SEDCL Retreat 2015 - Industrial Feedback Session

Enjoyed the talks
Two pieces of advice more with presentation format than the work itself
1. Wants to see related work slide Don't spend too much time on it but it would be nice to say these are the points that are related, just for reference later on
2. Wants argument or intuition about correctness of distributed protocols
Some of these presentations give a description of distributed protocol without presenting the key idea of why these

Liked diversity of content
Hope the slides can be available quickly
They need to write reports
Liked the fact that we are pushing latency and scalability (latency matters a lot for them).
Being able to identify use cases [for the low latency] is helpful.
Interesting that it's asked " does latency matter?"
Being able to highlight uses cases where latency makes a difference is good
If you don't care about latency you may care about other things
- Such as cost
- Problem can be run on SSD instead of DRAM you can use a lot less DRAM - may save a lot of cost
- Sometimes tradeoff is not absolute performance but instead efficiency metric
Beautiful spot, Well-run
enjoyed breakouts
- Liked the multiple professors talking

Works on Data center Power and usage efficiency issues
Christos talk very relevant to her.
Some talks today, especially Christos caused a lot of interesting
Enjoyed retreat
In regards to the visions of different degrees of speculativeness, wondering about time frames / time scales for each one of these [projects].

First time at a retreat
Moving into systems domain
Spent a year looking at end-to-end latency issues
He deals with existing issues (how things are) instead of how things should be (what the PlatformLab should aim to do).
Our focus seems to be in the right area
Nice mixture of things; well presented
Excited to see where this will end up
Facebook is focused on year to year issues, rather than 5-10 years down the line
If communication in the DC becomes "free," what would we be able to gain from that?

Works on measuring data centers capacity
Enjoyed the talks and discussions
Only comment is that observation -- people care about operational simplicity as much as about performance improvements and wants us to:
- Highlight performance implications of different designs
- Consider operational complexity as we add abstractions and layers in large scale platforms

He really enjoyed this
Meta-comment
Great to see more cohesive or integrated research agenda going forward
Right now, this is a transition from SEDCL to the Platform Lab
It is understandable that it is a snapshot of projects that have been ongoing
There is a broadening out of the research and the overlap / relationship
Forming a vision for how this could move forward
As it progresses, it will be interesting to see how the vision of the platform lab really comes to fruition
Really liked John's opening talk -- discuss stuff agreed to from Faculty offsite
- Set up 4 main goals / objectives that you identify as problems to solve
Definitely going in the right direction and encourage everyone to continue
Encourage more collaborations.

First time here
It exceeded his expectations
Breakouts
he liked the fact that it happened twice with a sleep in between
Slides if available would be great
Things to improve
- Structure the collaboration to make the interaction more frequent
- Use case driven design
- Maybe there's another way to structure a more closer partnership between industry and the faculty
Really liked the opening session
it gave a framework to the platform lab
The direction is very good
Pushing the physical limits is a worthy cause
Wants more framework based presentation (high level ideas and observations distilled)
- Give the big picture more
- Software-defined X
- Trends: Highlight them for industry

Research discussion is very fundamental and maybe a critical issue for the next 5-10 years of technology development
Consider semiconductor costs as part of the impact to technology
Comment: Industry is approaching to the end of Moore's age
Even so, we have to continue to improve cost and performance.

First time here
Liked the spectrum of the research, impressed by the tight coupling of research direction with industry.
Enjoyed Ana and Christos' talks about latency-sensitive services running together with best-effort tasks
Facebook is going after this direction
- Care about latency for web requests and care about failover redundancy so they build a lot of additional capacity
- Utilization is not close to 100% because they prepare for disaster recovery
- There are a lot of CPU resources sitting on the table for best-effort jobs
Facebook will have a new network
- improved network throughput
- reduce oversubscription from 1:133 to 1:13
- More network bandwidth to use
- Neutral to pull data from remote servers instead of push the compute to the data
- Very relevant to what they're looking for

Personally, almost every one of the topics is extremely relevant to VMWare's thinking for next several years. Pleased with selection of topics.
Break-out sessions were very unique and nice
- Gave an opportunity for us to actually discuss the nuances of a particular topic even though only half an hour
- More time would be beneficial
Only comment
3rd time coming here
He'd like to have a higher ratio of interns coming into VMWare