information flow between interconnected components in:

Telephone networks Internet Wireless networks (cell phones, pagers, etc.) Distributed databases (e.g., flight reservation systems)

Communication software is widely used Communication software is hard to develop and test


Overview

Main steps of the software development process.

Main tools used for each of these steps in industry today.

More detailed discussion on testing.


Software Development Process The waterfall model

In real-life, steps overlap and have feedback loops


Software Development Steps

Requirements specifications and analysis: Determine customer-visible features, feasibility study,

development costs, and price of the product Determine “what to do” Done by “system engineers”

Design: Determine high-level and detailed design of product that

will meet requirements Determine “how to do it” Done by the “architects”


Software Development Steps (Cont’d) Coding and unit testing:

Produces the actual code that will be delivered to the customer

Test individual modules in isolation Done by “developers” (aks “programmers”)

Integration and system testing Test the integration of individual modules and the whole

system Testing implies running the code Done by “testers”

Delivery and maintenance Deliver the product to the customer and provide

documentation, training, field support, and bug fixes “Product manager” manages the end-to-end process


Development Tools

What are the most common tools used at each step of the development process today in the software industry?

Warnings: The list that follows is not exhaustive Only general-purpose tools are considered, not application-

specific tools (for GUI, web, databases,…). Names of commercial products are used only as examples,

for illustration purposes only.


Tools for Requirements Specification and Analysis

MS Word and PowerPoint!! Done informally Requirements are often imprecise, ambiguous, and

incomplete This can be done partly on purpose…

“Formal Notations”: Use-cases, state machines, message sequence charts,

tables, decision trees, etc. Less ambiguous than English text Enable simple automatic analysis of specification (check

for consistency) Can only cover a subset of requirements In practice, used in conjunction with English text


Tools for Design

MS Word and PowerPoint… With diagrams, tables, state-machines, message sequence

charts, etc.

Modeling Languages (for design and high-level coding) UML, SDL, ObjectTime, VFSM, etc. Less ambiguous than English text Enables automatic analysis Code can be generated automatically of a template of the

implementation


Tools for Coding

Defect tracking and resolution managers: Track problem reports and the status of their resolution Record “history” of the system Also used by testers as well as during maintenance

Version control systems: Controls and coordinates the various versions of the

software (e.g., SCCS, CVS, etc.) Code browsers and editors:

Help navigate through the code Also help navigate the history of the code Help compare different version of the code (e.g., diff)


Tools for Coding (Cont’d)

Compilers Translate high-level source language to lower-level target

language Report syntax errors

Linkers Combine mutually referencing object code fragments Report errors at module interface level

Code reviewers (static analyzers) Examine source code to detect programming errors Provide suggestion on code structure and style (type

checkers, Lint, etc.) Automated tools for detecting semantic errors through

symbolic execution Colleagues !


Tools for Testing

Debuggers: Requires code instrumentation (usually during compilation) Control and examine code execution

Memory analyzers: Detect memory leaks and overflows

Memory leaks (= memory allocated, no more reachable but not freed)

Memory overflow (= access to unauthorized memory address), (unallocated/uninitialized memory, array out-of-bounds,…).

Parse and instrument source or binary code to check properties at runtime


Tools for Testing (Cont’d)

Performance analyzers (profilers) and code coverage tools:

Count number of occurrences of execution of program statements or procedures

Report time spent in each part of the program execution Parse and instrument source or binary code to record

runtime information Languages and platforms for test automation:

Example: expect Capture/replay tools:

Record/replay actions performed during manual testing at standard interface

Example: GUI/web testing


Tools for Testing (Cont’d)

Load Generators: Simulate environment through standard interface Example: network traffic

Test case generation from specification: Generate sets of tests from higher-level specification of I/O

behavior Easier test management with better coverage

Test management tools: Process: help record test plans, track, and report the status

of testing project Code: store and execute test code, compare, and store

results Used by testing organizations only


More on Testing Why test?

“To find errors” “The process of executing a program with the extent of finding

errors.”[Myers,1979]

What is an “error”? “Any problem visible to the end user” Programming errors, conflicts with requirements, unexpected

behaviors, features too hard to use, etc.

When to stop testing? In theory, when full coverage is reached Coverage can be defined versus requirements, formal I/O, code, or

state-space In practice, test until shipment date!


Tools for Testing: Summary

Three main types of tools for testing:

1. Code Inspection: analyses (parses) the code to find programming errors.

2. Code Instrumentation: analyses (parses) source code or binary code and inserts code (such as assertions) to check properties at run-time.

3. Code Execution: help generate, execute and evaluate tests performed by running the code in conjunction with a representation of its environment.


Different Levels of Testing

Level 1: manual testing Most testing organizations Some tests cannot be automated

Level 2: automated testing Automated test execution and evaluation Advantage: automated regression testing

Level 3: automatic test generation Automated test generation from higher-level specs Advantages: easier test management with better coverage


What is Communication Software?

Coordinates the information flow between interconnected components

Each component can be viewed as a reactive system I.e., continually interacts with its environment

Examples: telephone, internet, wireless, etc.


Peculiarities of Communication Software Revisited

Communication software is like any other software Same overall development process and general-purpose

tools Developing communication

Many possible sequences of interactions between components

Coordination problems Race conditions Timing issues

Testing communication software is harder! Traditional testing provides poor coverage

Debugging communication software is harder! Scenarios leading to errors can be hard to reproduce


Why Harder?

Implementation looks non-deterministic due to concurrency and real-time

Related to scheduling and processing speed, respectively “Non-deterministic” means unpredictable Same sequence of inputs does not imply same sequence of

outputs

Fundamentally, parallel composition is not “compositional”:

Given 2 function f(x) and g(x), f(g(x)) is easy to understand


Tools for Dealing with Concurrency Debuggers for concurrent/distributed systems:

Control and track the execution of more than one process/thread

Tools for detecting runtime coordination problems: Detect race conditions at runtime

simultaneous writes in same address Detect coordination problems at runtime

Deadlocks Instrument the execution of the processes/threads while

minimizing the impact on timing Record scheduling information (“trace”) for faithfully

replaying multiprocess scenarios leading to errors. Generate a consistent representation (snapshot) of the

state of a distributed system Example: Eraser, Assure (for Java)


Tools for Dealing with Real-Time

Schedulability analyzers: Analyze a set of real-time scheduling constraints and

generate a schedule if there exists one Constraints are imposed by the architecture and properties

to satisfy (specs)

Worst-case execution time analyzers: Determine worst-case execution time (WCET) of fragments

of code

Performance modeling tools: Analyze performance of an architectural model Uses queuing theory, stochastic processes, etc.


Another Approach: Formal Verification What is verification?

4 elements define a verification framework:

Verification: to check if all possible behaviors of the implementation are compatible with the specification

While testing can only find errors, verification can also prove their absence (= exhaustive testing)

Example of approaches: theorem proving and model checking


Theorem Proving

Goal: automate mathematical (logical) reasoning Verification through theorem proving:

Implementation represented by a logic formula I Specification represented by a logic formula S Does “I implies S” hold? Proof is carried out at syntactic level

This framework is very general Many programs and properties can be checked this way

However, most proofs are not fully automatic A theorem prover is actually a proof assistant and a

proof checker Limited usefulness


Model Checking Model checking is more restricted in scope but is fully

automatic Verification using model checking:

Implementation represented by a finite state machine M (called, state space)

Specification represented by a temporal logic formula f Example: Linear-time Temporal Logic (LTL)

Specify properties of infinite sequences s0, s1, s2, …. of states Temporal operations include: G (always), F (eventually) and X (next) Example: G(p -> Fq)

Does “M satisfies f” hold? (hence the term “model checking”) For LTL, do all infinite computations of M satisfy f?

Proof is carried out at semantic level via state-space exploration


State-Space of A Concurrent System The state space of a concurrent system is a graph representing

the joint behavior of all its components.

Each node represents a state of the whole system. Each path represents a scenario (sequence of actions) that can

be executed by the system. Many properties of a system can be checked by exploring its

state space: deadlocks, dead code,… and model-checking. Systematic State Space Exploration is simple:

easy to understand, easy to implement, easy to use: automatic!

Main limitation: “state explosion” problem! (State-space exploration is fundamentally hard)

Many tools are based on systematic state-space exploration: CAESAR, COSPAN, MURPHI, SMV, SPIN, etc.


Formal Verification Vs. Testing

Model checking (e.g., SPIN) can be very effective in detecting subtle design errors

In practice, formal verification is actually testing because of approximations:

When modeling the system When modeling the environment When specifying properties When performing the verification

Therefore, “bug hunting” is really the name of the game!


Formal Verification Vs. Testing (Cont’d)

Model based testing


Applications: Hardware

Hardware verification is a booming application of model checking and related techniques

The finite-state assumption is not unrealistic for hardware The cost of errors can be enormous (e.g., Pentium bug) The complexity of designs is increasing very rapidly (e.g.,

system on a chip)

However, model-checking still does not scale very well Many designs and implementations are too big and

complex Hardware description languages (Verilog, VHDL, etc.) are

very expressive Using model checking property requires experience


Applications: Software Models

Analysis of software models: (e.g., SPIN) Analysis of communication protocols, distributed

algorithms Models specified in extended FSM notation Restricted to design

Analysis of software models that can be compiled (e.g., SDL, VFSM)

Same as above except that FSM can be compiled to generate the core of the implementation

More popular with software developers since reuse of “model” is possible

Analysis still restricted to “FSM part” of the implementation


Summary: Formal Methods Formal methods are based on applied discrete mathematics

Set theory Logic Languages, analysis techniques, and tools

Applicable to the development of Computer hardware Computer software

Used for: Description

Specifying requirements Precise and unambiguous

Analysis Demonstrate that program function implements design Rigor of mathematics helps develop convincing argument Reduces reliance on human intuition and judgment


Summary: Reality Gap

Formal methods help fill the reality gap


Summary: Filling the Reality Gap

Informally: review, testing Formal methods are not complete solutions

How to fill the remaining space Formal specification of required properties Formal model of the delivered system Formal demonstration that the system model has the

required properties

Is it worth the trouble? Apply only for design phases Apply only for critical components and properties


Summary: Validation Techniques

Wide spectrum of languages and inference rules Process algebras: CCS, CSP Logical methods

Set theory, predicate calculus Temporal logics Rewriting logic: OBJ, Maude

Proving equivalence or entailment Not easy to fully automate

Theorem proving assistants Model checkers

SPIN, SMV, UPPAAL, KRONOS This approach can be fully automated

cse 555 protocol engineering

Documents

protocol design

protocol engineeringreference

protocol sajkowski

protocol model

protocol engineeringdr

initial protocol requirement

implementation tools

formal specifications