CST.93.7.

COMPUTER SCIENCE TRIPOS Part II

Tuesday 1 June 1993 1.30 to 4.30

Paper 7

Answer five questions.

Submit the answers in five separate bundles each with its own cover sheet.
Write on one side of the paper only.

1 Distributed Systems

A distributed software system follows the client-server model. The microkernel

on which it is based supports multi-threaded processes. A remote procedure call
(RPC) package is used for client-server interactions. The RPC system runs above
an unreliable, datagram-based communications service.

(a) Explain how timers may be used in the RPC protocol to achieve client-server
synchronisation. [10 marks]

(b) Discuss how the RPC system may support the location of remote procedures.
[7 marks]

(c) Discuss the requirements on the RPC system that follow from the use of multi-
threaded processes. [3 marks]

2 Distributed Systems

A distributed computation may involve related operations on a number of objects

which reside at different nodes of a distributed system.

(a) Explain why the concept of transaction is suitable for modelling such a
computation. [6 marks]

(b) Explain what is involved in committing a transaction in a distributed system.

[14 marks]

1 [TURN OVER
CST.93.7.2

3 Comparative Architectures

A revisionist view might be that a processor design is RISC if every transistor pays
its way in terms of global system performance, i.e. the design is in some sense a
local optimum. Explain how this view relates to the original definition of RISC.
Can a CISC instruction set be RISC under such a definition? [4 marks]

Given a certain silicon budget (e.g. 106 transistors) compare and contrast, under the
above criterion, alternative ways of spending the excess over the (say) 104 transistor
cost of a simple load-store one-accumulator machine. You might find it helpful to
consider the various features and instructions of common processors. [12 marks]

This budget is sufficient to build most, but not all, of a VAX. Which parts would
you omit and how could you arrange to execute full VAX code? [4 marks]

4 Comparative Architectures

Compare and contrast possible implementations of instructions which

(a) load a single byte from memory [8 marks]

(b) store a single byte in memory [4 marks]

(c) move a sequence of n bytes from one address to another [8 marks]

For each implementation, indicate briefly whether it conforms to the RISC

philosophy, its effect on pipelining, and its likely efficiency compared with any
alternatives.

2
 CST.93.7.3

5 Specification and Verification of Hardware

You are given components MUX, REG and COMP whose behaviour is defined by

MUX(sw,in1,in2,out) = (∀ t. out t = (sw t → in1 t | in2 t))

REG v (in,out) = (out 0 = v) ∧ (∀ t. out(t+1) = in t)

COMP(in1,in2,out) = (∀ t. out t = (in1 t < in2 t))

Use these to implement a device MAX that satisfies the specification

MAX(in,out) ⇒ (∀ t. out t = M ax in t)

where the function M ax is defined by

(M ax in 0 = in 0) ∧
(M ax in (n+1) = (M ax in n < in(n+1) → in(n+1) | M ax in n))

[10 marks]

Prove that your implementation meets its specification. [10 marks]

6 Specification and Verification of Hardware

Describe the simple switch model of CMOS transistors. [5 marks]

Draw a circuit diagram of a CMOS inverter and give a proof that it is correct in
the simple switch model. [2 + 3 marks]

Describe the unidirectional model of NMOS transistors. [5 marks]

Draw a circuit diagram of an NMOS inverter and give a proof that it is correct in
the unidirectional model. [2 + 3 marks]

3 [TURN OVER
CST.93.7.4

7 Numerical Analysis II

State a recurrence formula suitable for evaluating the sequence of Chebyshev

polynomials {Tn (x)} for an argument x. What are the starting values? [2 marks]

The error in Lagrange interpolation can be expressed in the form

n
f (n) (ζ) Y
f (x) − Ln−1 (x) = (x − xj )
n! j=1

for a suitable function f (x). Suggest a choice of the interpolation points {xj } which
tends to minimise this error over the interval [−1, 1]. [3 marks]

Hence justify and explain the method of economisation of a power series.

[5 marks]

In what sense is an economised power series a best approximation? [2 marks]

Suppose Pn (x) is a polynomial formed by truncating a power series after the term
in xn . Perform an economisation of the truncated power series

x2 x4
cosh x ' P4 (x) = 1 + 2! + 4! [5 marks]

Given that the maximum error in P4 (x) over [−1, 1] is approximately 0.0014,
compare the error in your economised polynomial with the error in P2 (x).
[3 marks]

4
 CST.93.7.5

8 Numerical Analysis II

If B is a real symmetric n × n matrix such that z̄T Bz > 0 for any complex vector
z, prove that any eigenvalue λ of B is such that λ > 0. Hence prove that the
eigenvalues of AT A, where A is any real square matrix, are real and non-negative.
[3 marks]

Let P, Q be real n × n matrices and let kPk22 denote the maximum eigenvalue of
PT P. State Schwarz’s inequality for kPQk2 . Explain how this is modified if Q is
replaced by a vector of n elements. [3 marks]

Derive the condition number K for solution of the equations Ax = b. Hint: start
by setting e = x − x̂ where x̂ is an approximate solution. [5 marks]

Describe the singular value decomposition

A = UWVT

and explain how you would use it to solve the n equations Ax = b when W has
rank n. [5 marks]

How may the singular value decomposition help in solving the equations Ax = b
when A has rank < n? Use the case n = 4, W = diag{1, 10−3 , 10−20 , 0} to
illustrate your answer. (You may assume that machine epsilon ' 10−16 .)
[4 marks]

9 Graphics II

When scan-converting items for display, a Z-buffer is sometimes used to avoid some
sorting. Outline its operation and limitations. [12 marks]

The use of an A-buffer will improve matters. Explain why. [8 marks]

5 [TURN OVER
CST.93.7.6

10 Semantics

An imperative language has boolean expressions be, integer expressions ie, and
commands C, whose abstract syntax is specified by:

ie ::= n | X | ie + ie | ie − ie
be ::= b | ie = ie
C ::= skip | X := ie | C ; C | if be then C else C | while be do C

where b is true or false, n is any integer, and X ranges over a fixed set of variables.
Describe the operational semantics of the language in terms of inductively defined
evaluation relations

be, S ⇒ b ie, S ⇒ n and C, S ⇒ S 0

where S and S 0 are integer-valued functions on the set of variables. [5 marks]

In what sense are these evaluation relations deterministic? What is meant by the
assertion that two commands are semantically equivalent? [3 marks]

For any choice of be, C and C 0 , which of the following pairs of commands are
semantically equivalent and which are not? Justify your answer in each case.

(a) ((while be do C) ; C) and (if be then ((while be do C) ; C) else C)

(b) (C ; (while be do C)) and (if be then (while be do C) else C)

(c) (while be do (if be then C else C 0 )) and (while be do C)

[12 marks]

6
 CST.93.7.7

11 Additional Topics

Describe, for a competent programmer, the decoding of a Lempel-Ziv data

compression code. Assume that the input is a stream of positive integers.
[10 marks]

Construct a Huffman Code for a seven-letter alphabet, given the following

approximate frequencies:

a b c d e f g
192 58 105 71 315 131 0 [4 marks]

Give an error-correcting code and explain how it works, excluding the simple
repetition code. [3+3 marks]

12 Additional Topics

Explain the distinction between Public- and Secret-Key cryptosystems. [4 marks]

Name an example of each type. [1+1 marks]

Is a One Time Pad a Public- or Secret-Key system? [1 mark]

Discuss the advantages and disadvantages of One Time Pads as compared with
alternative systems. When might One Time Pads be used and when should they
be avoided? [13 marks]