OpenBSD Journal

News about SMP development

Contributed by jose on from the multiprocesors dept.

Eduardo Alvarega writes: "Theo de Raadt wrote:

Date: Wed, 17 Mar 2004 02:25:08 -0700
From: Theo de Raadt <deraadt@cvs.openbsd.org>
To: tech@openbsd.org
Subject: SMP development

As some of you already know, we have someone working almost "full time" on making i386 SMP happen by the release after 3.5 (ok, maybe it will be delayed by another release, but I doubt it). That person is funded.

I'd like to bring another person up to doing that, fulltime, and I have someone in mind. If anyone out there is able to fund someone for a few months for this to happen, please get in touch with me for details.

Thanks.

(Comments are closed)


Comments
  1. By Anonymous Coward () on

  2. By Mike () on

    http://secunia.com/advisories/11139/

    Is OpenBSD 3.3 or 3.4 affected?

    Comments
    1. By Juanjo () on

      http://marc.theaimsgroup.com/?l=openbsd-cvs&m=107952658000854&w=2

  3. By kris () on

    Theo, instead of asking someone to fund him, why don't you setup a donation fund, you are more likely to get people, as opposed to asking people to just pay for it.

    Comments
    1. By Anonymous Coward () on

      > Theo, instead of asking someone to fund him, why
      > don't you setup a donation fund, you are more
      > likely to get people, as opposed to asking people
      > to just pay for it.

      making a wild guess here, but it's possibly because he does not want to ask someone to make a full-time commitment to this work unless he knows for certain they will be able to pay their bills.

      Comments
      1. By kris () on

        Completely reasonable of course. Point well taken.

    2. By Brad () brad at comstyle dot com on mailto:brad at comstyle dot com

      I'm so sick and tired of seeing people post the same bullshit like this all the time. The suggestions for special donation funds NEVER translate into anyone actually donating money to the special donation funds.

      Comments
      1. By kris () on

        Brad, calm yourself. Its not the same bullshit. If you just say, 'hey, this guy needs to be paid, someone help', the odds are going to be slim. If you setup a donation fund and advertise it a little more than just a small link on a page, you will probably get more donations. Its all about advertisment.

        Comments
      2. By Anonymous Coward () on

        "I'm so sick and tired of seeing people ..."

        Well, it seemed like a reasonable suggestion, even if experience has shown it doesn't work. In any case, it isn't an either/or situation. You can appeal for a single institution to the funding, and set up fund. Maybe sell a special shirt =)

        Comments
        1. By Anonymous Coward () on

          I like the idea...a swarm of Puffys moving off after having skeletonized a hacker?

        2. By Anonymous Coward () on

          Them OpenBSD hats are nice. ;)

      3. By Anonymous Coward () on

        They don't work because usually such donation funds are setup like crap, in logistics as well as mentality. Why? Because there's a single call, some people donate a little over a short while, then it's forgotten rather quickly.

        Instead, if you make a concerted effort and set up specialized fundS or, say, a special t-shirt section with a t-shirt selection, people would have a greater ability to choose what they want done WITHOUT hunting down the developer(s) to donate specifically to.

        A couple of years ago, I wanted firewire development pushed forward faster. But most developers were still focused on wrapping up USB support. In order for firewire development to get focused on, USB support had to become more solid. Those developers would have more time possibly to work on firewire if they so chose.

        I tried to donate (contacted Theo several times via email as he was listed at the donation contact at the time), got no response, searched through the mailing list archives, found a person working on usb/firewire, contacted him, and ordered him some requested equipment.

        Now, this doesn't sound very hard, but we're talking about something on the order of say, a couple of weeks, for this to occur. Email, wait, no response, email, wait, no response, archive search, web search, contact developer directly via email, wait, correspond over a few days on needs, order equipment.

        iow, WHAT A FREAKIN PAIN IN THE ASS. Wastes developers time, wastes people who want to donate time, and a fair amount of effort (not difficult effort, but you've got to be persistant).

        I would have much rather had a "fund page" "to support SMP, donate here", "to support Firewire support, donate here", "to support 64-bit processors, donate here". Have a disclaimer that leftover funds may be diverted and even there are no assurances of support except that money will be donated to the OBSD project and related efforts. Click, find address, send money order or PayPal payment, be done with it.

        This is really not so different than the OBSD donation web page where equipment gets listed that is wanted.

        Personally, I think this also would give greater feedback to developers on what users want. I know developers may not want to work on less exciting or what they consider less important aspects of the OS. And this does stink of "you pay, it gets done, don't pay, we ignore you" sort of environment some people may have a knee-jerk reaction towards.

        But it's a lot more directed than general donations, buying CDs and t-shirts, etc.

      4. By deichert () deeiche@digit.not on mailto:deeiche@digit.not

        OK Brad, you can't say NEVER here. I managed to scrounge US$405 towards Ken's AIC work, plus another US$127 from a vendor. However it IS a PITA to co-ordinate the US$20-30 donations, fortunately we got a couple of US$100-200 donations also. I can't ever see SMP funded in this manner, these people are OTL.

        I'm planning on posting the list of donaters to the misc@ list once Ken gets the last of the funds.

    3. By Peter Hessler () spambox@theapt.org on http://www.theapt.org

      how about, instead of asking for a donation fund, you donate money and use the "memo" field to say "smp". now they have a special donation fund, and no extra paperwork.

      Comments
      1. By kris () on

        That would work as well

  4. By click46 () on

    ...how much does this cost? ballpark figure...?

    Comments
    1. By Anonymous Coward () on

      Let's see. Assuming 6 months of work, 40 hour weeks and a pay rate of $40/hour:

      26 * 40 * 40 = $41,600

      That is a ballpark.

      Comments
      1. By Chad Loder () on

        Damn, $40/hr is a bit much to pay isn't it?

        Comments
        1. By Anonymous Coward () on

          Not if it's Canadian.

        2. By Anonymoose () on

          Half industry standard for someone who needs to have a LOT of knowledge about the guts of the OpenBSD kernel, and an understanding of how to lock and unlock parts of the kernel without breaking everything? Not a bad deal if you ask me.

          Comments
          1. By Anonymous Coward () on

            If $40/hr is half the industry standard, you're suggesting that most kernel developers make $80/hr? I think that figure is high -- that's $160,000/yr. Senior engineers make that much but I don't think you would need someone with 15 years of experience to do SMP. It's not as hard as you think...

            Comments
            1. By Anonymous Coward () on

              Heh, of course it's easy, that's why the SMP code is so stable in so many Unix (and non-unix) operating systems... Check the commits against FreeBSD, and Linux over the last year that deal with SMP. It's easy to get something that brings up the second CPU, but to do locking WELL, and to be sure you aren't introducing a zillion race conditions is a whole 'nother story...

              Comments
              1. By Anonymous Coward () on

                Whoops... there's a missing sarcasm tag after the first sentance... (Obviously?) :)

            2. By Anonymous Coward () on

              $80/hour... on my consulting side, that's about roughly where I start... :-) The more painful and nasty it gets, the more red tape to cut, the higher that number goes.

              $160'000 high? Maybe today... but certainly not 5 years ago. $40/hour is *very* reasonable, that's $80k/year... getting quality people that know what they are doing at that rate can be hard though...

              Comments
              1. By Anonymous Coward () on

                Contract work always goes for a higher per hour rate than steady work, because you aren't able to get a guarenteed 40 hours a week. $40/hr is pretty much right on for a 40 hour a week full time job doing kernel development.

            3. By Anonymous Coward () on

              that's $160,000/yr. Senior engineers make that much

              A senior engineer at a good company? The kind that also pays health insurance and retirement crap?

              I make 60k a year at my work, but i cost my company $140k once all my benefits are added in.

              Paying someone knowledgeable 40/hour to work on SMP while dropping everything else seems like a very, very good deal.

              Comments
              1. By Dan () on

                For that money, lets buy 4 full time programmers in Idia...

                Comments
                1. By Chris Humphries () chris@unixfu.net on http://unixfu.net/

                  Very true :)

  5. By j0rd () mits_rox@OHNOS.hotmail.com on http://j0rd.ath.cx

    Glad to see actual progress being made. I'm planing to pick up a copy of 3.4 when 3.5 comes out, that's my donation.

  6. By Anonymous Coward () on

    Niklas is doing good work already - it'd be great to see someone working with him on this.

  7. By Anonymous Coward () on

    I'd love to take advantage of the second processor on my sparcstation 20. :)

    Comments
    1. By Anonymous Coward () on

      it says i386 you plonker. sparc is totally different.

      Comments
      1. By kris () on

        haha he said plonker ...

        Comments
        1. By Anonymous Coward () on

          indeed i did :-)

      2. By Anonymous Coward () on

        WOW I DIDNT KNOW SPARC IS TOTALLY DIFFERENT THAN I386. DID YOU LEARN THAT IN COLLEGE?

        Jesus Christ why would I say "Any word on SMP for sparc32?" if I didn't realize that they are talking about i386 SMP.

        Posting serious questions about OpenBSD here is such a fucking waste of time.

        Comments
        1. By masouds () masouds@oeone.net on mailto:masouds@oeone.net

          Most of the locking required for SMP is in the C portions of the kernel; The rest, needs some one with enough knowledge. IIRC, Linux got the i386 SMP first, and then the other architectures.
          On a related note, Last time I tried SMP which was couple of weeks ago (yes, call me a cowboy) it failed just after enabling ioapic on doreti() function. Dunno whether it is my motherboard or it is always the same. Gotta try it again soon

    2. By Anonymous Coward () on

      it's i386 SMP

  8. By Anonymous Coward () on

    In one presentation, Asynchronous Multi Processing was proposed (allowing proc #2 to be dedicated to say... crypto).

    Is this being persued in the new multiprocessing work?

    MD

Latest Articles

Credits

Copyright © - Daniel Hartmeier. All rights reserved. Articles and comments are copyright their respective authors, submission implies license to publish on this web site. Contents of the archive prior to as well as images and HTML templates were copied from the fabulous original deadly.org with Jose's and Jim's kind permission. This journal runs as CGI with httpd(8) on OpenBSD, the source code is BSD licensed. undeadly \Un*dead"ly\, a. Not subject to death; immortal. [Obs.]