Re: Hyperthreading hurts 5.3?

From: Scott Bennett (bennett_at_cs.niu.edu)
Date: 01/13/05

  • Next message: Christian Hiris: "Re: gmirror problem on 5.3-R i386" 
    Date: Thu, 13 Jan 2005 02:21:11 -0600 (CST)
To: freebsd-questions@freebsd.org

         On Thu, 13 Jan 2005 02:11:27 +0100 Anthony Atkielski
    <atkielski.anthony@wanadoo.fr> wrote:

    >Scott Bennett writes:
    >
    >SB> I notice that the 5.2.1 boot messages refer to the second core as an
    >SB> AP, which I'm guessing stands for "attached processor". If that
    >SB> guess is correct, then it means that only the first core is able to
    >SB> perform certain functions, and the AP core has to get the first core
    >SB> to do those things for it when it needs them done.
    >
    >AP just stands for "application processor," from what I've seen. My
    >impression from snooping in the code and looking elsewhere is that an AP
    >is just a processor that is halted during system boot. The processor
    >that executes the boot sequence is the bootstrap processor (BSP). Once
    >the boot proceeds far enough to allow synchronization of multiple
    >processors, the other processors (APs all) are started by the BSP.

         Oh, good. That sounds much better than what I was thinking.
    >
    >This doesn't necessarily mean that the BSP is special in any other way
    >outside of startup or shutdown, and hopefully it is not, as that would
    >defeat much of the conceptual purpose behind SMP. I know that on
    >operating systems that insist on keeping one processor special for
    >certain tasks, the speed of this processor often becomes a bottleneck on
    >heavily loaded systems, as it tops out trying to handle all the
    >"restricted" stuff for the other processors and itself.

         Usually that sort of restriction has a basis in hardware. For example,
    IBM's MP mainframes *used to* require that the same processor that started
    an I/O operation be the one that fielded the interrupt(s) upon completion
    of the operation. Some machines also had the main processor/attached
    processor configuration, in which the attached processor had no access to
    the I/O hardware at all, so all I/O handling had to be done by the main
    processor because the AP had no way to do it.
    >
    >SB> What Intel claims is essentially that the HT-enabled CPUs allow
    >SB> snappier responses in interactive processes when a CPU-bound process
    >SB> is running.
    >
    >That I can believe. One of the great advantages to a multiple-processor
    >system is that it's much less likely to bog down if a process decides to
    >hog a processor (unless the process runs multiple threads). I think MP
    >is more interesting for its ability to run completely independent
    >processes or threads than it is for its ability to run multiple
    >threads doing the same thing. Few applications require multiple
    >high-speed processors churning through code all at once.
    >
         My main interest in such things at present is for dividing the workload
    in fluid dynamics and, most especially, geophysical fluid dynamics models.
    Those, of course, do immense amounts of number-crunching with occasional,
    massive bouts of I/O. I want to play around with making a two-threaded
    version of a GFD model (either atmospheric or oceanic) to see what, if any,
    savings there may be in elapsed time by running on both cores vs. just one.
    Such a model would have both threads doing essentially the same things,
    though operating upon different parts of the arrays involved.
         But first, I still have to get a usable FreeBSD system going. :-(

                                      Scott Bennett, Comm. ASMELG, CFIAG
    **********************************************************************
    * Internet: bennett at cs.niu.edu *
    *--------------------------------------------------------------------*
    * "A well regulated and disciplined militia, is at all times a good *
    * objection to the introduction of that bane of all free governments *
    * -- a standing army." *
    * -- Gov. John Han***, New York Journal, 28 January 1790 *
    **********************************************************************
    _______________________________________________
    freebsd-questions@freebsd.org mailing list
    http://lists.freebsd.org/mailman/listinfo/freebsd-questions
    To unsubscribe, send any mail to "freebsd-questions-unsubscribe@freebsd.org"

  • Next message: Christian Hiris: "Re: gmirror problem on 5.3-R i386"