Commoditization, not open source, killed Sun Microsystems

The patent-troll industry is in full panic over the consequences of the Alice vs. CLS Bank decision. While reading up on the matter, I ran across the following claim by a software patent attorney:

“As Sun Microsystems proved, the quickest way to turn a $5 billion company into a $600 million company is to go open source.”

I’m not going to feed this troll traffic by linking to him, but he’s promulgating a myth that must be dispelled. Trying to go open source didn’t kill Sun; hardware commoditization killed Sun. I know this because I was at ground zero when it killed a company that was aiming to succeed Sun – and, until the dot-com bust, looked about to manage it.

It is certainly the case that the rise of Linux helped put pressure on Sun Microsystems. But the rise of Linux itself was contingent on the plunging prices of the Intel 386 family and the surrounding ecology of support chips. What these did was make it possible to build hardware approaching the capacity of Sun workstations much less expensively.

It was a classic case of technology disruption. As in most such cases, Sun blew it strategically by being unwilling to cannibalize its higher-margin products. There was an i386 port of their operating system before 1990, but it was an orphan within the company. Sun could have pushed it hard and owned the emerging i386 Unix market, slowing down Linux and possibly relegating it to niche plays for a good long time.

Sun didn’t; instead, they did what companies often try in response to these disruptions – they tried to squeeze the last dollar out of their existing designs, then retreated upmarket to where they thought commodity hardware couldn’t reach.

Enter VA Linux, briefly the darling of the tech industry – and where I was on the Board of Directors during the dotcom boom and the bust. VA aimed to be the next Sun, building powerful and inexpensive Sun-class workstations using Linux and commodity 386 hardware.

And, until the dot com bust, VA ate Sun’s lunch in the low and middle range of Sun’s market. Silicon Valley companies queued up to buy VA’s product. There was a running joke in those days that if you wanted to do a startup in the Valley the standard first two steps were (1) raise $15M on Sand Hill Road, and then (2) spend a lot of it buying kit at VA Linux. And everyone was happy until the boom busted.

Two thirds of VA’s customer list went down the tubes within a month. But that’s not what really forced VA out of the hardware business. What really did it was that VA’s hardware value proposition proved as unstable as Sun’s, and for exactly the same reason. Commoditization. By the year 2000 building a Unix box got too easy; there was no magic in the systems integration, anyone could do it.

Had VA stayed in hardware, it would have been in the exact same losing position as Sun – trying to defend a nameplate premium against disruption from below.

So, where was open source in all this? Of course, Linux was a key part in helping VA (and the white-box PC vendors positioned to disrupt VA after 2000) exploit hardware commoditization. By the time Sun tried to open-source its own software the handwriting was already on the wall; giving up proprietary control of their OS couldn’t make their situation any worse.

If anything, OpenSolaris probably staved off the end of Sun by a couple of years by adding value to Sun’s hardware/software combination. Enough people inside Sun understood that open source was a net win to prevail in the political battle.

Note carefully here the distinction between “adding value” and “extracting secrecy rent”. Companies that sell software think they’ve added value when they can collect more secrecy rent, but customers don’t see it that way. To customers, open source adds value precisely because they are less dependent on the vendor. By open-sourcing Solaris, Sun partway closed the value-for-dollar gap with commodity Linux systems.

Open source wasn’t enough. But that doesn’t mean it wasn’t the best move. It was necessary, but not sufficient.

The correct lesson here is “the quickest way to turn a $5 billion company into a $600 million company is to be on the wrong end of a technology disruption and fail to adapt”. In truth, I don’t think anything was going to save Sun in the long term. But I do think that given a willingness to cannibalize their own business and go full-bore on 386 hardware they might have gotten another five to eight years.

104 comments

  1. I agree about Sun regarding commoditization and open source, which reminds me that it may be time for another post on the smartphone market to follow up on the discussions of a year or so ago. Despite some predictions that commoditization and open source would significantly erode their position, Apple still seems to be doing quite well. It looks the history of the desktop PC market is not reflected in smartphones (and to some extent, tablets).

    1. >may be time for another post on the smartphone market

      Maybe, if I can find reliable market share figures. But I’m declaring smartphones off-topic for this thread; otherwise contingencies of the Apple-Android war will certainly hijack the discussion.

      Let’s stick to considering the actual post topic.

  2. Agreed but I doubt there’s any need to dispel the myth. Anybody entitled to an opinion knows it. Reminds me of the reported reply when Isaac Newton was asked for an erratum sheet to go with Principia – anybody capable of reading it will make his own corrections.
    For a while my brother had a pair of high end Sun machines in his office. One to tie up with longer runs and one to be free when the first was busy. Now it’s all Linux clustered and R. There’s nothing on the market to compete.

  3. Wholehearted agreement.

    I watched the same thing years previously, as supermicros based on off the shelf components became powerful enough to do the same things minicomputers did, and minicomputer vendors died horribly.

    A prime example might be DEC: why spend $250K on a VAX when a $25K supermicro would do it as well, almost as fast, and a *lot* cheaper? DEC *did* try to cannibalize their existing market, with machines based on their Alpha RISC chip. But they couldn’t ramp up RISC workstation sales fast enough to stem the bleeding as the VAX market imploded, and it probably just wasn’t possible to do so: people bought supermicros because they were *cheaper*, and getting everyone who had a VAX to switch to an Alpha simultaneously would have only hastened the end. Something similar happened to DEC competitor Data General, whose Aviion workstation running DG-UX was intended to supplant their Eclipse mini line. (DG also introduced a mass storage line that lives on as an EMC offering.)

    The bottom line is that you can’t make money on hardware, and haven’t been able to in some time. Value comes from outside of the organization, and you survive and prosper by providing what the *customer* thinks is valuable. They aren’t buying hardware, they’re buying solutions. If you tie your solution to particular hardware, you are more or less asking to die down the road when cheaper hardware you don’t use allows someone else to undercut you with an equivalent cheaper solution.

    1. >I watched the same thing years previously, as supermicros based on off the shelf components became powerful enough to do the same things minicomputers did, and minicomputer vendors died horribly

      I almost quoted “No one will survive the Attack of the Killer Micros” from USENET days.

      >DEC *did* try to cannibalize their existing market, with machines based on their Alpha RISC chip.

      Not hard enough or soon enough. In late 1997 or early ’98 I remember accosting DEC’s product manager for the Alpha at a conference and asking him what plans they had for pushing Alpha machines in the low-end market. He said they didn’t in a way that made it clear that neither he or any of DEC’s C-levels had even considered why that question might be interesting.

      Right then and there I realized “DEC is finished. Done. Over.” It made me sad; I cut my teeth on DEC-10s and VAXen and loved that hardware. But it was clear that even if I clubbed this clueless git upside the head and yelled at him nothing was going to happen. It’s not just that he was oblivious, the whole management culture around him was.

      Maybe they got a clue later, but in 1998 it was already too late.

  4. As a former Sun employee I had to nod in agreement with this. The writing had been on the wall for a long time, but the belief among higher management was that the X86 could be just an amusing sideline – for enterprise customers (those that run Oracle, SAP etc), only the big iron Sparc systems would provide what they needed. Sun did try to get into the cheap X86 appliance server market with their purchase of Cobalt, but that was ill-timed and the market collapsed. Same with their in-house development of blade servers.

    If I remember correctly, at some point during the long development of Solaris 10 there were plans to completely drop X86 support and have it Sparc-only. Later on there was a quick turnaround, and full X86 support was back on the roadmap along with the Andrew Bechtolsheim designed Opteron boxes. By then, though, it was too late.

  5. The bottom line is that you can’t make money on hardware, and haven’t been able to in some time.

    Unless you’re Apple. True, Apple is selling solutions, not just hardware, but the components — hardware, software, and services — are designed to function well together as a cohesive unit. So few seem to mind that the one is tied to the others.

    It’s theoretically possible to undercut Apple with cheaper versions of their solutions, but Apple’s business model is to own all the good ideas in design and manufacturing, making them compelling for people who care about quality. So they’re still making money hand over fist in a hardware market where literally everyone else is struggling.

    There was a running joke in those days that if you wanted to do a startup in the Valley the standard first two steps were (1) raise $15M on Sand Hill Road, and then (2) spend a lot of it buying kit at VA Linux. And everyone was happy until the boom busted.

    Funny — we’re living in a time when buying physical hardware for your startup is a nostalgic, old-timey thing — like dialing a rotary telephone or replacing the tubes in your radio. These days, you skip that and just requisition server instances in the cloud until you’re big enough that maintaining actual compute infrastructure becomes economical.

    1. >Unless you’re Apple. […] Apple’s business model is to own all the good ideas in design and manufacturing

      One word: Bendghazi.

      Which demonstrates, if demonstration were still required, that Apple doesn’t make money selling hardware, it makes money selling cult memberships. Its glassy-eyed partisans might as well be Hare Krishnas. This is probably not a sustainable business model.

      Anyone remember the iPhone 5 and “You’re holding it wrong”? If it follows the trendline, the iPhone7 will explode lethally in the user’s hand. And Apple cultists will welcome this as a a catapult to Jobsian paradise.

  6. > “they did what companies often try in response to these disruptions – they tried to squeeze the last dollar out of their existing designs, then retreated upmarket to where they thought commodity hardware couldn’t reach”

    This actually sounds a bit like what’s going on in higher education these days.

  7. That we can buy a powerful ITX based solution for considerably less than a current Windows License, is a simple fact today. so why would we not use a cheaper option of software to make it work? Drop to the RaspberryPi/BeagleBoard which comes with a pre-packaged OS and replaces even the ITX solution then even the commodity end has a problem making any money?

  8. I agree with the main point but would put it slightly differently: Sun went under not because they embraced open-source software but because they failed to embrace open hardware platforms.

    To a large extent, this was what killed DEC as well, and had Apple hurting about the time they rehired Steve Jobs — who then moved the Mac onto Intel hardware. (Apple will tell you that their closed-source software is still a big part of their success, but I would predict, and I’d bet Eric also would predict, that sooner or later Apple will have to go open-source in software, too, or go under.)

    Which brings us to the last major part of computing that is still dominated by closed hardware — the designs of CPUs, GPUs, and other hardware below the component level. This area will be hard for open-source to crack, because the government will tell you that the secrecy of these designs is a national security matter. What I think they’re really saying is that the three-letter agencies have always put in back doors for themselves, and want to keep doing so undetected. But now that we’ve heard Snowden’s revelations, we the public need to demand these designs be open. And preferably that some means be made available for us to audit their accuracy ourselves, which I’m sure is not a simple problem.

    Once it’s all open-source, they’ll never be able to make us go back.

  9. Sun went under […] because they failed to embrace open hardware platforms.

    The ditches are littered with the corpses of generic PC makers. True, they weren’t fully “open” in the sense of having open source CPUs, but that’s pretty much a pipe dream. Intel and Apple put hundreds of millions of dollars into CPU development, so they aren’t going to open source the result. And nobody who wants an open source CPU has hundreds of millions of dollars to spend on designing and making them, nor are they foolish enough to try to compete with Intel and Apple. I don’t think it has anything to do with national security, it’s just economics.

    1. >“Bendghazi” has been pretty thoroughly debunked

      Every time Apple fucks up, the fallback position is that under carefully controlled conditions of temperature and marketing pressure the problem doesn’t replicate. Well, of course it doesn’t; that’s why they’re carefully controlled. Duh!

      It’s like an annual post-release ritual these last few years. What stupid-obvious design flaw did they miss? How many RPMs of spin will Apple marketing achieve this time while trying to airbrush the issue out of existence? How embarrassing will the CEO’s next on-stage appearance be?

      After multiple instances we can begun to notice a pattern. That is: insufficient testing with real users. I put this down to time-to-market pressure; the prototypes probably get a whole fifteen minutes of field testing by employees who are under tremendous implicit pressure not to upset the launch schedule.

      I don’t actually blame Apple for this much. Under present market conditions it’s pretty hard to avoid this kind of mistake. I do blame Apple cultists for their persistent delusion that the sun shines out of Cupertino’s asshole.

  10. The next casualty of the killer micro might be the mainframes, which still survive to this day in some legacy/enterprise applications. They’ll get disrupted from below by ‘big-data’-oriented clusters, which have the requisite scalability, data-processing throughput, and potentially fault-tolerance/reliability.

    That is, unless mainframe companies slash their prices in order to stay competitive, at which point they will be well positioned to enter the emerging big-data sector with their own solutions.

  11. > “Bendghazi” has been pretty thoroughly debunked.

    Yes and no. From what I have read, bending in back pocket was debunked (this was tested… I’d like to see the test setup for this :-)), bending in hands is confirmed (compare Galaxy Note 4, IIRC).

  12. Yup, OpenSPARC is a thing after all. Ironically, I think Sun might have been _more_ open in HW than they were in software. But fully open-source hardware has lots of challenges which may be insurmountable after all. Of course this shouldn’t stop hardware companies from pursuing things like open specs and product documentation (i.e. not requiring NDAs), which _are_ a significant advantage.

  13. Jakub, my link includes links to various articles which include pictures and details of the test setups used.

  14. > Same with their in-house development of blade servers.

    Were they too early for blade servers?

    1. >Were they too early for blade servers?

      I’ve actually had that thought myself. Too early, and with their attempt a bit too tied to bigger iron, was how I read it.

  15. >This actually sounds a bit like what’s going on in higher education these days.

    Please elaborate.

  16. IIRC, Larry McVoy was at Sun in the early 90’s and put out an internal white paper, “The Sourceware Operating System Proposal” that called for doing it *then* and basically preempting/becoming Redhat. It’s an interesting counter historical to consider…

    1. >IIRC, Larry McVoy was at Sun in the early 90’s and put out an internal white paper, “The Sourceware Operating System Proposal” that called for doing it *then* and basically preempting/becoming Redhat.

      This is true. I have read the white paper. McVoy was substantially ahead of the curve there.

      McVoy has been heard to imply that I might have got the fundamental idea about open source from that paper. That is not true, the timing was wrong, but it wasn’t a crazy guess to make if he didn’t know exactly when I read it. Richard Gabriel, via his Worse is Better paper, has a better claim to influencing me.

  17. Oh, come on. Not only does Apple test things thoroughly (they recently released a video showing iPhone durability testing procedures), but SquareTrade and Consumer Reports also have tested them, and found them fine.

    So on the one hand we have Apple (makers of 500 million iPhones and counting) and two established companies that are in the business of testing such things, all of whom who say they are fine. On the other side we have one YouTube video and a total of (AFAIK) *9* reports of bent phones. I think the first side has a tad more evidence. Unlike Benghazi, this really is a fake scandal.

  18. “The bottom line is that you can’t make money on hardware”

    This is overstated. Hardware is still needed, and someone has to manufacturer it. But to consistently make money in hardware, you have to be the lowest-production-cost commodity supplier, and you need to have a sustainable manufacturing advantage to keep you there.

    Offering the lowest price isn’t sufficient if your costs aren’t also the lowest, as K-Mart found out when they tried to compete with Walmart by lowering consumer prices.

    In other words, you have to be the Walmart of the manufacturing space.

  19. RE: DEC cannibalizing their own market, I wasn’t born yet, but it strikes me that the best timeframe and manner for DEC to have cannibalized their market would have been to have introduced a PDP-11 based micro along the lines of the DEC professional a year before IBM introduced the PC, rather than a year after.

  20. @Cathy:
    I’m sure that’s entirely true, either. I’d point out that BMW, Lex, etc., all have thriving markets in cars which can be replaced nearly identically for half the price. Part of this goes back to solutions as previously mentioned.

  21. I, too, thought “attack of the killer micros!”

    I do find myself wondering how a properly-marketed Alpha would have fared against the x86 line. I was an Alpha bigot for several years, and still have a pretty well loaded Alphaserver 4000 in the basement (though it got retired in favor of an Itanic).

    There was a time, as well, when I gave serious consideration to buying one of the Sun Opteron boxes. They looked that good and that well priced.

    I’m not sure Solaris has anything to offer over Linux, though, nowadays…

    1. >I do find myself wondering how a properly-marketed Alpha would have fared against the x86 line.

      So do I. It was a beautiful architecture. I would have loved to write a compiler for it.

  22. It’s like an annual post-release ritual these last few years. What stupid-obvious design flaw did they miss? How many RPMs of spin will Apple marketing achieve this time while trying to airbrush the issue out of existence? How embarrassing will the CEO’s next on-stage appearance be?

    Apple sold over 10 million iPhone 6s in its first weekend, most of them falling into the hands of users who are probably not so careful with their devices as you or I might be. So far it has received a total of nine Bendghazi complaints. That is well beyond six-sigma tolerances for this particular defect and an excellent quality track record. The Bendghazi kerfuffle supports the thesis that Apple is an exemplar of manufacturing quality. An Apple product failing under rare, pathological conditions is like a bit of cellulite on Scarlett Johansson: the press will publicize the shit out of it because they just can’t resist cutting the tall poppies down to size because they know the masses will eat it up. Just like you did because you’re just hoping Apple will take a huge fall for daring to defy the laws of commoditization and openness for as long as they have.

    But just because you desperately want something to be true, doesn’t mean it is.

  23. I’m not sure Solaris has anything to offer over Linux, though, nowadays…

    ZFS. I think that’s about it. And I believe ZFS on linux is now very close to being reliable. It may even be considered fully reliable, I haven’t checked for a while though.

    Apropos Apple. It is worth noting that Apple has yet to release a fix to the Bash bug (aka shellshock) – see http://arstechnica.com/security/2014/09/apple-working-on-shellshock-fix-says-most-users-not-at-risk/

    I could be wrong, but my understanding is that you can get an Apple to run a bash script with a cunningly chosen DHCP option so the “most users not a risk thing” is a total lie. You’d just need a Raspberry Pi to hand out DHCP leases faster than the wifi router at your local starbucks to infect any hipster who shows up to mooch wifi

  24. >I do find myself wondering how a properly-marketed Alpha would have fared against the x86 line.

    So do I. It was a beautiful architecture. I would have loved to write a compiler for it.

    DEC Alpha’s ran Win NT significantly better than Intel boxes did back in the mid 90s.

  25. “DEC Alpha’s ran Win NT significantly better than Intel boxes did back in the mid 90s.”

    This is damning with faint praise…

    OTOH, I did run Linux on a couple of the little Alpha desktops (the ones with the tantalum caps that died and took the machine with it). Ran quite well.

  26. This article makes me wonder if RIM could have saved its bacon by going from Blackberry to , say “Openberry” ;) I’m guessing not because I don’t think RIM had enough of a mobile OS to compete with iOS and Android even if they made open source. On the other hand, if a bunch of hackers jumped in like they did with Linux two decades before, it might have made a difference, and maybe we’d have bit more freedom on our handhelds.

  27. “DEC Alpha’s ran Win NT significantly better than Intel boxes did back in the mid 90s.”

    This is damning with faint praise…

    Well yeah, but shortly thereafter, despite the performance issue, Win NT ate Novell Netware’s lunch and then Win NT (and successors) in combination with Exchange led to Microsoft pretty much owning the file/print/email server market (and client market) for the last almost 2 decades. That’s actually quite an impressive achievement.

    If someone like Sun or DEC had managed to get a reliable corporate backed Unix distro that was as affordable out there in the mid/late 1990s before the domination of NT became unstoppable things would have very different. Unix already had good email systems, decent file sharing (NFS), TCP/IP network stacks plus built in security etc etc. Either could easily have done such a thing and if priced right a baby workstation using as many industry standard components as possible (i.e. EISA or PCI hardware bus, PATA hard disks, VGA graphics etc etc.) would have been fairly simple. There were plenty of Windows client programs that could connect to unix servers over TCP/IP and thanks to netware people were used to installing the client drivers on windows machines.

    Heck if Novell in 1996 or so had properly got behind a Unix instead of sticking to the propriatary Netware OS for Netware 4 maybe they could have done it. They had UnixWare at the time I think.

    VA linux was probably (IMHO) a couple of years too late to be able to have got mindshare, plus their buisness model was, as I understand it, to eat the low end unix workstation/server market rather than go after the much bigger SME file/print/email server market. But that could have been another option. Problem was, once companies standardized on Exchange and Outlook for mail and calendaring they were locked in and still are today.

  28. DMcCunney:
    “””The bottom line is that you can’t make money on hardware, and haven’t been able to in some time.”””

    Dell
    SuperMicro
    HP (sort of)
    IBM (sort of)

    Hell, Oracle is still selling Sun hardware.

    You *can* make money selling pure hardware. You can’t make money selling a server operating system.

    Apple will, to our hosts chagrin, continue to sell well because they do for their user interface what they do for their hardware. In a way they do the interface what he does to his datastructures and algorithms.

    I’ve been running linux personally or professionally off and on since about January of 1993. Kernel version .99p5, and rather than deal with train wreck that is the Linux User Interface these days, I do almost all my work sitting in front of a Mac, because, as Blank Frank once put it on the CypherPunks list:

    Interfaces matter. You need mathematical bones; engineering muscle; but you
    won’t replicate without beautiful skin. Bits, transistors, wires, code,
    gummint velveeta is free. Will is expensive. Gutenburg. Smith. Ford. Moore.
    Postel. Steam engines were neat. Steam engines pulling trains were amazing.
    Computers were neat. Computers networked were amazing. Warning grunts are
    useful. The ability of a charistmatic speaker to fuck with your head is
    disastrous.</blockquote.

    Ok, only about 1/2 of that was relevant, but Blank was like that.

  29. “This article makes me wonder if RIM could have saved its bacon by going from Blackberry to , say “Openberry” ;)”

    The question is when they could have done this. If they did it after Android was already well-established, it would be too late. But if they had gone down that path when Android was barely out of the gate, who knows?

  30. Wasn’t Blackberry QNX-based? If so, wouldn’t this have gotten in the way of open-sourcing it?

  31. @Jeff Read: “‘The bottom line is that you can’t make money on hardware, and haven’t been able to in some time.’

    Unless you’re Apple. True, Apple is selling solutions, not just hardware, but the components — hardware, software, and services — are designed to function well together as a cohesive unit. So few seem to mind that the one is tied to the others.”

    Apple *isn’t* selling hardware in the sense I was talking about. Look at a teardown on an Apple device, and you see off-the-shelf components – ARM or X86 CPUs, other chips to do other things – tied together in an Apple design as part of a solution. Precisely what parts get specced for inclusion is a matter of fierce competition among suppliers making them, and industry analysts read the tea leaves trying to guess who will rise and fall in future iterations, because Apple is such an enormous market. But none of the hardware is proprietary and unique to Apple, and Apple’s success is not tied to proprietary hardware.

    DEC was tied to the proprietary LSI-11 architecture used in the PDP and VAX lines. On a deeper level, this was the RISC vs CISC battle, with the VAX as a spur.

    The VAX had a splendid instruction set, but people realized that many if not most of those instructions never got used, because compilers did not generate object code that used them. The question arose “Why have them at all? Design a chip with a minimal set of basic instructions, make them as fast and eficient as possible, and let the compiler combine them to create the higher level operations.” So we got a generation of RISC based machines. DEC shifted to the Alpha. Sun bet on SPARC, which was available for others to use, but few did. Along similar lines, IBM had the PowerPC CPU, and HP had the Precision Architecture among others.

    It worked or a while, but eventually, CISC, in the form of Intel’s X86 architecure won, providing chips with as much speed and power as a RISC CPU, but cheaper. This is the overall trend in just about any area. As markets mature, things increasingly become commodities, and competition devolves to price.

    The trick is finding things that don’t commoditize like that. Apple largely has. The hardware and software underlying thier product are largely commodity products. The solutions they sell are not. Other vendors can produce similar products and price them cheaper, but they won’t be as well designed, thoroughly integrated, or intuitively usable as Apple’s. In the process, Apple has created what the customer sees as sufficient value to command a premium price, get enormous revenue and proits, and a stock price in the ionosphere.

  32. Only BlackBerry 10 and the Playbook version were QNX-based. QNX has been around for so long, and is so critical and valuable — being a vital component of many embedded/SCADA type systems — that open-sourcing the thing would actually be a really good idea. It’s not like selling the OS itself is such a moneymaker for BlackBerry; like Darwin, it’s the hardware/OS/shiny-layers-on-top combo plate that really matters.

  33. @Jay Maynard: “Wasn’t Blackberry QNX-based?”

    The current version of Blackberry’s OS is based on QNX, replacing what they had used before. Since RIM bought the outfit that develped QNX to get and use it, they likely *could* make it open source. The question is how they would benefit by doing so.

  34. But none of the hardware is proprietary and unique to Apple

    The A-series CPUs in iOS devices are proprietary, even though they are based on ARM designs. That’s rather important. The magnetic power cord on MacBooks. The cases, of course. There might be others.

  35. a) I thought you wanted smartphones to be off topic. If you want to talk Bendgahzi and how Apple sells a cult and not hardware perhaps you should do so in the context that it does so to 42% of the US market in July BEFORE the iPhone 6 launch. That’s a lot of cult members.

    b) Apple is the number 1 unix vendor today. 16M unix hardware sales in 2013. Of course they sell solutions. Sun sold solutions too. Oracle does today. Apple is just better at it for their target market (desktop of all things) than any other unix vendor.

    c) Linux didn’t kill Sun’s low end market. Microsoft did. Wierdly IIS share is growing in 2014. In any case, many of the lower and mid tier servers were lost to Windows Server (for AD, Exchange, SQL, Sharepoint, etc). They moved up market not because of commodity linux but because of commodity windows. Sun’s fumbling of Solaris x86 gave MS the window (heh) to devastate them in the lower end server market.

    d) the bash bug is not critical to OSX because the default shell is tcsh. First, OSX DHCP doesn’t run a shell anymore. Even if bootpd did invoke a shell it would most likely be sh (or perhaps tcsh…but /bin/sh is far more likely). If you are running apache with a vanilla CGI with bash or SSH server are you vulnerable. That’s not most folks. Unlike RHEL where DHCP does run bash scripts. Even the CUPS web interface is disabled by default on recent OSX version.

    1. >I thought you wanted smartphones to be off topic.

      I did. I should have known the cultists would make this impossible.

      1. >Linux didn’t kill Sun’s low end market.

        VA’s customers sure thought it did; these were not people migrating to Windows. Sun’s people thought it did too.

  36. The A-series CPUs in iOS devices are proprietary, even though they are based on ARM designs. That’s rather important. The magnetic power cord on MacBooks. The cases, of course. There might be others.

    Apple tends to have stuff that isn’t quite normal. EFI vs BIOS, thunderbolt vs USB 3, PCIe SSD vs SATA SSD. Firewire vs USB2. Unibody construction. High resolution displays (aka retina). mDP vs DP (or hdmi) connections.

    Most is not proprietary but the combination of use is rare until folks start to copy it. Like that whole ultrabooks category copied from the MBA. Or Intel NUCs are copied from the weird little Mac Mini (I love it but what does it have to be that small?). Or the AIO form factor for desktops.

    But usually by the time they copy it Apple has some other not-normal thing that is slightly better but more expensive and commodity makers simply can’t turn a profit doing that yet and even if they did Windows wouldn’t support it anyway.

  37. Since someone mentioned the mainframe, this has gone from smelling funny to smelling awful quite recently.
    In the course of the last year, the company I work at has shown that the business software of a major insurance company can be successfully migrated from an IBM MF to a Linux server that runs a framework (supplied by Microfocus) which emulates the MF environment (file system w/ catalog, PL/1 programs, JCL,…).

  38. I did. I should have known the cultists would make this impossible.

    Stop being a dick. The first person to talk smartphones after your declaration is YOU when you brought up bendgahzi as if it were a real thing.

    What? We weren’t supposed to mention apple computers at all in the context of Unix, computers and the ability to sell hardware for good money despite commoditization either? Who’s the cultist?

    These are all things Sun could have done but instead they choose open source and it killed ultimately them from a combination of revenue loss and missed opportunities.

    If they has not open sourced Java then Google would have had to pick a different language for Android and not embraced, extended and extinguished Sun’s Java ME business. Plus Android would have been slowed down a bit. perhaps enough for Sun to copy Apple’s successful mobile app platform for J2ME handhelds.

    They should have worked harder to make java faster and better for devs and not more open so IBM could make money on Java using AIX and Linux.

    For the OS side they got nothing for opening up Solaris. They should have worked harder on ease of use and ease of administration and deployment rather than all that effort and focus to court the open source world that never liked them anyway.

    Finally, look at the server OS market share. Microsoft owns 40%+. All those departmental unix installs for mail, disk and database servers went x86 + windows server and not Linux.

    VA Linux was hardly a pinprick in comparison the the share lost to MS.

    1. >If they has not open sourced Java then Google would have had to pick a different language for Android and not embraced, extended and extinguished Sun’s Java ME business.

      You should at least try to keep your story consistent. Last time around Sun’s problems were Microsoft’s fault.

      >For the OS side they got nothing for opening up Solaris.

      That much is almost true. They were at least three years too late doing it, probably five years. Though it would have been too much to expect, I guess, for them to have seen the necessity that soon.

      >VA Linux was hardly a pinprick in comparison the the share lost to MS.

      In the Valley it was way more than a pinprick. MS didn’t compete effectively against either Sun or VA. This I know because I sat in the board meetings and talked with our marketers. The ex-Sun people I know don’t know how much VA Linux itself cut into their workstation business but they are very clear on how much Linux did.

  39. Uhm, Nigel…the default shell on every Mac I’ve set up in the last, say, 7 years or so has been bash. Maybe longer. You might have tcsh because wit’s migrated your settings…but I’ve set up OS X VMs from scratch, no migration done, and gotten bash when I’ve fired up Terminal.

    With that said, though, I’m highly skeptical that the security hole affects OS users anyway: as you point out, OS X no longer uses bash to run DHCP setup, and damned few OS X systems have enabled web servers on them at all.

    Kurt, I new Micro Focus had that kind of environment. I didn’t know they’d been successful in using it to migrate major systems, though. Mainframe-to-<smaller server> migrations are notorious for taking far longer, costing far more, and being far less successful than the marketing types claim, and often being abandoned completely when the customer realizes that it’s cheaper just to upgrade the damned mainframe; I’m happy to see that, at least, yours got to the finish line.

    1. >Uhm, Nigel…the default shell on every Mac I’ve set up in the last, say, 7 years or so has been bash.

      I meant to mention that Debian gets kudos for foresight here. Years ago they defaulted the shell used for system scripts to dash, a derivative of the old Almquist shell optimized for small memory footprint. This means that Debian and its derivatives – notable Ubuntu and Mint – are, so far as anyone not privy to still-private CVEs knows, immune unless the user has reconfigured to use bash places where he shouldn’t.

      The AskUbuntu entry says: “However, Ubuntu’s /bin/sh is not bash, so only programs that explicitly invoke bash and not the default scripting shell are affected.”

      ssh to a user account configured with bash as login shell could use the exploit, but if they can do that they have your password or private key and you have bigger problems than shellshock.

  40. nht wrote:
    >If they has not open sourced Java then Google would have had to pick a different language
    >for Android and not embraced, extended and extinguished Sun’s Java ME business.

    You’re got two errors there:

    1. Android does not use the (GPL-licensed) Java code. Instead it was based on Apache Harmony, a clean-room re-implementation.

    2. Sun kept J2ME effectively proprietary by releasing its libraries as GPL without the classpath exemption. That meant that any platform running J2ME without a license from Sun would only have been allowed to run applications with a GPL-compatible license–no proprietary apps allowed.

    Arguably, Sun refused to provide the TCK that would have allowed Harmony to call itself Java precisely to protect J2ME from open source Java competition. Android worked around that issue by abandoning Java compatibility and branding.

  41. “Sun kept J2ME effectively proprietary by releasing its libraries as GPL without the classpath exemption.”

    And Stallmanites complain when I point out that the GPL is actively harmful to their cause.

  42. @jay yep, my bad. The default is bash.

    Guess I should have checked my new mbp with the clean install and not my old mini that has been upgraded since forever (it’s an old core 2 duo one).

  43. @jim arguably (and that’s what I was doing :)) if it were not for the JCP and suns efforts to allow for an open java infrastructure then IBM would not have jumped on board with Java but gone their own route. Without IBM harmony probably wouldn’t exist.

    You can counter with Java not being as popular as it is today but that popularity never translated into a significant competitive advantage for Sun.

    Just having Java on windows, Unix and Mac (the latter being an iffy desire) would make it more superior to C# as a cross platform managed language.

    C# and Visual Studio is a competitive advantage for MS. Java and NetBeans could have been for Sun…although probably more like ObjC and Xcode in scale.

  44. Yeah, open-source didn’t contribute to Sun’s downfall. The fact stuff like Dtrace, Lustre and ZFS were given out to competing OS vendors for free when they could have been a competitive advantage to Solaris didn’t play any role.

    What was the thing FOSS advocates wanted from Sun anyway? Keep Jonathan Schwartz in CEO position ’till he drives the company to the ground, open sourcing the company’s crown jewels that could have been a Solaris exclusive? Well, duh, that’s exactly what the FOSS advocates wanted, but investors, aka the people who actually have a stake, didn’t want that.

    And let’s not get started on OpenOffice, a project that was completely irrelevant ever since Sun lost interest in the workstation market and focused on servers and supercomputers. But you see, OpenOffice was flipping the bird to evil MS, and that’s how you build value for the company, by flipping birds.

    Meanwhile, under Ellison rule, Sun’s hardware department is doing much better than before (see the Sparc M7). As regards Solaris, the desktop market is gone because of X.org (Schwartz didn’t want to invest in something like Quartz), but it’s gaining relevance on the server again, because of better intergation with Oracle applications…

    …in other words, *exclusive* features. That’s how you beat commoditization.

  45. There are two critical things that Sun did, or didn’t do, that I gathered were critical in its failing at “commodity” hardware:

    Long before that became a big thing, they changed their sales structure so that “small” (less than mainframe, multimillion dollar) orders had to go thought “VARs” and the like, who were generally incompetent and/or uninterested. By the time of the post-dot.com revival, I read the tales of woe of many startups that found it impossible to buy Sun gear beyond what they could put on a credit card on the web site, in the period where they were successful but not yet “large” in Sun’s eyes. Of course, by the time they got that big, they’d already gotten sufficient experience making hardware such as Dell’s work for them.

    The name of Dell frequently came up because they were, oddly enough, willing and able to sell their hardware to anyone; it wasn’t of the quality of Sun’s, but you could actually procure it.

    Then Sun started lowering the quality of their x86 systems. I remember reading that they substituted a chipset? Marvell? Ethernet chip for one of the standard, very well supported and reliable Intel one that everyone loved. And that Joylent stopped buying Sun when it silently started fairly randomly changing the lights out hardware (IPMI or the like).

    Based on this, I think it remains unproven that they couldn’t have succeeded with “commodity” hardware, they just had too many internal problems that were clearly quite capable of killing them.

  46. And before someone says there was some great plan by Schwartz that didn’t have time to realize, here is a quote from him:

    “What’s the business model? I don’t know. But if you don’t have adoption, it won’t matter what business model you use. Companies that sell open source are prioritizing community and adoption over instant monetization. We will win.”

    http://www.cnet.com/news/q-a-jonathan-schwartz-on-suns-open-source-business-strategy/

    So, doing crazy experiments on a 5 billion dollar company (with lots of employees and lots of smaller firms with their fortunes tied on the company), based on some wacky theory that adoption is the only thing that matters, while business models are not (hint: it’s both, you must have both, see Oracle). What a d-bag.

    1. >So, doing crazy experiments on a 5 billion dollar company

      …is, when you have irrelevance looking like your near-term future, sometimes the only recourse you have.

      I wasn’t a huge fan of Jonathan Schwartz – I recall him blue-skying about selling Sun boxes in supermarkets in a way that seemed astoundingly dumb – but Sun as a company was on a fast road to nowhere. If they hadn’t open-sourced the things you think they could have kept proprietary their customers would simply have defected to Linux faster.

  47. ZFS is great and all, but I do believe Linux will get there eventually. The one thing I really miss as an ex-Solaris guy is Dtrace. It’s much more potent than any other production-usable tracing system I know of, and the Linux equivalents don’t really have the same sort of architecture.

  48. Nb. Apple uses to open source projects (and contributes to at least some of them): GNU tools, Darwin, WebKit, Mach^W XNU kernel,…

  49. @Mike E: there are efforts to port ZFS to Linux, as far as license allows. And there is Btrfs.

    I thought that SystemTap would be DTrace equivalent, but it is forever out of tree; perf events (who the *&%# named this thing to be un-Googleable) and BPF for tracing comes further and further.

  50. So the most accurate (if not overly useful) statement is lack of direction and poor execution on top of that is what really killed Sun and not commodification because other companies have flourished in that environment (like say apple). Likewise other companies have strategically open sourced IP without going bankrupt (like say apple).

    One wonders what Sun would be like if they bought NeXT and made Jobs CEO rather than Apple.

  51. If they hadn’t open-sourced the things you think they could have kept proprietary their customers would simply have defected to Linux faster.

    No. Customers don’t like change. And if Sun had kept their competitive advantages intact it would have been harder to destroy them.

    But with My Little Pony who drank the FOSS kool-aid in charge they were doomed.

    1. >And if Sun had kept their competitive advantages intact it would have been harder to destroy them.

      In 1998-1999 Sun didn’t have any competitive advantage that could justify their nameplate premium. I know this because I was on the board of a company that was capturing Sun’s Silicon Valley customers and prospects about as fast as we could ramp up production and distribution. One of our challenges was moving product fast enough to meet demand.

      This is exactly what a technology disruption looks like. The sustaining company (Sun) thinks its gold-plating is boosting demand, when in fact the engineering has overshot actual customer demand. The second a lower-cost vendor gets just good enough the sustaining company haemhorrages customers,

      You can spin all the counterfactuals you like; I was there, and I had plenty of contacts inside Sun to tell me what the score was.

  52. Customers hate change, but they like money. It was the crushing price/performance advantage of 2 & 4 socket x86 boxes that strangled Sun’s traditional business model. It didn’t matter how good Solaris was, once Linux was adequate they were living on switching costs.

    1. >they were living on switching costs.

      That could be inscribed on the gravestones of dozens of tech-disruption victims.

  53. Sun was already dying when they open sourced Solaris.

    Aside from that, it’s not like people were leaping with joy at the prospect of using Solaris for free. It’s not like people were buying Sun machines in order to get access to Solaris. Rather, they liked those Sun machines, and the supported OS to use with it is Solaris.

    I was in a shop that made the transition from SunOS 4 to Solaris, and I remember it feeling like a huge waste of time to us. The new fancy OS had nothing in it that objectively helped us; no new and exciting APIs, no obvious improvement in performance, no obvious improvement in reliability. It was just a bunch of random breakage we had to track down, because they didn’t maintain backwards compatibility very well.

    1. >Rather, they liked those Sun machines

      Everybody liked Sun machines. I liked Sun machines, pre-386, when I could get my hands on ’em.

      But I saw the handwriting on the wall in 1985 when the 386 shipped first silicon. Even then I understood that Intel’s ability to ship volume and drive process improvements was bound to swamp the fuck out of Sun and Apollo and half a dozen other now-forgotten workstation vendors – and wrote up that prediction for a 1986 SAMS anthology called Tricks of the Unix Masters.

      There have been many times I’ve been called a nutcase and lived to laugh at the critics. That may have been the first.

  54. Sun’s situation was even worse than Eric describes. In 1990, as the 64-bit SPARC architecture was being defined, some critical technical decisions were made that compromised the competitiveness of the architecture. The biggest mistake was to keep register windows, because it required a lot of chip area that didn’t really contribute much to performance. Processors at that time were starting to do multiple instruction issue, which meant you needed to duplicate the registers. So you take a lot of registers, and you double their width, and then you replicate them, and pretty soon you are looking at 25% of your chip area for this one thing. Competing processors got to put that area to use on things that actually improved performance.

    This led to a long period in which SPARC performance was progressively less competitive. The whole SMT push was an attempt to “hide the pea” on this issue. The final nail was when somebody did a full-system SPARC emulator that ran on a Pentium server and out-performed the real thing by a noticeable margin on a box that cost half as much. Given that the two machines have different byte orders, that’s not an easy thing to do. Sun bought the company and buried the technology.

    Sun compounded bad hardware decisions with bad decisions in the compiler group that *further* weakened their performance. Eventually they built a disastrously bad storage subsystem that corrupted customer data. How? Their much-ballyhooed operating system suffered terrible race conditions.

    It wasn’t *just* commoditization that killed Sun. It was also a string of poor technical decisions and poor execution that progressively weakened their competitive position for more than a decade. IBM has done just fine in the Pentium-class server business, and Lenovo will do better with it. Sun had every opportunity to own that business. Instead, over a ten year period, they managed to snatch defeat from the jaws of victory every 15 to 18 months.

    1. >The final nail was when somebody did a full-system SPARC emulator that ran on a Pentium server and out-performed the real thing by a noticeable margin on a box that cost half as much.

      Holy shit. I knew the price-performance gap was bad – I had to, I was helping run VA’s strategy (albeit in a relatively minor way). I didn’t know it was that bad. I say again, holy shit.

  55. Eventually they built a disastrously bad storage subsystem that corrupted customer data. How? Their much-ballyhooed operating system suffered terrible race conditions.

    That reminds me of another non-technical decision that a lot of people predicted would kill their enterprise business (that is, the line of machines they were actually willing to sell to you). They shipped systems in the very late ’90s and maybe the year 2000 that had a nasty cache corruption issue, then tried to hide this by demanding customers sign NDAs (as well as other class acts like blaming the company/ies they got the chips from).

    Whatever the performance or lack thereof, this demonstrated they weren’t truly a vendor of “enterprise” level computers.

  56. The final nail was when somebody did a full-system SPARC emulator that ran on a Pentium server and out-performed the real thing by a noticeable margin on a box that cost half as much.

    That’s got to be high on some list: “Top Ten Signs Your Technology is Doomed.”

  57. “In 1998-1999 Sun didn’t have any competitive advantage that could justify their nameplate premium”

    So, instead of being overpriced and non-competitive, they decided to go for overpriced and even more non-competitive, because whatever little exclusive Solaris features (read: competitive advantages) they had were open sourced and given away to the world for free.

    Look, I never meant to say Sun was a well-run company. Urban legend has it that Dilbert corp was modeled first after Sun and afterwards after Nokia.

    But pony-tail dude’s embrace of the Open-source Underpants Strategy of
    1. Open-source all the crown jewels.
    2.???
    3. PROFIT!
    accelerated the decline.

  58. Urban legend has it that Dilbert corp was modeled first after Sun and afterwards after Nokia.

    Scott Adams worked at Pacific Bell, and having done some contract work for them, I have no doubt that was his first inspiration. I worked with one women who worked with him, and used to have some of his early cartoons and doodles. She kicks herself for not keeping them.

  59. Whatever the performance or lack thereof, this demonstrated they weren’t truly a vendor of “enterprise” level computers.

    Can we stop pretending that “enterprise” (in a computing context) means anything besides software or hardware that is marketed to people much higher on the org chart than its actual users, and specifically that the term implies NOTHING about quality, reliability, or performance? Enough fuck shit stacks have been foisted on unsuspecting office drones under the rubric of “enterprise” now that the use of the term should no longer cultivate an expectation of quality, but rather, its opposite.

  60. Jeff Read: That makes me think of Sharepoint. Having once helped build an internal company website based on it, I decided it was something that no office drone would actually ask for, but was simply a tangled mess foisted on them because it was from Microsoft.

  61. Jeff Read: while I didn’t make it at all clear, my emphasis on “enterprise” was vendors, not computers. In the sense that IBM was legendary as an “enterprise” computer vendor because they did a good job of taking care of their customers, and would move heaven and earth if the hardware they were leasing you had a problem like this. Sun was trying to take this market and showed it simply wasn’t in their DNA.

    As for your general point, I at least notice differences I’m willing to pay a premium for in “enterprise” disk drives and “sever” and/or “workstation” motherboards and CPUs. Minor details like EEC, or 5 year warranties for the disk drives, and for the true enterprise ones, SCSI/SAS support and 10K or 15K rotation rates.

  62. Regarding Debian and dash — sure, they’ve escaped the bash bugs, but that’s really just security through obscurity. They’re vulnerable to any exploits in dash, and since dash is a whole lot less used than bash (I hadn’t even heard of it until now) there are probably a lot fewer eyeballs looking at the code.

    As for the original topic of commoditization vs open source, yes, Sun was killed by commoditization, but it was commoditization on two fronts at the same time. Intel’s x86 commoditized their hardware away while at the same time Linux commoditized the OS away. The open source nature of Linux is what enabled it to be a commodity OS, so yes, in that sense “open source” *did* kill Sun.

    Hardware commoditization without software commoditization would have hurt Sun, but probably wouldn’t have killed them in the long run. Look at cisco — Broadcom has been commoditizing the network chips used in routers and switches for a decade. At this point, Broadcom builds chips that are good enough and cheap enough that even cisco uses them in their own low-to-mid range products, whereas a decade ago every cisco box had a cisco forwarding ASIC in it.

    So why is cisco still going strong compared to where Sun was a decade after the 386 really became popular? It’s the software stacks in IOS (the real one, before Apple swiped that trademark as well) and now NX-OS. The software to run a switch or router is not (yet) a commodity. SDN may be trying to change this, but it hasn’t happened yet.

    1. >The open source nature of Linux is what enabled it to be a commodity OS, so yes, in that sense “open source” *did* kill Sun.

      To see where I think this goes wrong, consider an alternate history in which OSes commoditized but hardware did not. Would Sun still have a viable business in that timeline?

      I think the answer is “Hell, yes!” The computer market would be stratified by hardware type – almost everything would run Linux but there would be large differences in performance and reliability between a Sun Linux box and a generic PC running Linux. Sun would live on those differences

      An interesting feature of this scenario, BTW, is that Sun probably decides it doesn’t need to spend the NRE on Solaris and transitions its own customers over to Linux. Because if your hardware designs are truly a fortress around your business, why not?

      So: Hardware commoditizes, Sun goes pfft. Hardware doesn’t commoditize, not only is Sun still strong, it’s strong enough not to need proprietary lock-in from its software.

  63. @dht The JCP launched in 1998, Sun open sourced Java in 2007. If you’re arguing that Sun could have better managed Java licensing you might have a point, but that has nothing to do with Eric’s argument about open source versus commoditization.

  64. The computer market would be stratified by hardware type – almost everything would run Linux but there would be large differences in performance and reliability between a Sun Linux box and a generic PC running Linux. Sun would live on those differences

    An interesting feature of this scenario, BTW, is that Sun probably decides it doesn’t need to spend the NRE on Solaris and transitions its own customers over to Linux.

    POWER?

  65. @Ian:
    > Regarding Debian and dash — sure, they’ve escaped the bash bugs, but that’s really just security through obscurity. They’re vulnerable to any exploits in dash, and since dash is a whole lot less used than bash (I hadn’t even heard of it until now) there are probably a lot fewer eyeballs looking at the code.

    Dash is designed to be lightweight Bourne and POSIX compatible shell. This means lack of features such as executable code in environment variables – minimal features means smaller area of attack. Because the code weight less, eyeballs per line of code is probably not that much smaller…

  66. Eric: My point is that Sun’s death required the commoditization of all of their competitive advantages — both the Sparc hardware architecture and the Unix-based software. I agree that in your hypothetical world of Linux + non-commodity hardware Sun would still be around, but I also maintain that in alternate history where Linus Torvalds decided to become a WRC rally driver instead of a hacker and AT&T/USL won the BSD lawsuits, we’d also see Sun going strong, selling Solaris for people to run on white box servers.

  67. As for the original topic of commoditization vs open source, yes, Sun was killed by commoditization, but it was commoditization on two fronts at the same time. Intel’s x86 commoditized their hardware away while at the same time Linux commoditized the OS away. The open source nature of Linux is what enabled it to be a commodity OS, so yes, in that sense “open source” *did* kill Sun.

    The OS was already a commodity, though.. Even if one deletes both Linux and *BSD, the multi-vendor nature of SVR4 and the commercial clones (like OPENSTEP) pretty much meant the OS was already commoditized in the early 1990s.

  68. (And that’s assuming we’re deleting the whole FSF/GNU project from history, not just having Linus go off racing. Because somebody would have eventually added a kernel.)

  69. @kurkosdr
    …in other words, *exclusive* features. That’s how you beat commoditization.

    You assume that “exclusive features” can be stable over any period of time, when in fact there is competitive pressure for everyone else to figure out out to include them as well. Even if the company can constantly figure out new exclusives there is still the problem that anyone who doesn’t need *those* features will go to the disruptor, which reduces the resources that the disruptee has to work with, making it harder to innovate……

    The nature of a disruption means that as long as the entrant has even the tiniest foothold he can expand. This means that the giant must be willing to completely gut itself on a regular basis in order to fill Every. Single. Niche.

  70. The first thing to say about Sun’s failure is that it all depends on where you want to start; Sun’s failure was baked in long before the 2000 fall from profit. Since the eventual winner in IT CPUs is Intel, perhaps Sun should not have doubled down on SPARC in the 1980’s.

    Windows NT drove Solaris off the corporate desktop, partly because they didn’t fight. They’d started moving off the desktop towards the data halls by then.

    Linux ( & Windows ) enabled x86 CPUs to enter the Datacentre and compete with Sun’s Server offerings in their, by then home space. Sun moved up market, with the E10K and its successors and looked to monetise their intellectual property through hardware sales. They recruited a lot of marketing and sales from DEC who only understood proprietary strategies.

    They lost the Universities and their students to Linux for two reasons. The enclosure of Solaris made it unattractive to Universites since they couldn’t re-engineer (or even tune), you had to buy a compiler and it ran best on SPARC which they didn’t want for their own use, mainly for cost reasons.

    Sun’s Linux experiments were just that. History shows that Sun should have built a desktop and a platform. It’s what customers see. It’s why Apple are a successful UNIX company.

    With respect to Blades and x86, the sales teams wouldn’t sell them against SPARC, the blades were killed in the corporates by the network i.e. backplane (software) architectures adopted, and in the Universities by cost. The number of people in Sun who spoke Grid could be counted on the fingers of one hand which was another reason the blades didn’t take off. Sun and its customers were wedded to the RDBMS, which was a scale up, not scale out solution and it was taken by surprise by both the HPC compute grids and the new age data grids as they became the future volume deployment paradigms.

    Announcing the non delivery of Solaris 9 on x86 was another act of stupidity.

    I am firmly on the side of those who argue it was commoditisation not open source that killed Sun. HP’s survival is a proof point, also showing that dreadful management isn’t necessarily fatal. Sun recognised the sedimentation and commoditisation tendencies but chose to try and stay in a high margin business; scale up hardware.

    I am also influenced by the arguments of Carlotta Perez and that the tendency to monoploy and the squeeze on super profits i.e. you can’t make money from hardware, is just part of the evolution of the silicon age Kondratiev curve.

    While the long term opportunism of Sun’s leadership didn’t help, I reckon open sourcing Solaris and other products, although not all of them, staved off the inevitable, I can’t see a counter factual, if that’s the word we’re using, which suggests that a proprietary software strategy would have saved the company. Too many decisions that killed the company had already been taken. Innovation was happening elsewhere.

  71. Java is precisely what turned “$5 billion company into a $600 million company”.

    Smart companies commoditize their complements. Sun commoditized their core business (hardware) which is obviously totally insane choice. Open source had nothing to do with that.

    Java was designed to make sure you could write you software once and run “anywhere”. Practically speaking “anywhere” never materialized, but you COULD run it on Windows, MacOS, Solaris, and, [later] Linux. The hope was that Sun will build superpowerfil servers and that everyone would eventually migrate on them from Windows.

    The only problem: that hope never materialized and people migrated from Sun servers to Windows and Linux servers instead. Java is half of what of made it possible (the other half was obviously Linux, but while Linux allowed migration without Java, Windows migration needed Java to succeed).

    Note: Java was actually proprietary at that time. When Sun open-sources Java and [later] Solaris the writing was already on the wall: most guys have already abandoned Sun hardware at that point.

  72. Steven: “(And that’s assuming we’re deleting the whole FSF/GNU project from history, not just having Linus go off racing. Because somebody would have eventually added a kernel.)”

    Really/ how long did it take the HURD to reach even a minimally bootable state, and how much Linux infrastructure was needed to get there? The FSF is constitutionally incapable of doing an OS. If they had, they would have gotten there long before they did, and wouldn’t have to horn in on Linus’s act to get credit.

  73. Jeff Read: while I didn’t make it at all clear, my emphasis on “enterprise” was vendors, not computers. In the sense that IBM was legendary as an “enterprise” computer vendor because they did a good job of taking care of their customers, and would move heaven and earth if the hardware they were leasing you had a problem like this. Sun was trying to take this market and showed it simply wasn’t in their DNA.

    I still contend that “enterprise” is a weasel word, intended to make you think the company is as reliable as Big Blue circa 1969 and that their products are specially designed to be more scalable and reliable under heavy loads. It’s like “AAA” in gaming: it actually means a game wuth a huge bloated development team working fuck-your-personal-life hours and an even bigger marketing budget. But game marketroids would rather you believe it means “a fantastic game”. And Sun was far from the only guilty party here. I heard Microsoft mentioned, and I recall at least one company that set out to build proprietary “enterprise” blog and wiki software, in an era when WordPress and MediaWiki were already handling an internet’s worth of traffic.

  74. I still contend that “enterprise” is a weasel word, intended to make you think the company is as reliable as Big Blue circa 1969 and that their products are specially designed to be more scalable and reliable under heavy loads.

    They are. Big companies pay a lot of money for enterprise class support.

    It’s like “AAA” in gaming: it actually means a game wuth a huge bloated development team working fuck-your-personal-life hours and an even bigger marketing budget. But game marketroids would rather you believe it means “a fantastic game”.

    It may or may not be a fantastic game but typically AAA titles have very high production values in terms of cutscenes, rendering and voice acting.

    It’s like hollywood blockbusters. It may or may not be a great movie but it will typically have great FX and lots of explosions.

  75. I was working for Intel in the early 1990s and we were worried. We had poured everything into the Pentium, and it was doing great against x86 competition.

    Then we were shown a performance chart of an alpha running windows and various apps through software emulation and it was 2 to three times faster than our top Pentium running windows and the apps native.
    To say we worried is an understatement. We just concentrated on getting sales now until the market figured out how bad x86 was.
    Intel had several RISC projects at all stages of development waiting for that day.
    I was gone when the alpha group went on sale, but it is no accident Intel went out to buy that group with laser focus.
    We also we were about our Ethernet card business being taken out by alpha since their Ethernet card was mostly software compared to an Intel system. 3com ended up getting us there until moores law integrated Ethernet into the south bridge.

  76. Jay: Note I said “somebody”, I did not specify the FSF/GNU project as the somebody.

    If it’s 1994 and there’s no Linus and nobody else has started their own toy kernel project in the interim and the BSDs have lost to USL, well, somebody (likely from the *BSD community) hacks together some sort of kernel for the GNU userland on x86 (perhaps by taking Mach and expanding it into a normal kernel that looks something like XNU). And since the same university Internet-and-Usenet infrastructure is around, you get a system that the FSF calls GNU/[Kernelname] and everybody else calls [Kernelname].

  77. I risk being laughed out, but: the i386 also provided a different kind of disruption for the “bitty boxes” like the Commodore 64 and the actually not so ridiculous Amiga product line. It was not from below, for the C 64 it was from above, and for the Amiga… sideways?

    My point is, as an Amiga fan back then I saw it as something better getting replaced by something worse and it was not even cheaper, the graphics in my games were worse (the resolution chunkier), musical abilities were worse, I had to use DOS when the Amiga already had a point-and-click multitasking OS with overlapping windows, and so on.

    This was, I think, a disruption of modularity and flexibility. The PC had this component based architecture, you could buy in parts, from different vendors, and build yourself, or replace what needed, the Amiga was a lot like todays notebooks and tablets, buy in whole, send to the service center, and at the end throw away.

    The point is that sometimes people seem to prefer this kind of modularity EVEN IF the price is higher and the features lower? How does that actually work, does anyone know?

  78. >Everybody liked Sun machines. I liked Sun machines, pre-386, when I could get my hands on ‘em.

    Er, not true. Which Sun machines? Some of the people who cut their teeth on diskless Sun 3’s, no they did not like Sun machines.

    Before it sold big RISC iron, Sun got its start as a purveyor of nasty, cheap-but-good-enough crap.

  79. The 8-bit boxes were wholly obsolete by the time the i386 was a factor. The Amiga was not, but yes, it suffered due to not using Intel CPUs and being a closed platform based on custom multimedia chipsets – thus, high NRE costs. Even the later Amiga 1200 / 4000+ model had the same sound chip as the old Amiga 1000 or 500, for instance.

    DOS actually had lots of point/click GUI shells besides Windows – think GEM, DesqView etc. Similar in some respects to current Linux, with its variety of window managers and desktop environments. But yes, it was less focused on multimedia and soft-realtime interactivity, compared to the Amiga.

  80. *Bump*

    esr wrote: “And, until the dot com bust, VA ate Sun’s lunch in the low and middle range of Sun’s market.”

    I wonder who else ate their lunch. Microsoft, HP, Dell? I weren’t there, im just asking.

    “Note carefully here the distinction between ‘adding value’ and ‘extracting secrecy rent’.”
    Why can’t a company do both? Isnt that what IBM does with their linux server offerings? Isnt that what Apple does by giving away Darwin source code, and hope that most customers are happy using the pretty jail on top of it? Or they can just throw out the entire OS X and run their own OS of choice.

    1. >Why can’t a company do both? [add value and extract secrecy rent]

      Many do. My point is that while both of those look good from inside the company itself, only one of them is attractive to customers. The other is actually repellent.

      The relevance here is that Sun open-sourced a bunch of stuff at least partly as a signal to customers that it was giving up secrecy rent in favor of genuinely working to add value. (There were other reasons, too, including reassuring potential customers about the future-proofness of their purchases.) This was a sound general choice that happened to fail in this case for particular reasons.

  81. Hardware commodity is definitely the big culprit here.No one wants to pay a premium for something they have to pour tons of resources into for it’s life cycle to operate it.Apple’s premium price isn’t subject to this,because it’s a different market,it’s not “enterprise”.There’s no TCO shadowing overhead.I don’t know if open source can be argued to have helped sustain Sun or not,but IBM’s Eclipse Java certainly helped pour salt into the wound.Solaris and ZFS had (and still has) a lot of acolytes, but those people don’t necessarily have a say in purchasing enterprise hardware.There are many institutions that simply refuse to use anything but Unix for their workloads,but they will cling to their legacy systems out of tradition as well.The place where hardware is most likely to sell is in the low to mid range, and x86 simply dominates.Sun tried too hard to remain a hardware company,they couldn’t adapt,so they died.Oracle has the diversity and customer support+marketing to continue to sell no matter what.

Leave a comment

Your email address will not be published. Required fields are marked *