664 Midterm Sample Questions
- Define each of the following terms and
describe how it is reflected in "Application-Controlled
Physical Memory Using External Page-Cache Management" (max 5
sentences each)
- Goal
- Technique
- Metric
- Workload
- Statistics
- Microbenchmark
- Macrobenchmark
- Richard Gabriel's paper "Lisp: Good News, Bad News, How to
Win Big" discussed two alternative schools of thought in
approaching the design of systems. What are they? Select a
paper we covered in which each philosophy can be seen and
explain how it impacts the decisions made in that paper.
- We discussed the unified buffer cache (using memory for
either a file system cache or page cache depending on current
demand) several times in class. If you must choose between
either a BSD Fast File System or LFS for both your files and
to store memory pages, how would you alter the buffer cache
algorithm for each type of filesystem?
- Some researchers have proposed letting multiple computers
share the same hard drive and filesystem by placing multiple
computers on the same SCSI bus. Obviously this sharing is at
a much lower layer than an NFS protocol. Briefly discuss how
you might implement such a system. How could you incorporate
optimistic techniques similar to those discussed by Kung and
Robinson into your system?
- Both "Converting a Swap-Based System to do Paging in an
Architecture Lacking Page-Referenced Bits" and "Virtual Memory
Management in the VAX/VMS Operating System" discuss paging
algorithms. How are they similar? How are they different?
- In its conclusion, the paper "Concurrency Control
Performance Modeling" claims to have evaluated a physically
justifiable model to compare different concurrency algorithms,
including "On Optimistic Methods for Concurrency Control."
They argue their results indicate the optimistic methods do not
perform well. Have Agrawal et al. studied the problem
thoroughly, including all realistic workloads? Are there
differences between the workload Agrawal uses and workloads
where optimistic methods might be appropriate?
- Considering the results presented in "A Comparison of
File System Workloads," would the Sprite LFS filesystem be
good for the envionments studied by Roselli et al.?
- The performance improvements used by many of the papers in both the
File Systems and Virtual Memory portions of the course involved
minimizing the number of occurences of one particular type of
operation. What was it? Pick one paper from each section and
explain how it reduces this type of operation.
- Compare and contrast the fairness and sharing algorithms used in
"Application-Controlled Physical Memory Using External Page-Cache
Management" and "Memory Resource Management in VMware ESX
Server."
- The papers "Application-Controlled Physical Memory Using
External Page-Cache Management," "Scheduler Activations: Effective
Kernel Support for the User-level Management of Parallelism," and
"Safe Kernel Extensions Without Run-time Checking" all try to
give the user control of functions normally considered to be core
kernel functions. How does each handle security and fairness? What
are the performance impacts? Could Necula and Lee's idea be applied
to the problems in the first two papers? How? Would it make a difference?
- "FUSE: Lightweight Guaranteed Distributed Failure
Notification" provides us with a way to detect failure in a
distributed system and "Flashback: A Lightweight Extension for
Rollback and Deterministic Replay for Software Debugging" provides
a way to rollback a process after it encounters an error. Could
these two ideas be combined into a distributed system that supports
rollback? What problems would this solve or not solve? What
additional features or services would be required to make it work?
- "Journaling Versus Soft Updates: Asynchronous Meta-data
Protection in File Systems," "Flashback: A Lightweight Extension
for Rollback and Deterministic Replay for Software Debugging," and
"FUSE: Lightweight Guaranteed Distributed Failure Notification" all
present methods for ensuring consistency. What is consistency
and how does it apply to each of these ideas?
- Both "Myths and Realities: The Performance Impact of Garbage
Collection" and "The Design and Implementation of a Log-Structured
File System" discuss locality, although for different reasons. Why
is locality important to each and what methods can be used to ensure
locality?
Bruce B. Lowekamp