Objections

1. Motivation

1.1. "Software engineering / debugging is trivial"

1.2. "The PC thought that solving the ultimate problem of finding the bug will be very difficult. "

2. Related Work

2.1. How is this different than general distributed systems debugging? i.e., No discussion of D3S, WiDs checker, et al.

2.2. Didn't frame well against "The criticial transition"

2.3. "Really a variant of common SE techniques"

2.4. Compare to program slicing

2.5. Didn't frame well against x-trace

3. Content

3.1. Not enough detail provided on approach

3.2. System isn't all that impressive

3.3. The paper doesn't show real SDN deployments, only simulation

4. Evaluation

4.1. Scale

4.2. Not demonstrating helpfulness for programmers

4.3. Not enough bugs

4.4. Boring bugs

4.5. No eval of the extent non-determinism mitigation tricks helped

4.6. Don't demonstrate that replay/dd work on very long traces

4.7. Are MCSes actually minimal?

5. Practicality

5.1. Fuzzing

5.1.1. Catching irrelevant problems

5.1.2. Not being able to catch relevant problems

5.2. Too much work to achieve sufficiently "real" simulation

5.2.1. Simulating middleboxes

5.2.2. Simulating legacy devices

5.2.3. Simulating traditional routing protocols

5.2.4. Simulating vagaries of h/w failure modes

5.2.5. Not being able to reproduce asynchrony in a single process

5.3. Controller instrumentation difficult

5.4. Non-determinism considered fatal

5.4.1. "Exponential probability of triggering bug" via retries assumes i.i.d, which doesn't hold