View Source

Brent Welch - Google

Is worried the lab is building a bunch of components without an overarching problem, which would really help with the design of an end-to-end big control systems.
Suggests that the the researchers should talk more with industries that have big control systems like SDN's, data center job scheduling, airplane logistics, and FedEx deliveries.
Asserts that testing, simulation, verification, and validation are important.
- What-If analysis without crashing the production system is invaluable.
Suggested research question: What are commonalities that do and do not carry over to other types of systems (consider trains vs. SDNs).
- Additionally, Federations of control systems may be important sooner than you think; coordinating multiple control systems like traffic lights with train schedules. These are entities that individually run big control systems, but need some common API to coordinate.

Appreciates the ambition the lab has to do something large and exciting, as evidenced by the talks and the breakout sessions, but would appreciate if the talks had given more information about the inspirations and related work for some of these systems.
- In particular, he likes how some talks cite their related works or describe how their work builds upon earlier work of others students.
- Would like to see how big projects are inspired and how ideas are formed, so that we may someday come to understand how we can recreate big successes.
Would like to see further progress integrating the projects into an actual platform and more communication between groups.
Would like to see 1-2 keynotes about the past/present/future of that platform.

Would like a faculty member to work on the security and privacy of these big control systems.
It's hard to test large scale control systems and so if you can do this and make it open source, it'd be extremely helpful.
Would nice to have some research into optimizing a data center in respect to power, cooling, and other aspects like network traffic.

Appreciates that his feedback last time was taken into consideration (i.e. seeing more related work and the retreat was broader in scope).
Enjoyed the early-work talks and the vision on big control is becoming clearer and cleaner

Would like to see concrete use cases for big control systems (like Amazon's drone delivery service).

Big control is a very good vision to have for the whole community
In the next couple of months, it'd be a good idea to find a couple of good use cases and drill down on what computation can be centralized and what needs to be pushed to the edge.
The neural network, societal networks, and analytical monitoring talks were interesting.

Pratap Subrahmanyam - VMWare

Would be good to emphasize which ideas are inspired by industry collaboration
He felt that RAMCloud was always a solution looking for a problem and worries that Big Control may also be following that path.
- He would prefer more emphasis on "top-down" research (i.e. one with a high level goal) rather doing "bottom-up" by building small components.

Breakout section has interesting discussions
Feels like some talks on flash applications and machine learning can be immediately be applied to some of NEC's systems.

Enjoyed learning about big control and is eager to talk to their colleagues about how to leverage big control with several vertical applications.

Huawei Attendee

Would like to see how the different (seemingly unrelated) projects stitch together into a big platform in the future.
It would be nice to have more use-cases to help identify the problems and drive the design/architecture.
It would be interesting to see research into hierarchical control planes.