Date: Sat, 10 May 1997 19:58:49 -0400 From: Michael Alwan <alwan@rma.edu> To: chaos@tgci.com Cc: questions@freebsd.org.shawn@luke.cpl.net Subject: Re: advantages of symmetric processing Message-ID: <3.0.1.32.19970510195849.006c38d8@rma.edu> In-Reply-To: <199705102241.PAA21385@train.tgci.com>
next in thread | previous in thread | raw e-mail | index | archive | help
At 03:13 PM 5/10/97 +0000, you wrote: >> Date: Sat, 10 May 1997 17:25:46 -0400 >> To: questions@FreeBSD.ORG >> From: Michael Alwan <alwan@rma.edu> >> Subject: advantages of symmetric processing > >> To anyone interested: >[snip] > > >> On the other hand, there are other socket 7 processors like the AMD K6 with >> all the 32 bit optimizations, faster clock speeds, and lower prices than >> newer Intel stuff. The upgrade path (beyond 266 MHZ) seems unpredictable, >> and as far as I know, it can't be multiprocessed. > >Sure it can with the appropriate motherboard. > >> >> Here's my question. All other things being equal (version of operating >> system, system bus speed, amount of ram, kernel configuration, disk speed, >> etc.) which runs a given app faster--symmetric multiprocessing or faster >> clock speeds with one processor? > >All things being equal, if you double your clock speed you will double >your on-chip processing capability. Your on board i/o capacity stays >the same. > >If you double your processor, you may or may not double your >processing capability. Assuming you're looking to increase the >performance of a single application, as opposed to increasing >performance of a system running multiple applications (or instances >of same), the application must be written to take advantage of SMP. > >Assuming such an application, it *may* also take better advantage of >disk i/o and memory i/o. Maybe. > > >> Compare, say, two 120 MHZ Pentiums to one 200 MHZ Pentium Pro. Do >> something CPU-intensive in a database. Which will come out ahead? What is >> the break-even point? I'm less likely to be networking or using my machine >> as a server and more likely to be image-processing or DTP or using a database. > >Depends on the database ( or app) and what you're doing with it. A >single threaded app will run faster on the 200 Mhz PPro. A >multi-threaded app *might* take better advantage of the 2-120s, >although you are comparing apples and oranges. The PPro can really >scream with certain apps. > >> >> I haven't been researching this for long, but everything I've read seems to >> suggest adding a second processor doesn't increase the speed of a given >> operation more than 50%. There is a lot more information about the impact >> of cpu clock speeds, obviously because most people have one cpu. I also >> realize that at the rate new hardware and software is coming out, any >> prediction now might make no sense in 2 months. I'm just looking for the >> most bang for the bucks I have now. >> >> If anyone has any answers experience, or opinions, I'd be really >> interested. It's hard to get a straight answer from a vendor. >> >> Thanks, >> >> Michael > >You really need to define what you want to do and what application >you're using to do it with before you can get a real grasp of this >question. > >Ciao, > >Riley > Riley, Alec, Shawn: Thanks for the input. You folks brought up just the point I hoped to hear something about--multithreading and apps written to take advantage of multiprocessing. Since I do nothing critical (yet) with FreeBSD, I hesitated to get too specific about apps running on other platforms, being too removed from questions@freebsd.org. On Windows NT Workstation I might be running Adobe Pagemaker and Photoshop--not necessarily at the same time. I think the latest Photoshop IS multhreaded, but not Pagemaker. I asked the question here because, in general, people really know what they're talking about and most "tech support" people know less than I do. Alec Kloss gave me another scenario in which more than one process is running in a given app; in that case, it seems I would benefit from multiprocessing in terms of increased throughput but not in terms of latency per process. And that is, I guess, the answer I was looking for. If, in general, one program runs one process at a time, then a specific cpu type (i.e. PPro vs Pentium) and cpu speed are more important than number of cpu's. I was trying to get a handle on exactly how symmetric multiprocessing worked. About the K6--I'm just parroting what I've been told by vendors of systems, none of whom offer a multi-K6 system. As you suggest, an approriate motherboard will probably be forthcoming soon, though I don't understand why an existing one with the correct voltage wouldn't work, if the K6 decodes all x86 instructions and plugs into a socket 7. Thanks to all--I would be happy to hear anything else you have to say. Michael
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?3.0.1.32.19970510195849.006c38d8>