RTFA: http://sourcemaking.com/antipatterns

40 Reasons Why Software Projects Die

Wow! This is surprisingly detailed, and it’s also totally great stuff.

RTFA: http://www.doe.gov/news/5849.htm

The U.S. Department of Energy’s (DOE) Office of Science today announced that 265 million processor-hours were awarded to 55 scientific projects, the largest amount of supercomputing resource awards donated in the Department’s history and three times that of last year’s award. The projects-with applications from aeronautics to astrophysics, and from climate change to combustion research-were chosen based on their potential breakthroughs in the science and engineering research and their suitability of the project for using supercomputers. These awards will allow cutting-edge research to be carried out in weeks or months, rather than years or decades, giving scientists access to some of the world’s most powerful supercomputers at DOE national laboratories.

What really surprised me was the allocation of time for porting Plan 9 to Blue Gene:

Proposal Title: “BG/P Plan 9 measurements on large scale systems”
Scientific Discipline: Computer Sciences

INCITE allocation: 1,000,000 Processor Hours
Site: Argonne National Laboratory
Machine: IBM Blue Gene/P
Allocation: 1,000,000 processor hours

Research Summary:
This project will provide a new software environment for supercomputers that makes the supercomputer appear to be part of the user’s desktop system, instead of a remote and hard to access external computer. Initial work will expand on the version of Plan 9 that was ported onto BG/L for the FASTOS program by SNL, Bell Labs, IBM, and Vita Nuova. Because the Plan 9 operating system was built with networks in mind, it requires less system administration support than other operating systems. In Plan 9’s environment, files and directory trees can be imported from other machines, and with all resources defined as files or directory trees, sharing resources is greatly simplified. New drivers will be tested for the BG/P tree and torus networks. These new drivers make it possible for native file systems to use, e.g., the tree reduction network to make file systems very efficient. We plan to scale the Plan 9 implementation out to the full machine at ANL and measure performance for applications of interest. We will test all aspects of the Plan 9 environment and modify Plan 9 as needed for this large scale machne, in preparation for future systems with 10 million CPUs.

I’ve reviewed Plan 9 before, and it certainly has architectural features that look great, but it’s anything but a production operating system. Plan 9 corrects many of the errors of Unix. There is no application base whatsoever, although Plan 9 has a complete toolset for building applications.

…but think about this in relation to Unix. In the same way that Unix eventually created a single environment across different minicomputer models, Plan 9 could marry certain “desktop paradigms” with supercomputer-scale resources. It’s not like there aren’t already C libraries for managing parallel computation (which could be said to provide a consistent environment).

It’s that Plan 9 is a single operating system that could run on the slowest Pentium and scale up to the fastest known supercomputer… and between these two computers of dramatically different scale, there would be transparent, file-based resource sharing, provided by Plan 9. It’s not that this is specifically possible, but that this is generally possible.

Plan 9 doesn’t deal with hardware device nodes or networking sockets - these are concepts that are deeply tied to Unix (e.g. Linux-esque) that are deeply confused in the sense that they are not files, but are sometimes mapped onto the filesystem. In Plan 9, everything is a file in a conceptually similar way, so it will be possible to manipulate supercomputer resources using very familiar metaphors.

Totally wild! Is Plan 9 destined to become the universal Cloud Computing OS?