Rohan Padhye:

1) Describe what fork consistency is.

Fork consistency is a property that enables multiple users to ensure that they have a consistent view of the global state (e.g. the filesystem) by communicating with each other to verify each other's view of the global state.

2) It turns out it is very hard to guarantee fetch-modify consistency
(sometimes called freshness) without some trust assumption. There are
even some impossibility results. Sketch a simple design for ensuring
 time-bounded fetch-modify consistency, namely, that the server did not fork
 two clients for more than some time bound B.  

[Not sure I understood the question/content to answer this correctly.]

3) What are the main techniques in SUNDR that achieve a smaller amount
of bandwidth and computation as compared to the expensive strawman file system?

To avoid transmitting and storing entire file version histories, SUNDR uses a version vector which only contains hashes of user/group i-tables instead of the data itself.

SUNDR also avoids the use of a global lock that would cause every client to wait when another is performing an operation on the file-system, but ensures that a partial ordering (a happens-before relationship) can be enforced on modifications to the version vectors (through signed hash chains) and conflicts can be detected when they occur.