Follow Slashdot blog updates by subscribing to our blog RSS feed

 


Forgot your password?
Close
typodupeerror
×
Compaq

According to Compaq 53

Posted by Hemos from the what's-their-plan-exactly dept.
Joseph DiLascio has written up a recent speech/interview with a Compaq engineer, dicussing their Linux plans. Given recent news about what's going over there, very interesting read.

According to Compaq

Joseph DiLascio
11 Sep 99


At the last FLUX (Florida Linux Users' Exchange) meeting, the guest speaker was an engineer from Compaq who came to discuss the company's support of Linux on the Alpha processor. With all the discussion on Slashdot recently I thought I'd try to answer some of the questions brought up, because I think the people inside Compaq pushing Linux on Alpha are doing good things.. despite what anyone may think of the company as a whole.


1. What's Compaq's "OS plan"? (Are they dropping Tru64 and moving to Linux.. etc.)
In the server area, at least, they're focusing on three:
  • Tru64 on Alpha for high-end, high-performance, high-etc. applications
  • Linux on Alpha for not-as-high-end-etc.-etc.
  • SCO on x86 for.. well.. the x86 market
In their role as a hardware vendor, they are officially "OS Agnostic", though.

1a. What about Linux on x86? Not likely to be really supported by them any time soon. They make good money with SCO and that's what they'll keep trying to do.


2. What's the deal with this new C/Fortran compiler? (Are they using glibc.. Is it GPLd.. etc.)
A number of customers liked the idea of Linux on the Alpha, but didn't like the performance and/or lack of source-compatibility with other Unices in the area of threading, etc. of GCC and the standard libs[1]. So Compaq figured it would be a good idea to make the development tools consistent between Tru64 and Linux... so they undertook the task of porting their Tru64 compiler and runtimes (no, not glibc) which is what that story on Slashdot was all about. The result is code that more fully exploits the power of the Alpha processor. As far as the GPL... the Alpha people tend to prefer BSD-ish licenses; as far as opening up the compiler and libs.. not right now, but maybe someday.


3. Does Compaq really give a damn about freedom and openness.. or are they just jumping on the bandwagon in the hopes of quick profit?
The impression I got was a bit of both. They've already released source and hardware specifications to get Linux running onAlpha... but they can't necessarily do that for every component of their systems (i.e. video controllers which aren't Compaq's, but come from other companies). Yes, they're out to make money... but, like many of the other big players who have been getting into the Linux game (IBM, etc.), there are people inside the company that like Linux and free/open software for what it is, not just because it'll make them a buck.

3a. Isn't Compaq a slave to MS like so many other hardware vendors? Not when it comes to servers (see #1). I don't think they're looking at putting Linux on desktops at the moment (like Dell seems to be). Compaq and Microsoft do go a long way back, and the relationship apparently is a great one for both of them.


4. Will they kill Alpha if/when IA-64 becomes viable?
The three main goals of the Alpha are:
  • Speed
  • Performance
  • Going Fast
That said, Alpha will be around as long as it's the quickest thing out there... which it probably will be for the foreseeable future.

4a. Will the loss of NT on Alpha affect the viability of the Alpha platform? Not likely. NT on Alpha wasn't selling very well (but that isn't all the fault of the Alpha people *ahem*MS*cough*).


So it looks like Compaq's support is a pretty good thing for the Linux community on the whole (at least, for the adoption of Linux at the high end... those who value freedom (open code, etc.) above all else may not be satisfied). That doesn't mean I'll be buying a desktop PC from them any time soon... but the Alpha is a neat processor and Linux can do neat things on it. Advancing the development of 64-bit Linux programs and Linux in general is something I personally can respect and appreciate.

Final note: The above information may or may not be completely accurate. I base all this on my impressions of the talk and Q&A session with a representative of Compaq who spoke at the last FLUX meeting. If you know of any errors, please let me know.
Also, thanks to the other FLUXers on the mailing list for their thoughts.
[1] - One of the more interesting parts of the presentation was a historical view of the development of "Unix". I personally learned quite a bit about what makes up the thing we call "Linux" today. One thing Linux currently lacks is a 100% complete kernel threads implementation. Pthreads are mostly there but they don't work the same way as on other Unices, and Solaris threads (arguably more "standard" than pthreads among many software vendors) aren't there at all.

Joseph, all Slashdot contributors now receive *free t-shirts* from Copyleft. To get yours, please send your shipping address and shirt size to roblimo@slashdot.org.
This discussion has been archived. No new comments can be posted.

According to Compaq More

According to Compaq

Comments Filter:
  • At a presentation for my computer graphics class, one of the individuals involved in the production of Toy Story (the way he talked he was the master of Pixar but his name wasn't very high in any of the credits) came to talk and show videos. I believe there were 200 suns working on rendering for about two years. They said it was more economical than buying a cray.
    I don't know what an Onyx is or if it was even available in those speeds when the movie was rendered. Toy story was made with some pretty cpu intensive effects, NOT things that can currently be made realtime. Although he did say if moore's law proceeded we would be able to render toy story in real time by... 2005 I think he said. (The presentation was in 1996 I think).
  • that was a deficiency in the earlier libc5.

    glibc aka libc6 has built-in posix threads. so this problem should be solved.
  • A bit more information on COMPAQ, Alphas, and Linux can be found in the UK Register here [theregister.co.uk].

    Especially interesting is the fact that even when IA64 comes out, Alphas are still expected to be running faster.
    --
  • It sucks that I got rated down as a troll - but really, of the innovative things being done with Linux by Compaq, the Itsy should come out near the top of the list.

    Why can't we get a palmtop that runs Linux?
  • The Linux threading model is not terribly clean, and there are still way too many huge locks in the current kernels.

    I had the impression that the "threading" being talked about here is the threading model as made available in userland (e.g., Solaris's model with N threads per process implemented atop M LWPs) rather than the degree to which the kernel itself is multi-threaded.

    Linux had kernel threading retrofitted instead of woven into the OS like Solaris

    I was under the impression that the SVR4.0 kernel Sun started with had neither MP nor userland threading support, and that Sun added both of them. (Yes, there were SVR4.x releases that had both MP and threading - I think there were even SVR4.0 versions with them - but, as far as I know, that wasn't what Sun started with; as of when I left Sun in 1988 we were getting pre-release loads of SVR4 from AT&T.)

  • I would be curious to see a transcript or some paper of Linus making an absolute fool of himself by saying this. The Linux threading model is not terribly clean, and there are still way too many huge locks in the current kernels. Solaris thread granularity leaves Linux's in the dust; Linux had kernel threading retrofitted instead of woven into the OS like Solaris or even BeOS.
  • um, gee, you think they ain't going to use a firewall??????
  • The article was not saying that Linux didn't have threads. It was saying that Linux threads were not compatible with the threads provided by other Unices. And that's true.
  • I'm not so sure I agree with that statement. Linux has an exceptionally low latency level, and our context switches for normal processes are at least as cheap as Solaris's threads. So Linus & co. made the reasonable decisions and made threads a special process. Now you can argue all day about whether you want a separate run queue for threads or not, but I'll trust Linus Torvalds, DaveM and Alan Cox on OS design over someone who doesn't have numbers to back their claims up.

    And if you want numbers about linux threads, trawl through the lkml archives. There was a discussion about this not so long ago.

  • For the last two years, I have been watching x86 hardware manufacturers and application software vendors to see what they would do with respect to the portion of their base that has (historically) run on SCO. This is what I have seen:

    SCO raised their license fees, and (theoretically) passed the margin dollars on to their partners.

    Some hardware vendors (like Compaq) have continued to back SCO. These vendors tended to be the ones that had the largest base of x86 *NIX sales prior to the general acceptance of Linux.

    Some software vendors jumped to platforms like AIX / RS/6000 to maintain their high margins.

    Most of the others who have embraced Linux had little or no share of the preceding *NIX market on Intel.

    I would guess that the people that embraced Linux and include a service component in their marketing strategy are generally making more money than they did in SCO. The rest are hoping that the SCO market doesn't collapse.

    I find it interesting that most people I talk to about SCO qualify most of their comments in revenue terms. That ought to help us understand why the vendors are still supporting it. It also gives us an idea of what would be required in order to displace SCO as an OS platform, if that is what we wanted to do.

  • Don't get me wrong...I'm not saying that Toy Story wasn't impressive, its just that it fit the example of what I was saying.

    Onyx's are SGI's gift to the movie industry. They are large servers that do very well with graphics/special effects rendering. Most of the movies with heavy fx in them use Onyx's, for example Twister.

    With the example of Toy Story, and the amount of machines they had to use (I remember when Sun had a story on their site about it, they were using Sparc 20's which at the time of the movie with the options they had would cost about $20000). Using 200 sparc 20's at US$20000 would cost about US$4million. You could probably do the same thing with 3 to five Onyx's which at that time cost around US$200000, so figure 5 and you still are at half the budget of the sparcs.

    I am sure that Pixar got some good kickback from Sun on the machinery because Sun used that as a huge advertising gimic, so that would save the budget on that movie, but it still doesn't account for the fact that if you were just to go out and buy the machinery, you wouldn't be saving anything.

    I talked to someone who used to do film remastering at one of the imaging giants here in Rochester. He said that they used three SGI Onyx's and it was more power than they would ever need for removing the dust and scratches off the last master of the movies turning it into the final master. They didn't do any special fx with them, but they did remaster every single frame for a large number of movies (any movie shot on this company's film) and they Onyx's work much more efficiently than any Sun solution they could get that competed in price with buy the three Onyx's (I think he said they paid US$200000 a machine).

    Here's a link to the new Onyx's. [sgi.com] So that you can read the specs on them.
    ---
  • All right, my first non AC comment, and I get moderated down to -1. Come on, you idiots, it was supposed to be funny! And I thought all funny comments got moderated up. Jeez..
  • Is the 2nd Gen. 64 bit intel even still x86?

    I really have not idea but it seems to me that it's not.

    Personally I'm keeping my eyes on the Alpha and the PowerPC. Out of curiosity I tried out one of the prerelease PowerPC linux distros on an old 7200 mac on it's way to the bone yard and was almost disappointed. It was pretty much just like sitting at a redhat box (accept slower - times have changed a bit since that 7200).

    I think that if Compaq works with the Linux community a bit with regards to compilers and the like they could have an excellent future. I'd love to have an alpha to beat up on my code and warm my feet :).

    Anyway even if 2nd generation Intel/HP cpu is the roadmap for all hardware (it sounds good but Compaq and IBM/Motorola are not sitting on their hands) how much money do you think compaq could make before that sees the light of day?

    And how much is that thing going to cost? Will I be able to get an Alpha for 1/2 as much by that time?
  • Actually, I believe the first dissenter was confused by my language. I don't work on internal servers. I work in an engineering group. "Compaq does have customer support for Linux on x86" is what I should've said.
  • My take on Dell selling Linux is that Bill Gates most likely talked with Mike Dell and said his lawyers need a big OEM showing that they are selling non-MSFT OS's. In comes Dell and look, you have to pay extra for the Linux machine. A MSFT mouse you say? I'm not surprised that MSFT mice are your only choice because MSFT wants their name on that box somehow. I say that Dell will drop Linux when the DOJ vs MSFT case is finished. Really now, Bill and Mike are buddies and Mike stands up for Bill at any chance he gets. I'd expect Linux on desktops from Gateway, Compaq and IBM way before Dell. So Dell is a lion in sheeps clothing if you ask me.

  • Yeah, sure, so whatever Intel does is the way to go now? Isn't that what is pissing everyone off about microsoft? I am honestly amazed that the entire world hasnt completely ditched intel after the whole PIII thing (you know what I'm talking about..). Especially since AMD now has FAR superior x86 processors. I dont get. Maybe those retarded intel ads really do affect people with IQ's > 10.
  • I'm glad that someone is working on making Linux more accessible on other platforms. Lots of work has gone into x86 versions, with Sparc and Alpha progress still lacking. Maybe Sun will follow suit and perfect Linux for Sparc. That would be a good move for them. They could have Solaris run servers, and use Linux as a workstation OS.
    Brad Johnson
    Advisory Editor
  • that bit about threads is wrong.

    linux does support threads, and has since about 1.2. There was a recent discussion about it on linux-kernel. What i understood from it was that threads and processes are both created by the kernel's 'clone' call - which takes arguments to specify whether to share memory/stack/etc..

    everything else uses that, in fact the traditional Unix process system creation calls are just C lib wrappers around linux clone(). Just as the thread library calls are.

    So it seems linux supports threads pretty well, just the userspace library wasn't up to it until glibc.
  • Oh ya... Think of the Beowulf cluster you could make with this!

  • Conflict of interest (Score:5)

    by scumdamn ( 82357 ) on Thursday September 30, 1999 @05:31AM (#1648747)
    There are a couple of companies that seem to "get it" and a couple that don't. I'd (and, of course, at least 50% will disagree with me, but that's Slashdot) put SGI and IBM in the group of companies that "get it" and Sun and Compaq in the group that doesn't. The latter two are treating Linux as the little brother to the OS they produce. Linux will never be sold by Sun as a solution by itself. They'd rather it be the "helper OS" to Solaris. Same with Compaq. Tru64 and VMS are always going to be the main NOS's they want to sell. Linux? Oh, it's that "low end" operating system we support on our smaller machines.
    At least SGI is contributing code to help Linux scale better, have a journalling file system, and be optimized for a UMA architecture. IBM seems to want Apache to be a good solution for everything and also seems to be more OS agnostic. It's too bad that Linux is in coopetition with Compaq and Sun because they could potentially be strong allies. As of yet, they are only half-hearted helpers. If either of them concentrated on making their platform the friendliest, most optimized platform for Linux, they would be in a strong position when all the optimization comes together to make Linux the strongest server OS bar none.
  • So they want to continue pushing SCO on the x86 market. Kinda funny that on some level they are actually competing with themselves, by pushing SCO in one hand and linux on the other.

    As to alpha, I will be very surpised if it leaves anytime soon. The new chips are very fast and (last time I checked) still going to be faster than the initial offering of IA-64.

    -- Moondog
  • I don't think they're looking at putting Linux on desktops at the moment (like Dell seems to be).

    Dell isn't "looking" to put Linux on desktops.. They already have. Unfortunately, due to their "Winmodem" policy, you can't get a desktop with a modem (if you prefer Ethernet, you likely don't care, especially if you live in a dorm). A quick skim through their site [dell.com] should back that up. (I'd post a direct link, but their site loads too slowly for my tastes)

    Of course, I still think it's funny that they appear to offer Logitech mice, but in the custom config page for their Linux systems, you aren't given a choice. It's MS IntelliMouse or.. well.. MS IntelliMouse. Yay!

  • Lets do that today!

  • Solaris threads etc. (Score:1)

    by Anonymous Coward
    If I remember correctly Linus officially decreed that solaris threading model sucks. And Linux model is much cleaner. This ofcourse is debatable. But from the application programmers point of view (which I am not) you really do not care what kind of threading model your kernel implements as long as you have POSIX interface to it. Ville
  • My understanding is that IBM is going with the "linux as little brother to our OS" approach too. They intend to have their own OS (replacing AIX) that'l run on merced (and a bunch of other chips... !!) for super-high end machines and recomend linux for the not-quite-so-high-end stuff.

    This doesn't bother me, frankly. They intend to support binary compatibility w/ linux binaries, etc. And so long as linux is around and not subjected to overpowering FUD (a little FUD is ok...) it will grow or not on its own merits. One of linux's greatest strengths is that it isn't subject to the market in the same way other OS's are.
  • Not so sure if seperating their efforts in development to target to different environments is such a good idea. Divide and Conquer still remains, so why divide yourself? Trying to seperate the desktop / server market is getting harder and harder anyways, seems like it would be better to have apps that will run on both. Of course what am I thinking, actually being able to run one single app on multiple platforms....Java, yeh right, that worked well.
  • Compaq DOES support Linux on x86. I work there and work on Linux problems on our servers.
  • You must not know much of what goes on in some of the internal units at Compaq, then. Just because every service is run on something other than linux doesn't mean that compaq support for linux is lacking. If anything enthusiasm among the engineers is quite high.
  • They're all by ACs and they all read 'hey, [article topic] would be great for beowulf clusters.' Or something along those lines. Some poor sod is just getting his rocks off.

    Isn't there a moderation tag like (Moron)?

    - dom
  • In a couple years, Compaq will be supporting SCO on IA32, Monterey on IA64, Tru64 on Alpha, VMS on Alpha, probably Linux on all three, and NT on both IA32/IA64. That's lots of products that function in the same market space. The bottom line is that something has gone horribly wrong with the Compaq DEC merger, from a strategy perspective.

    The theory was that Compaq-DEC would be able to compete with IBM. While supporting lots of OSes, IBM at least has a coherent product lineup from top to bottom. Compaq probably just has confused salespeople.

  • Re:Conflict of interest (Score:4)

    by sbuckhopper ( 12316 ) on Thursday September 30, 1999 @06:01AM (#1648759) Homepage Journal
    Linux seems to fall under the same cometetition standards as the server market did for a while.

    I remember a conference I went to 6 years ago called the SGI/Oracle Webmaster Survival Conference. In this conference, SGI made it well known that even though they would like to see one of their workstations on every desk top and one of there servers in everyone's server room, they knew that it wasn't possible. They knew that there would be a market split and that certain server/workstation solutions made sense in certain situations.

    IBM seems to get this too. They portray that although we wouldn't really mind if you bought all this computing machinery from us, it isn't a big deal if you don't, especially if it doesn't make sense to do so. At a company I used to work for we would use IBM's high end printing solutions. Although we ran the printer with an RS/6000, we ran the rest of the network with Sun equipment. IBM would have liked to see us running the rest of the network (which was all *very* dependant upon this printer) with IBM equipment, but it wasn't necessary and they understood that the Sun/HP solution we had in place made more sense.

    On the same respect, it seems as though Sun does not support this mentality. They think that every computer in your network should be a Sun computer. PC's? Forget it, replace them with XTerms. SGI's? Why do you need one of those when you could use an enterprise 10000 to do the same thing? Remember the movie Toy Story? They had to use (I forget the exact stats, but something like) 20 farms of 100 Sparc 20's to do the same thing that 2 or 3 SGI Onyx's could do. Sun doesn't seem to care if something doesn't make sense as long as it puts money in their pocket. Although I think that Sun has some great products, I think that this mentality will haunt them.

    That also seems to be the same way that these companies approach Linux as mentioned before so I won't mention it again. You are right, and the evidence has been around longer than all of these companies have been embracing Linux. Articles like this [sun.com] are good indicators that Sun still believes that every computer in the world should and will be a Sun. Perhaps Sun is the one that needs the reality check in this article and not IBM.
    ---
  • From now on on slashdot, if you do now talk about how cool "current technology" would be for a beowulf cluster you will me moderated down to "dumb"
  • i think you and the previous poster are not quite understanding each other. Linus has been quoted repeatedly saying that he likes Linux's implementation of the threads API for user processes better than the Solaris one. this doesn't have much to do with how either OS uses threads within its kernel.

  • Re:Solaris threads etc. (Score:1)

    by Anonymous Coward
    The Solaris threading model is good in some ways and not so good in others. For one thing, having the two level heirarchy (processes/threads) greatly complicates kernal code. It also requires rewrites of user space utilities ( like ps ), if you want information about individual threads. It also tends to make full process switching slow ( around 100 times slower than a thread switch ). Meanwhile, Linux only has processes ( threads are just processes that share VM maps, etc. ), and can concentrate on optimising this case: as a result, its context switches are reportedly about as fast or faster than Solaris's thread switches. In the past, the Linux pthreads implementations have been somewhat slower than they need to be because the implementers in user space had to jump through hoops to get some subtle signal behavior to be POSIX compliant. This is changing, from what I've read on the kernel list, with some concessions from Linus to help make a fully POSIX compliant implemention easier.

Slashdot Top Deals

All seems condemned in the long run to approximate a state akin to Gaussian noise. -- James Martin

Close