Tip #1:
Don't spend time regurgitating obvious stuff from the papers! The point of these questions is to
think,
and then to write down what it is that you thought about. I read 20 to 30 write-ups, so the more interesting you are, the better! If you find yourself just repeating a lot of details from the paper, you are going down the wrong path.
Tip #2:
Don't spend time criticizing how they wrote the paper. That is, I don't need to know whether you thought the paper was well written or not. We'll have plenty of time to talk about those types of things later in the class.
04/16:
No questions this week. Just read the papers before class (as always), and work on your projects!
04/08:
When does NFS performance match the performance of a local file system (such as FFS)?
04/10:
Describe two workloads. One that runs much better on AFS than NFS, and one that runs much better on NFS than AFS.
03/31:
No question. Your last break, as it is.
04/02:
The VMWare paper describes how ballooning is used, requiring no OS change to achieve its goal. How would the OS have to change to support an explicit interface instead of this implicit approach?
04/04:
RPC makes a number of shortcuts to simplify its design. Discuss one that is the worst design decision that they make. Why is it so bad, in your opinion?
03/10:
No question for Monday. Read the paper (about web servers), and study for the exam.
03/14:
3/26:
In the Vax/VMS paper, a number of performance optimizations are discussed. Which one is the most useful? Which is the least?
No reading questions this week. However, PLEASE read the papers carefully before class (as always).
Q10, 02/25:
When people talk about RAID performance, they are usually talking about performance when all is working well. How is the performance of a RAID system (e.g., parity-based RAID-5, or a mirrored system) affected when a single disk has failed?
Q11, 02/29:
Writing synchronization code is complicated and messy. Monitors are supposed to help with this. Do monitors make writing synchronization code easy? In what ways do they help? What could be better?
Q8, 02/20:
IRON File Systems are the best. But, you might not think so. Thus, name three reasons you wouldn't want your desktop file system to include IRON techniques such as checksums, parity, and the like.
Q9, 02/22:
Anticipatory scheduling shows us that sometimes, when we wait, we get better choices and thus can make better decisions. Name and describe two other situations in computer systems where this happens.
Q6,
02/13:
02/18:
FFS uses cylinder groups to co-locate "related" files using various heuristics. When do these heuristics do a poor job of grouping related files?
Q7, 02/15:
How does the LFS file allocation policy compare to FFS?
Q4, 02/06:
Disco virtualizes hardware, thus exporting an interface that looks like hardware to the OS. Exokernel kind of does the same thing. How are they different?
Q5, 02/08:
Nooks uses hardware isolation to provide boundaries between the main OS kernel and device drivers, assuming that drivers are more buggy. When might this worsen reliability?
Pilot
Q2, 01/30:
Pilot uses hints to great effect (e.g., even the file system index as built by the scavenger is a hint of sorts). What are the advantages and disadvantages of such an approach?
Q3, 02/01:
Exokernel in many ways repeats the arguments made by Nucleus. How is exokernel different than Nucleus?