CSE 552 Spring 2019 Problem Set 2
Due Friday, May 10 at 5pm
On page 1, the Corfu paper highlights "enabling new designs that are
impractical on disk or RAM infrastructure."
Is Corfu sensible if its backing store is disk storage instead of
flash? Why or why not?
Is Corfu sensible if its backing store is RAM storage instead of flash?
(ignore durability; assume the RAM is battery-backed)? Why or why not?
When implementing a trusted log with TrInc, you have the choice of
assigning semantic meaning to counters or not assigning semantic
meaning. An example of assigning semantic meaning is (as A2M does) to
say that, in a log representing COMMIT messages sent by a node, the
first $x$ bits of the log entry number represent the view number and the
remaining $y$ bits represent the command sequence number. An example of
not assigning semantic meaning is just having the $n$th message in the
log assigned counter value $n$.
Give one advantage and one disadvantage to assigning semantic meaning to
In Bitcoin, a transaction can be invalid for various reasons,
including representing an attempt to spend a unit of currency a second
time. Miners check validity of transactions before putting them on
the chain, and clients refuse to accept blocks that contain invalid
transactions. Why do they do this, instead of just treating such
transactions as no-ops during execution?
On page 9, the Chord paper says:
"Since the number of queries handled by a node is roughly
proportional to the total identifier space covered by that node,
the worst-case number of queries handled by a node does not change."
What important assumption is embedded in this assertion that's often
false in real workloads? What's a more reasonable assumption, and
how do you predict it would affect a Chord deployment?