[cadynce] Pre-telecon Rant

[Cross, Joseph]

Gentledudes -

Anybody read the comic strip Sherman's Lagoon? In it, Sherman the shark
counsels his young nephew that only two questions need be asked of
everything he encounters: "Can I eat this thing?" and "Can this thing
eat me?" If the answer is no to both, then the thing is irrelevant.

Regarding the draft presentation, for every bullet on every slide, Carol
will want to know "Can this thing help me?" (i.e., can it make my job
easier or expand my empire), and "Can this thing hurt me?" (i.e., can it
decrease my empire or make me look stupid.)

For many bullets in the draft presentation, the answer is no to both.
E.g., on slide 6, "A multi-dimensional binpacker capable of constraint
satisfaction is at V0.8." Or the beautiful graph of NTP performance on
slide 14. Or on slide 9, we announce an upcoming report that describes
possible paths to solution of the Generalized Sporadic Arrival Model
problem (surely Carol has a long list of problems, but she probably
doesn't have one with that name.)

If we want Carol to back us up in our presentation to RADM Frick, we'll
need to show her crisply what the goal program will deliver that will
help her, and then show seedling results that make it credible that
those deliverables will actually appear.

I recommend focusing on the head of the octopus rather than on its
tentacles. I.e., we should say explicitly that the goal program will
deliver:

1) Design-time tools that generate allocations, estimate their
certifiability by model checking,  automatically test those that pass,
and present the test data to the certification board; 
2) Rules that define equivalently-certifiable configurations, such as
interchanging apps in two blades with the same number of cores on the
same shelf;
3) Run-time tools that will select a certified configuration from the
list that 
   a) Fits on available hardware
   b) Optimally supports currently desired missions
   c) Minimizes configuration-transition costs such as shutdowns and
startups.

Then for each item, we should give whatever evidence we have that the
goal program will get there. E.g., the multi-dimensional bin-packer
progress lends credence that we will be able to generate a lot of
potentially certifiable configurations, and the NTP data lends credence
to the idea that we'll get good test data.

Doubtless Carol is smarter than I am, and she'd be able to draw the
lines from what's in the presentation to the benefits to her. But just
to be safe, we should make those connections blazingly obvious.

- Joe