50 likes | 265 Views
What we have learned:. Barrelfish:Abstraction vs. Performance trade-offGet hardware designers to think of system's level problems (e.g., protection) when building hardwareSnappleAdapt both to H/W and to workloadOctoPOSNeed abstractions for resource reqs on very different H/W CielMany-core ma
E N D
1. SFMA Wrap-up
2. What we have learned: Barrelfish:
Abstraction vs. Performance trade-off
Get hardware designers to think of system’s level problems (e.g., protection) when building hardware
Snapple
Adapt both to H/W and to workload
OctoPOS
Need abstractions for resource reqs on very different H/W
Ciel
Many-core machines are the new data-centres
3. What we have learned: Parallel Design Patterns
Wrong multi-threading worse than none at all
Shared-Nothing Transactional Memory
Avoiding sharing can lead to better performance
Dataflow transactions in Transactional Memory
Transactional memory can provide an abstraction for incrementally adding data-flow computation
On the Nature of Progress
It’s possible to impose an order on progress conditions
Panel Discussion
Systems researchers have very strong opinions (and don’t always agree!)
4. Thanks due: All authors who submitted
All of the PC
Keynote: Andrew Baumann
The session chairs – Tim Harris & Derek Murray
Panel Members - Maurice Herlihy, Stephan Diestelhorst & Paul Barham
Co-Chairs – Joe Sventek, Tim Harris & Timothy Roscoe
Everyone who attended!
5. Next Steps Seems a lot of interest in this topic
2nd Workshop on Systems for Future Multi-Core Architectures
Co-located with Eurosys 2012? Other conferences?
If you are interested in being involved, talk to myself, Joe, Tim or Mothy
rmcilroy@microsoft.com
Pub!