2012 9 27

###NEXT MEETING

• 9:30 - 11:00 next week
• SOSP straw-man outline.
• HotOS -> SOSP. HotOS paper can be a primality section of the SOSP. Orran and JA to work on next HotOS draft.

###TODO

• BOSCH - Open source VMware middleware layer.
• Move reviewer status into MultiLibOS
• Revisit Barrelfish HotOS paper. What is their approach? How do they provide needed motivation/background in 6 pages?
• Convert current MultiLibOS in whitepaper, put it on the web.

---

##SOSP

• Drive the paper by the notion of a system in constant change. Howto define and quantify the change from hardware, from software. How to implement it.
• Distribution as first class primitive.
• Discuss primitives & how they work. Thing changing dynamically.
• System strength functionality with respect of interesting
• Performance at large scale
• Explicit elasticity (hardware perspective, or app-defined workload)
• Fault Tolerance - is this within scope?
• (concern) the only platform we can run these experiments on is BlueGene. KD is sandbox.
• Multi-node Primitives
• We need micro -benchmarks (AMD benchmarks) and macro benchmarks (BG)
• Demonstrate asymmetry in process and load. (load of front-end web service, CPU load nodes) ####What do we build
• HASH table is self-created / self contained. Best potential example. Critical path.
• Medical Imaging Application to demonstrate scale & elasticity. Hook up to a web server (cool factor)
• Node.js software runtime. Take the distributed/event-driven application language (interface) of Node.JS and integrate it into EBBs. Port Google's V8 js engine (C++) and distribute their "jitcache", install distribution as a low-level primitive we can optimize all layers of the application software. Run micro-benchmarks on AMD.

##MultiLibOS Paper

• Move paper in a direction of a more in-depth discussion of motivation and characteristics driving a new "cloud distributed OS architecture." Discuss MultiLibOS in the abstract.
• Remove MultiLibOS examples, minimize the discussion of specific architecture.
• Need to more clearly discuss the degrees of freedom of the MultiLibOS Possible Direction (from Jonathan & Orran)
• This is the architecture we propose. Why is it an architecture? Otology of the work (roles, entities, members)
• "First class distribution support for kernels"
• Here are the fundamental changes/direction we see happening. What are the research challenges/open questions. This paper seems appropriate for HotOS but could provide common/weak paper overall (i.e., 'we've heard this before…').

##Statuses ###Dan

• ...

###Mark

• Exemplar program build on BlueGene. Toolchain problems.
• Sitdown w/ JA, figure it out. Send emails to Dan.

###Jim

• Resurrected development on SEHASH.
• Simple Shared Array running within an EBB based test suite running on Linux. Building on X86_64.

###Orran

• BlueGene Guy: HPC no role in cloud, because cloud needs the "cheap stuff". Rebuttal: Cloud will be heterogeneous, HPC will be part of a bigger "cloud"
• Working with Dan on ethernet code.