[Beowulf] Moores Law is dying
Many of your questions may have already been answered in earlier discussions or in the FAQ. The search results page will indicate current discussions as well as past list serves, articles, and papers.
ariel sabiguero yawelak asabigue at fing.edu.uyThu Apr 9 06:48:14 PDT 2009
- Previous message: [Beowulf] Moores Law is dying
- Next message: [Beowulf] Moores Law is dying
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Hi all, I am not sure that all those GHz are useless in a Desktop. All the users of the Atom processor I know are kind of disappointed.... 1,6GHz and a single core does not seem to be enough to run their preferred OS... They pay that price because they can carry it, but as far as I can understand, they need more. I can do really a lot with that system, but I had to understand -not painlessly- that I am not an average user. Just last year 55000 new security threats were discovered, and thus, their antivirus systems should try to do something about them... don't forget about personal firewalls working at the speed of P2P applications, and all the FX required to run a desktop today. I don't think all the GHz we have are enough for a regular Vista user (9x% of the market of multicore personal systems?). Unfortunately we have not figured out how to use the capabilities of parallel processing for a Visual Studio programmer, thus, 99% (please, add more nines as required) of the applications cannot deal with multiple cores. As we know, a single threaded application today might run even slower than a few years ago. In some way, we "earned" a few years in learning how to go parallel thanks to multicore systems. We can allocate one core for the system, another for the antivirus, another for the application and we still have one more.... but not on most notebooks, and even less for uses whose OS put a limit on the number of cores they are licensed to handle.... What we did? we just "released" all the power of a core to a single-threaded program while using the other cores to "serve" the application in foreground. But we still don't know how to solve general programming things in a parallel system... well, sure we know for certain applications!, we spawn more threads as more connections arrive for a web server or for an IMAP server.... and that can really use a multi processor system. Most of the readers in this list really know how to solve applications in parallel, even beyond my imagination! All the science -in parallel- you do really squeezes our systems, but even though we use COTS systems (and I suffer when you buy equipment from server vendors instead of building from scratch as original Beowulf was about -and Google learned to do), we don't have COTS programmers. Think of the training and know-how required for a skilled parallel programmer. Parallel programming is still done by artists. A good parallel program is a rare piece of art. What about other programs? Well, we need a programmer that has taken, at least, "101 - parallel programming" or "101 - High Performance Computing" lectures, which covers only a tiny subset of the humans writing lines of code. And please remember all those (barely) humans who specify their problem in some sort of 4GL language, that is utterly translated into rules, interpreted and finally executed -singlethreaded- on a processor. Those guys fly high on GHz and always need more, more, and more GHz. Their only way to run faster is to get more GHz I think that, in the end, until we learn how to translate from single threaded programs written in standard programming languages (yes, Visual Basic, Java, C#, and so on) into fine-grained parallel code, we are constrained to pray for GHz to have our systems running faster. Most of IT is based on reuse (otherwise we might have moved from Fortran and DOS...) and we have to reuse single threaded things for the years to come. The users will not be amazed playing some variations of Space Invaders and Pacman on their mobiles for long. regards ariel Bruno Coutinho escribió: > I think that even if they stop scaling down size of desktop processors > due lack of interest in more performance, > someone will continue doing it (even at a much slower rate) for HPC > market. > No matter how much computing power future processors will have, > someone will invent a application that needs more.
- Previous message: [Beowulf] Moores Law is dying
- Next message: [Beowulf] Moores Law is dying
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
More information about the Beowulf mailing list
