On Jul 10, 2007, at 4:02 AM, Richard L. Hamilton wrote: > ons to access them in different ways are what's really needed? > > But before going crazy, there's also the question not only what is > there > that _could_ be described now or in the foreseeable future, but how > much > of that would it be _useful_ to describe? Aside from simply > documenting > hardware configuration, how much might some program running on a > system > need to know to make best use of the system, or alternatively to > balance > its needs with a desire to be a good citizen and not take unreasonable > advantage? >
For large HPC "capability" systems the answers are, I think, fairly obvious. Everything should be exposed (we can't know a priori what will or won't be useful for programmers willing to cater ot the raw iron). Applications can assume they have the entire machine to themselves and their necessary system infrastructure. Of course, often they will share the machine, but via partitioning ... when you absolutely positively need to solve the worlds largest problems in the shortest wallclock time, timeslicing or otherwise sharing the computational infrastructure is neither desirable nor helpful. Admittedly that is a "niche" but it would be highly desirable to have our observability infrastructure scale from small to huge ;> Naturally most programmers should NOT have to grovel through the lowest level, and most complete observability features for most of their work. Keith H. Bierman keith.bierman at Sun.COM | khbkhb at gmail.com Sun Microsystems Microelectronics Group | sun IM: khb AIM: kbiermank 5430 Nassau Circle East | 650-352-4432 voice+fax Cherry Hills Village, CO 80113 | sun internal 68207 http://blogs.sun.com/khb | 303-997-2749 <speaking for myself, not Sun*> Copyright 2007 ------------------------------------------------------------ NOTICE: This email message is for the sole use of the intended recipient(s) and may contain confidential and privileged information. Any unauthorized review, use, disclosure or distribution is prohibited. If you are not the intended recipient, please contact the sender by reply email and destroy all copies of the original message.
