CS261 Fall 2016: Quiz #2
45 points: 15 per question
This exam is closed notes, closed papers, closed interwebs, closed everything.
Please use whatever word processing software that you'd like
to write your answers, but please send me either plain text
(≤ 80 characters per line)
or PDF. I will be grading the quiz blind (that is, I don't want to know
whose quiz I am grading), so please do not put your name anywhere in your
exam. Instead, name your exam lastname.{txt,pdf}.
Please email your quiz to: margo@eecs.harvard.edu.
Here are the papers we've read since the first quiz:
- Waldspurger: Memory Resource Management in VMware ESX Server
- Haggmann: Reimplementing the Cedar File System Using Logging and Group Commit
- Rosenblum: The Design and Implementation of a Log-Structured File System
- Sandberg: Design and Implementation of the Sun Network Filesystem
- Howard: Scale and Performance in a Distributed File System
- Min: SFS: Random Write Considered Harmful in Solid State Drives
- Rumble: Log-structured Memory for DRAM-based Storage
- Schroeder: Experience with Grapevine: The Growth of a Distributed System
- Anderson: Serverless Network File Systems
- Stoica: Chord: A Scalable Peer-to-peer Lookup Service for
Internet Applications
- Rowstrom: Storage management and caching in PAST
- Ghemewat: The Google File System
- Mickens: Blizzard: Fast, Cloud-scale Block Storage
for Cloud-oblivious Applications
- Dean: MapReduce: Simplified Data
Processing on Large Clusters
- Baumann: Shielding Applications from an Untrusted Cloud with Haven
- Tips for Building Distributed Systems
We've read many papers on distributed systems, mostly storage systems. Many
of these papers include principles for building distributed systems that scale.
List five such principles and cite the paper from which each came.
- Tradeoffs in log-structured storage
Part A
We read three papers that use log-structured storage. Which three were they and for what media
was their design intended?
Part B
Each of the systems chose log-structured storage, even though they were targeting radically
different environments.
Explain how the applications of log-structured storage differ between the papers and explain
how the differences reflect the different environments in which the storage was deployed.
- Statelessness versus Stateful Systems
NFS and AFS take exact opposite positions about whether a server should retain state about its
clients.
Part A
State the position taken by each system. Then give one example where each system is simpler, better,
or higher performing than the other, because of their decision regarding state.
Part B
Pick one other distributed system that we read about and indicate whether it is stateful
or stateless. Then discuss how it addresses the challenges (or the negatives) that arise
for a system of that type.
BONUS PROBLEM (2 points)
What is the answer to the "coming to class on time" question?
Paper Survey
After you complete the quiz, please complete
this form .