SLCentral - Your logical choice for computing and technology
Latest Deals   
Navigation
  • Home
  • Search
  • Forums
  • Hardware
  • Games
  • Tech News
  • Deals
  • Prices
  • A Guru's World
  • CPU/Memory Watch
  • Site Info
  • SL Newsletter
    Recieve bi-weekly updates on news, new articles, and more


    Forum home My SLBoards (Control Panel)View the calendar View the members list Read the FAQ Search the forums

    Go Back   SLCentral Forum > Feedback > Content Feedback
    User Name
    Password


    View Poll Results: Did you like this article?
    Yes 43 86.00%
    It was alright 4 8.00%
    No 3 6.00%
    Voters: 50. You may not vote on this poll

    Reply
     
    Thread Tools Search this Thread Display Modes
      #1  
    Old 06-14-01, 06:16 PM
    Dave's Avatar
    Dave Dave is offline
    ActiveTuning Partner
     
    Join Date: Dec 2000
    Location: University Of Maryland
    Posts: 1,873
    Dave is on a distinguished road
    Send a message via AIM to Dave
    Default Fundamentals Of Multithreading

    Post your comments, questions, and suggestions on the Fundamentals Of Multithreading article.

    Link: http://www.systemlogic.net/articles/01/6/multithreading
    __________________
    ActiveTuning - Partner & Director Of Sales & Marketing

    Reply With Quote
      #2  
    Old 06-14-01, 08:12 PM
    Unregistered
    Guest
     
    Posts: n/a
    Thumbs up

    Awesome paper--a must read. Wouldn't be suprised if it were slashdotted.
    Reply With Quote
      #3  
    Old 06-14-01, 08:17 PM
    Dave's Avatar
    Dave Dave is offline
    ActiveTuning Partner
     
    Join Date: Dec 2000
    Location: University Of Maryland
    Posts: 1,873
    Dave is on a distinguished road
    Send a message via AIM to Dave
    Default

    We're submitting it, and also praying Thanks for the comments.
    __________________
    ActiveTuning - Partner & Director Of Sales & Marketing

    Reply With Quote
      #4  
    Old 06-15-01, 12:19 PM
    Dave's Avatar
    Dave Dave is offline
    ActiveTuning Partner
     
    Join Date: Dec 2000
    Location: University Of Maryland
    Posts: 1,873
    Dave is on a distinguished road
    Send a message via AIM to Dave
    Default

    For those if you interested in hardcore techie stuff, we've made a new forum just for you guys that is highly moderated and kept in order and in control. Feel free to head on over there and join the discussions, or start your own:

    http://www.systemlogic.net/boards/f...p?s=&forumid=43
    __________________
    ActiveTuning - Partner & Director Of Sales & Marketing

    Reply With Quote
      #5  
    Old 06-15-01, 12:41 PM
    Unregistered
    Guest
     
    Posts: n/a
    Default Minor mistake

    ----QUOTE----
    There are exceptions to this rule, such as the Mac, where Photoshop has a patch which has multithreaded support for G4's even though no Mac OS below Mac OS X supports multiprocessing. Generally speaking, the computer needs OS support as well as application support to take advantage of multiprocessing.
    -----END QUOTE----

    This is incorrect. Every Mac OS Release since 8.6 supports SMP multiprocessing. Just like in Mac OS X, the Classic (Blue-Box) environment runs in a thread on the 8.6 - 9.2 nano-kernel. This is a normal thread that is sliced with other threads. 8.6 - 9.2 are much more limited in what they can do MP wise, but certain applications can benefit greatly.

    QuickTime, Photoshop and Soundjam MP are some applications that use SMP on Classic Mac OS.
    Reply With Quote
      #6  
    Old 06-15-01, 07:09 PM
    Paul's Avatar
    Paul Paul is offline
    Student-for-life
     
    Join Date: Dec 2000
    Location: College Park, Maryland
    Posts: 1,294
    Paul is off the scale
    Send a message via ICQ to Paul Send a message via AIM to Paul
    Default

    I looked up what you said, and I took for granted what I was told by a few mac zealots....that'll teach me to trust them even when they talk about their own "favorite" system!

    I'll get the article corrected -- they use an interesting (interesting in that it's primitive) smp support...but still cool that it's been around since <8.0. You're completely right!

    Thanks for the feedback,



    Paul
    Reply With Quote
      #7  
    Old 06-15-01, 10:12 PM
    Unregistered
    Guest
     
    Posts: n/a
    Default inaccuracies

    The article contians a number of incorrect assumptions, for instance in the first paragraph it say's that app have to be written specially to improve from MP. this is bull, that is an OS issue and not an app issue, witness some of IBM's os's. Even the lowly OS/2 SMP (10 years old nb.) does not require the app to have been specially written or enhanced for the SMP (there are in fact no OS/2 SMP enhanced apps) since the os handles the thread splitting etc etc etc

    The rest of the article sounds like a collage or a digest of information gained from technical reports rather than from a person with a deep understanding of the subject

    Skip this article and read :

    The design of OS/2 by Deitel & Kogan (a general OS book despite the name)
    Principles of Parallel and Multiprocessing by George R. Desrochers
    Reply With Quote
      #8  
    Old 06-15-01, 11:09 PM
    Unregistered
    Guest
     
    Posts: n/a
    Laughing

    One thing, guys: Tera is/was spelled with one r.
    The MTA architecture still lives as the Cray MTA-2.

    -Wendy Thrash, Cray Inc.
    Reply With Quote
      #9  
    Old 06-16-01, 04:29 AM
    Paul's Avatar
    Paul Paul is offline
    Student-for-life
     
    Join Date: Dec 2000
    Location: College Park, Maryland
    Posts: 1,294
    Paul is off the scale
    Send a message via ICQ to Paul Send a message via AIM to Paul
    Default

    Yeah, we noticed that one too It was spelled right the 1st time...and then fscked up 7 subsequent times. Also, the source I had on it was old, and was not called "MTA"....yes, I indeed should have gone to cray's website to look up more info.

    Another employee of Cray has already contacted me about (both) the issues. My apologies, we will be updating the article accordingly.



    Paul Mazzucco
    Reply With Quote
      #10  
    Old 06-17-01, 12:18 PM
    voodoobunny voodoobunny is offline
    Registered User
     
    Join Date: Jun 2001
    Location: Durban, South Africa
    Posts: 1
    voodoobunny is off the scale
    Question Combining multithreading forms for higher performance?

    To overcome the latency problems with current memory (and ignoring the dangers of dependent processes), what would the performance advantage be of Multi-Grained Threading? i.e. combining FMT with CMT, with threads sharing functional units, then dropping out of the active set when they encounter a cache miss or execute a specific number of continuous instructions.

    Two possible optimisations could be:

    1) an optimisation stage which allow a thread to continue beyond its instruction threshold as long as it uses (exclusively?) functional units that no other instructions would use, or "drops in" instructions which need exclusively a FU which would otherwise be empty);

    2) Using multiple memory units and a deliberately unmatched number of threads (and a lot of threads) you could get cacheless operation similar to the way Tera computer did, and additionally avoid threads inherently competing for the same cache lines or memory units (which the might otherwise do e.g. for data which is aligned). For example, take a system with 64 threads and 17 memory units: in the time taken for a thread to be reactivated, the memory subsystem would have time to *calculate* where in the memory units the data is located, and to retrieve it.

    My guess is that for supercomputing-type applications, where as is mentioned in the article much of the multithreading theory is most useful, this could be used to achieve significant results for moderate increases in production cost, e.g. without using specialised reordering units, a die the size of the Pentium 4 or Itanium could pack a formidable number of register sets and simpler functional units...

    So is anyone out there doing this now?
    Reply With Quote
      #11  
    Old 06-18-01, 12:28 PM
    Paul's Avatar
    Paul Paul is offline
    Student-for-life
     
    Join Date: Dec 2000
    Location: College Park, Maryland
    Posts: 1,294
    Paul is off the scale
    Send a message via ICQ to Paul Send a message via AIM to Paul
    Default

    I'm gonna reply to this, just give me awhile. I'm kinda swamped right now with lots of things, both computer related and not.
    Reply With Quote
      #12  
    Old 12-06-01, 07:32 AM
    Unregistered
    Guest
     
    Posts: n/a
    Default Multithreading article

    Good stuff - thanks. Any thoughts about a scenario where a uniprocessor is augmented with some small number of hardware contexts with the intent of reducing the cycles spent in context switch software?
    Reply With Quote
      #13  
    Old 09-22-03, 01:26 PM
    stevecentral's Avatar
    stevecentral stevecentral is offline
    Webmaster
     
    Join Date: Nov 2002
    Location: UK
    Posts: 1,329
    stevecentral is on a distinguished road
    Default

    Thought I'd read the article in total, very impressive.
    __________________
    SLCentral does it better www.slcentral.com/

    Reply With Quote
      #14  
    Old 09-22-03, 02:40 PM
    Dave's Avatar
    Dave Dave is offline
    ActiveTuning Partner
     
    Join Date: Dec 2000
    Location: University Of Maryland
    Posts: 1,873
    Dave is on a distinguished road
    Send a message via AIM to Dave
    Default

    I think Paul is a little overdue in a reply Or he's been REALLLLLLY busy
    __________________
    ActiveTuning - Partner & Director Of Sales & Marketing

    Reply With Quote
      #15  
    Old 09-22-04, 12:55 AM
    Unregistered
    Guest
     
    Posts: n/a
    Cool

    Good article. How about some more on DSMT and where things are at currently? IE who is using it if anyone?

    I can't help but think when I read such articles that the proc designers are trying to do everything ignoring the compiler writers, when plainly there has to be integration and great leaps will come about when this issue is addressed.

    I have just read a lengthy article about GHz, performance, multi core processors etc. that touches repeatedly on DSMT. The article claims DSMT is the "solution" however how can you solve an issue of what is essentially a compiler generated problem without looking at what can be done in the compiler to support the processor in its implementation. Ditto for the languages used.


    Dave, what are you doing with my car and what have you done to the wheels?

    - Tim

    PS thanks for cleaning it.
    Reply With Quote
    Reply


    Thread Tools Search this Thread
    Search this Thread:

    Advanced Search
    Display Modes

    Posting Rules
    You may not post new threads
    You may not post replies
    You may not post attachments
    You may not edit your posts

    vB code is On
    Smilies are On
    [IMG] code is On
    HTML code is Off
    Forum Jump



    All times are GMT -8. The time now is 10:49 AM.

    Archive - Search Engine Friendly URLs by vBSEO 3.0.0 RC6 © 2006, Crawlability, Inc. Top
    Browse the various sections of the site
    Hardware
    Reviews, Articles, News, All Reviews...
    Gaming
    Reviews, Articles, News...
    Regular Sections
    A Guru's World, CPU/Memory Watch, SLDeals...
    SLBoards
    Forums, Register(Free), Todays Discussions...
    Site Info
    Search, About Us, Advertise...
    Powered by: vBulletin Version 3.0.1
    Copyright ©2000 - 2017, Jelsoft Enterprises Ltd. Search Engine Friendly URLs by vBSEO 3.0.0 RC6 © 2006, Crawlability, Inc.
    Legal | Advertising | Site Info