Views on CPS Ganesh Why CPS is important

Views on CPS Ganesh

Why CPS is important • Tomorrow’s massively complex digital systems won’t “just work” • Their correctness is a multi-faceted issue – Concurrency problems are largely responsible for the sheer complexity (cf the Mars Pathfinder story) – Concurrency is unlike anything other engineering methods have successfully dealt with – Fortunately there are many solutions in the works to properly manage concurrency – There are things beyond concurrency (e. g. system-wide behavior, stability, ability to predict, security properties, …) that are so brand new that a proper mathematical foundation remains an unmet goal for two reasons • There are too many such proposals • Most proposals are left unfulfilled – “Toss over the fence” hoping someone will use the research is an unworkable proposition

Why Standardization is important • No engineering discipline evolved without – Standardization – Mathematics supporting the standards – Without BOTH in place, we will either be re-inventing the mathematics or be re-inventing the artifacts • Standards must be evolved bottom-up on a “need felt truly” basis – Too many academic standards – Too many standards that serve as grounds on which compromises have been struck between vested interests • When this happens, a standard simply serves as a white flag of truce, but then everyone goes about doing their own separate things!! • Sign of an enduring standard – Comes on the scene at the right time – Its proponents remain faithful to the cause for decades – It gets imitated / adapted for more than it was originally meant!

Are there any “enduring” standards? • With CS so young, it’s too early to tell! • In a way, mathematical logic is an enduring “standard” for human reasoning • Others (e. g. concurrency formalisms) are still not adequately vetted – Some evolved amidst the chaos and impatience that commercial interests impose on CS – Some were purely academic thought experiments of “ideal goals” whereas in most cases those who propose such goals are probably overlooking something serious – People try to forget the golden rule of simplicity • Example of simplicity : using a formalism that is “just right” • One anecdotal example : Ed Lee’s pursuit of stream programming for TWO decades now, and his use of this notation to teach a Robotics class that hacks the Roomba Robot !! • We need lots more such examples! • But there ARE some enduring “mundane” standards (next slide)

Example of a robust standard: MPI • While many say “yuck”, it is often an ill-conceived “yuck” • MPI is the de facto standard of distributed programming • Today’s weather simulation codes written in it can’t easily be replaced for ‘decades’ – Petascale machines will need MPI – Are we planning for exa-scale? If so how? • MPI has become the model for the creation of a multicore communication standard between the CPUs of future multi-core CPUs! (at least one standard that is under active consideration, called MCAPI – http: //www. multicore-association. org – Imitation is the best flattery! – There is something simple and natural in MPI that allowed MCAPI creators to model after it

Yet, one should plan for extrication from the “legacy mire” • Nothing remains “shiny” forever – The assumptions under which MPI was created, the hardware, the compilers, the user base, the user education level, future apps, … all change • What CPS needs can be distilled into TWO proposals – Be developing the replacement standard in a very concerted fashion! • E. g. (maybe) Yelick et. al. ’s work on Global Address Space languages – be trying to demo “MPI replacement” over a prolonged period. • People should be encouraged to port applications risk-free (current resistance comes from not knowing whether GAS languages will be there by the time the apps are built) – Be funding CPS researchers in a “true commitment mode”

True commitment mode • • A group of PIs should be hand-picked They (the whole team) should commit to be wedded mainly to the CPS project for about 4 -5 years There should be active monitoring by a consortium of industrial partners who have vested interest in the positive outcome There must be a massive attempt to experiment with curricula (e. g. hand-picked students studying under innovative new curricula) Without doing the above, we have plenty of “toss over the fence” research already underway – we don’t want more of it Sort of a hybrid of the NSF-style selection, DARPA-style funding, and SRC/GRC style of mentoring !!! This is what the program funding / deliverables model should be. Ideally the proposal also should be SRC-style in terms of proposing achievable goals, and not lofty visions of the DARPA era (what you do is what you say should happen more and more) The vacuum is not in people’s imagination and ideas, but the stamina to run the last mile !!