Thursday, 10 May 2012

2.00 pm - 4.00 pm
(2 hours)

DEGREES OF MRes, MSc, MSci, MEng, BEng, BSc, MA and MA (Social Sciences)

ADVANCED OPERATING SYSTEMS (M)

Answer 3 out of 4 questions

This examination paper is worth a total of 60 marks.

You must not leave the examination room within the first hour or the last half-hour of the
examination.

INSTRUCTIONS TO INVIGILATORS
Please collect all exam question papers and return to School
together with exam answer scripts
1. (a) Digital process control systems generally operate in a control loop, periodically sampling
the state of the system, comparing that with a reference input, performing a control law
calculation based on the difference, and then driving the state of the system using some
form of actuator. Effective control depends on the reference input, correct implementa-
tion of the control law, and on the accuracy of the sampled measurements of the system
state. Discuss the trade-off between the sampling period of the measurements, processor
overhead, and stability of the system. [4]
(b) The execution of process control systems of the type outlined in part (a) can be modelled as
a periodic task with four timing parameters (two of which are optional, and have assumed
default values in many tasks). What are those parameters? [2]
(c) The timing constraints of a system can variously be described as hard or soft real-time.
Explain what is meant by these terms. Discuss why most real systems should be considered
to be on the spectrum between fully hard real-time or completely soft real-time, rather than
at one of the extremes. [4]
(d) A range of algorithms exist that can be used to schedule periodic tasks in a priority-driven
system with a single processor. These include the rate monotonic algorithm and the dead-
line monotonic algorithm. Explain how these algorithms assign priorities to tasks. Discuss
under what circumstances you would prefer one algorithm to the other, and when it makes
no difference which algorithm is chosen. [10]

1 CONTINUED OVERLEAF
2. (a) The jobs in a real-time system may execute as part of a periodic task, or in an aperiodic
or sporadic manner. Describe the behaviour of these three types of task in terms of the
schedule of their release times and their deadlines. [3]
(b) Aperiodic jobs sometimes need to be scheduled as part of a system comprising mostly
periodic tasks executed according to a fixed priority scheduling algorithm. These aperiodic
jobs can be executed using slack stealing, or they can be executed by a periodic server task.
Explain how these two approaches to executing aperiodic jobs work, and highlight their
advantages and disadvantages. [5]
(c) The deferrable server algorithm can be used to implement a periodic server task for the
purpose of scheduling aperiodic or sporadic jobs. With reference to the underlying periodic
task, and the budget consumption and replenishment rules, explain how a deferrable server
works. Discuss how you can determine the schedulability of fixed- and dynamic-priority
systems containing a deferrable server. [7]
(d) The deferrable server is one example of a bandwidth-preserving server algorithm. Another
example is the sporadic server algorithm, which trades algorithm complexity for ease of
proving schedulability. Do you believe the complexity of these servers is worthwhile for
periodic systems that include some aperiodic jobs, or should those jobs be scheduled as
simple, low-priority, background jobs? What about systems of periodic tasks that also
contain some sporadic jobs? Justify your answer. [5]

2 CONTINUED OVERLEAF
3. (a) Some programming languages, for example C++, support allocation of objects both on the
stack and on the heap. Other languages, such as Java, allocate all objects on the heap.
Discuss the implications of this choice for memory management. [5]
(b) One widely used approach to automatic memory management is to add reference counts
to objects, and to update the runtime system to maintain these counts, and reclaim the
memory allocated to an object when the reference count for that object is decremented to
zero. Outline the main advantages and disadvantages of reference counting as a means of
automatic memory management. [5]
(c) Garbage collection and virtual memory can interact in ways that cause extremely poor
system performance. Explain what is happening when this occurs, and why. [6]
(d) Explain why it is difficult to build a garbage collector for the C programming language,
but relatively straight-forward for Java. [4]

3 CONTINUED OVERLEAF
4. (a) Modern computing devices increasingly include multiple processor cores, and these cores
typically operate with a non-uniform memory access (NUMA) model. Outline what is
meant by a NUMA model, and explain to what extent a NUMA system hides the under-
lying memory heterogeneity, and how the effects of the non-uniformity are visible to the
programmer. With reference to the organisation of the underlying hardware, explain why
such systems exhibit non-uniform memory access. [6]
(b) Some computing systems are designed to have separate memory regions available to
each processor, with either no shared memory, or only limited cache coherency. Using
a message-passing communications protocol, it is possible to implement the traditional
abstraction of threads with shared memory across the heterogeneous cores of a processor
that does not have hardware-based cache-coherent shared memory. Discuss the advantages
and disadvantages of faking shared memory in this way, explaining your reasoning. [4]
(c) Rather than using hardware memory protection to isolate processes, some research operat-
ing systems (e.g., Singularity) employ the concept of software isolated processes. Explain
what is a software isolated process, and describe how the isolation between processes is
enforced. Discuss the extent to which you believe software isolation is sufficient to enforce
separation between processes. [10]

4 END OF QUESTION PAPER