

Study with the several resources on Docsity
Earn points by helping other students or get them with a premium plan
Prepare for your exams
Study with the several resources on Docsity
Earn points to download
Earn points by helping other students or get them with a premium plan
Material Type: Notes; Professor: Bridges; Class: ST: Prog Analy &Mechanization; Subject: Computer Science; University: University of New Mexico; Term: Spring 2004;
Typology: Study notes
1 / 2
This page cannot be seen from the preview
Don't miss anything!


This is a closed-book, closed-notes exam with a total of 75 points. You may not use any external source for answering these questions. Please direct any questions about this exam to Professor Bridges, who may be reached either in person in his office in 345G Farris, by phone at 277-3032, or by email at [email protected]. From 10:00am to 1:00pm, Professor Bridges will be available only by email or cell phone at 363-8798. Turn your exam in to Professor Bridges or the front office by 5:00 PM MST on Tuesday, January 13, 2004. Exams will not be accepted after this time except by prior arrangement with Professor Bridges. Type or write your answers to the stated number of questions in each of the following three sections. Make any reasonable assumptions necessary to answer the question, but be sure to state any assumptions that you make.
Briefly answer 3 of the following 4 questions. Your answer should be no longer than one paragraph.
Provided detailed answers to two of the following three questions. Be sure to state any assumptions you make and to fully justify your answers.
Provide a detailed answer to one of the following two questions. Be sure to state any assumptions you make and to fully justify your answer.
Modern laptops frequently spin down (stop rotation) their hard disks when they have been idle in an effort to conserve battery power. Because spinning up (start rotating again) these hard disks is a potentially expensive operation, the operating system may want to delay writes to disk. Consider two different approaches to delaying having to spin up the hard disk in such a system:
For both of these options, describe a potential implementation, how it would work, and its potential advan- tages and shortcomings. Be sure to address which applications are likely to benefit most from each of these approaches.
4.2.1 Background
Large-scale, long-running scientific computations^1 generally use some form of recovery to tolerate crashes, as failures happen relatively often on large-scale machines. Checkpointing is by far the most common form of recovery in these systems; in this approach, a copy of the state of the entire system (a checkpoint) is periodically written to global stable storage. In the case of a crash, computation can be restarted from the most recent checkpoint. Unfortunately, checkpointing is generally costly due to the amount of data that must be saved to shared storage at each checkpoint. As a result, a variety of techniques for reducing the cost and frequency for checkpointing have been proposed. In this question, you are asked to evaluate the tradeoffs of one such potential optimization.
4.2.2 System Optimization
Assume we have a distributed system where each node is normally diskless but could optionally have non- volatile local storage installed in every node if desirable. We propose optimizing checkpointing performance by logging received messages to this optional local storage; after a failure (a failure entails loss of all volatile state), each machine would recover from the most recent global checkpoint and then use the local log of received messages to roll forward from the checkpoint without having to wait for messages from other machines. (This optimization is normally referred to as message logging.) Evaluate this proposed optimization, making sure to address the following issues:
(^1) Applications spanning hundreds or thousands of nodes and running for days or weeks