What I find fascinating about Sun is how fast its ride was. They launched their MVP in 1982 which was really just a bare 68000 board with a kluged together software suite. The second generation Sun 2's were like a year and a half later, running virtual memory on 4BSD, the 68020 made the Sun 3's in 1985 faster than a VAX, and suddenly Sun was The Premier Unix that everyone targetted.
The next few years (up through 1991 or so) would see the launch of SPARC[1] and all the Unix goodness we all still work on: shared libraries, NFS, RPC, pervasive IPv4 networking, basically everything about the modern datacenter software environment dates from these few years at Sun.
And then, sort of out of nowhere in the mid 90's, Linux distros running on P6 boards had essentially cloned it all on hardware 1/10th the price and the end had begun. Sun would continue to make a lot of money through the doc com boom, but their status as the thought and innovation center of Unix hit a brick wall.
The story of the end was all about Java and Oracle and datacenter markets. And IMHO it's not that interesting. What the hell happened to Unix?
[1] In hindsight it was just a flash in the pan, but the RISC arrival in the Unix world was shocking at the time. Even though in hindsight the workstation vendors had at most a 3-4 year lead on Intel at the peak and would rapidly fall behind.
Yeah in retrospect, it feels somewhat inevitable to me that Linux (or something similar if that hadn't happen) would displace it all and demolish the business model of "Unix as commodity", given Unix itself was clearly initially aimed at trying to popularize/democratize a set of technologies/techniques/concepts that had been previously locked up inside larger corporations and projects. The motive force of "getting this out there" was there, and was bound to escape the workstation maker's clutches.
I didn't live through the minicomputer era, but definitely grew up in the "Unix [and then Linux] ascendant" era and was an early adopter (as a user) of Linux on my 486. We just wanted what all the cool kids [err, adults] had. I spent many hours fine tuning my X11 environment to look like the screenshots I saw in UnixWorld of real Unix workstations, etc. ... without doing any actual "real work" with it...
Looking back, it was inevitable that Unix would become less and less a sale-able commodity and more and more a free standard that hackers would just ... assume.
I'm not sure how Sun could have saved itself without just turning itself into a services company, just too hard to win on economies of scale making actual hardware. They made hay while the Sun(tm) shone, I guess.
> Looking back, it was inevitable that Unix would become less and less a sale-able commodity and more and more a free standard that hackers would just ... assume.
I wonder if the operating system[1] has turned out to be the ultimate expression of Steve Jobs's quote about Dropbox: "feature, not a product". A means to an end, with the end being where all the value is.
Everyone talks about Microsoft retaining the rights to market DOS independent of the IBM license being the most important business deal of all time, but Microsoft producing its own applications may be even more important in retrospect.
[1] I wrote "Unix", but of course Windows has been de facto free, even when not purchased with a computer, for some time
> Microsoft producing its own applications may be even more important in retrospect.
I remain convinced that Microsoft Excel is the most important thing they ever built. You could replace Windows for Linux or vice versa and the world would hum along more or less the same. But entire economies are essentially running on what people do with Excel.
I did a podcast with ex-Sun Bryan Cantrill and sjvn a few years back about the inevitability of open source as part of a series. Bryan’s take was basically, if not Linux, BSD. Of course, there’s also the school that Microsoft basically wins which many assumed at the time.
I mean, if you look at commercial UNIX, well to start it all sources from AT&T at some point; they weren't permitted to sell it, so they gave it away more or less.
BSD (and others) took it and improved it.
Everyone (including Microsoft) took at least the BSD socket stack, at least for a while.
Commercial UNIX competing against free community UNIX is a hard battle to win. There's a question of UNIX vs alles, but if UNIX lives, it's going to be community UNIX (or well Linux which is community UNIX alike).
I suppose there's an angle for commercial UNIX on specialized hardware; Apple is doing fine with that model; but it stopped being compelling for Sun --- commodified x86 servers are good enough that you can't build a business to support commercial UNIX on specialized server hardware (x86 or not) alone. Oracle Solaris exists, but as a non-customer, it looks like development has slowed significantly.
I think the root issue here is Joy’s Law [1]: “No matter who you are, the majority of smart people do not work for you.” Sun had a whole lot of very talented engineers working for them, but ultimately they were building a proprietary, vertically integrated system. When compared with the best memory makers in Japan and the best CPU makers at Intel and AMD and the loosely knit coalition of OS engineers working on Linux and all the Linux desktop engineers, they eventually found that the best engineers did not work for them.
[1] Ironically coined and named after Sun Microsystems founder Bill Joy.
I recall that in about '92 Intel had launched a project called Eclipse which was an x86-based workstation they were developing to compete against Sun. As with many Intel projects, it didn't get anywhere.
Never heard of that particular product, but in point of fact Sun's original core workstation market had been essentially destroyed by the late 90's by x86 boxes running Windows NT. Intel didn't have the product in the channel in 1992, but by 1996 it was clear SPARC's days were numbered.
Wrecking the workstation market was easy, because workstations had small numbers of CPUs, and small pools of RAM. Intel started fucking up that market in the late 1980s.
x86 wasn't competitive on the server side until K8 (AMD) and Nehalem (Intel).
So you have a real long stretch -- 1987-ish to 2008-ish -- where proprietary UNIX on proprietary architectures owns the server side, even though the workstation market is eroding the whole time.
Again though, after 1995-ish Sun just stopped "doing Unix", abandoned the community they created (who all trotted off happily to Red Hat et. al.), failed in their core workstation market, and basically spent their time milking server sales to conservative[1] IT departments who wanted to do "internet".
Their swan song ended up being Java, an interesting (but again poorly exploited) technology that had next to nothing to do with the environment on which it was incubated. Frankly Sun ran away from it so hard that Java ended up running best (!) on Microsoft Windows.
So basically it was 13 years, as I see it, from kids-with-soldering-irons-and-a-dream to world-changing-behemoth to company-your-grandparents-buy-from. That's fast even in Silicon Valley.
[1] The cool kids, obviously, were all running Linux in their datacenters already. Only the S&P 500 dinosaurs were buying Sparcservers, but there were a lot of dinosaurs.
Sun greatly revitalized "Unix" in the 00s! Need I refer you to Bryan Cantrill's screed about how OS research was not boring? The list of features that shipped in the 00s is amazing:
- DTrace
- FMA/FMD
- SMF
- ZFS
- the unified process model
- NFSv4
- CIFS
- and more
and this was while being hamstrung by a crappy SVR4 networking architecture that the networking team was able to kill off (thank goodness).
Some of these are things not yet re-invented elsewhere, others re-invented poorly:
- systemd is a bad SMF
- SystemTap is a bad DTrace
- eBPF is pretty cool but in
some key ways not as good as
DTrace
- ZFS remains unparalleled
Yeah, this turns into senseless flaming very quickly. But to be blunt, the fact that puts the lie to your point that all those technologies are "revitalizing" or whatever is that basically no one uses them[1]. They're interesting ways to win an argument on the internet (with which I won't engage), but not evidence that you're doing something actually important.
[1] Obviously people use them! But not at scale and not in such a way that it provides meaningful advantage over the people who don't use them. Again, they're fun things to argue about but not transformative in the way that early SunOS was.
The fact that they've been copied is telling enough: others needed things like those. Either Sun was too early with some of these or Sun couldn't capitalize on them, or both. Sun definitely was too early with certain things like cloud (the Sun Grid). Was SunOS 4 as transformative as BSD?
Again, 100% not going to engage on another senseless ZFS/dtrace/whatever platform flame. But this point is worth responding to:
> Was SunOS 4 as transformative as BSD?
SunOS was BSD! 3BSD was absolutely transformative. But 4BSD launched after Joy and much of the team were already working at Sun, and almost all future "BSD" innovations of note were actually SunOS features.
Sun took a historically important but otherwise obscure academic platform (total VAX BSD deployments as of 1982 were what, in the dozens?) and turned it into an industry-defining software environment that we still use today (albeit in cloned form).
This is revisionist. ARM didn't break out as an embedded architecture until a full decade later. At the time it was entirely forgettable, with no competitive parts in the workstation market and no software worth running (again, the center of the universe at the time was SunOS).
It's popular now to imagine that ARM had some magic ISA back in the 80's, but it was very much an also-ran through most of its life. The magic is inside Apple Computer, and quite frankly they could have made anything fast. They simply happened to have an ARM OS core running already, so they picked the architecture that wouldn't force people to recompile their iPhone apps.
The magic of super fast ARM cores is inside Apple, but ARM's general success has little to do with Apple. It seems like a large part of ARM's success is offering licenses for good hardware at a pretty low price. ARM doesn't "capture value" much, it seems to me.
Right, but SunOS on SPARC changed the world forever[1]. It's not really a comparable discussion.
[1] And then promptly imploded, and has been forgotten now even by people[2] living and working every day in the environment Sun created. That's the bit I was pointing out upthread.
[2] Who apparently think that the important story of that era is somehow the emergence of ARM?!
DEC indulged an enormous number of also-rans. It is from the perspective of 2025 that we remember some of the good stuff and forget all the bizarre mis-fires.
Off the top of my head:
* Two duplicate "high-end" VAX architectures (VAX 8000 vs VAX 9000), because no one wanted to choose between CMOS and ECL
* Three duplicate systems targeted at the high-end (Alpha, VAX 9000, VAX 8000)
* Two duplicate RISC+UNIX systems, because DEC was extremely late to market (MIPS 3000/5000 series vs Alpha)
* Two duplicate UNIX software packages, because DEC was really late to market (1970s ULTRIX ported to MIPS, OSF/1 on Alpha, and the never-fucking-released OSF/1 on MIPS because DEC just could not get their shit together)
* Four duplicate low end systems (MIPS, PDP-11, NVAX, Alpha were all sold simultaneously at the same price point!)
* A dozen utterly-failed microcomputer projects (Pro/3xx, Rainbow, etc)
DEC was not a particularly well-managed company. Their approach, for decades, was "throw shit at the wall and see what sticks." This worked fine right up until it didn't work at all.
It is also worth noting that Alpha, the "good" DEC initiative, was a failure. It lost a lot of money! market share never got out of the single digits.
Do I need to? StrongARM is pretty much the definition of an "also-ran" product, no? It had no particularly notable design wins, and while it sold in reasonable volume was a distant second to MIPS in the "consumer junk" segment[1]. DEC unloaded it to Intel, where it becamse Xscale, and Intel dumped it on Marvell. At no point did anyone really care much about it.
Even within the ARM world itself, it was ARM Ltd's CPU cores (also Qualcomm had some decent designs) that powered the architecture's way back to relevance on phones, out of which Apple would grow to dominate.
That's a little unfair. StrongARM did well in the WinCE market (which I assume you are referring to as 'consumer junk') and did very well in the embedded market especially as Xscale over several generations (PXA, IOP, etc). As an embedded chip with a relatively short lifespan that's reasonably impressive.
However, the idea that somehow it (or any of it's contemporary ARM kin) could somehow 'replace SPARC and save Sun Microsystems'...well, that's just laugh out loud silly.
1996 was far too early for what they were trying to make it for:
> The StrongARM was designed to address the upper end of the low-power embedded market, where users needed more performance than the ARM could deliver while being able to accept more external support. Targets were devices such as newer personal digital assistants and set-top boxes.
They'd be able to power a faster PalmPilot or proto-TiVo with it but this was years before the mobile design advances, let alone battery and screen improvements, that led to the iPhone.
What should we 'explain' about a false equivalence? Different processors for different markets? It was never much other than a mobile and embedded processor. Yeah, I suppose some folks thought is could be workstation PC, but how many RiscPC 700s were sold? By 1996, Sun had SPARC for, what, 10 years, and had just introduced UltraSPARC? StrongARM was never in the same performance ballpark on any dimension other than performance/watt.
I thought all the uncritical ARM fanbois had defected to RISC-V. Good to see some still carrying the torch.
The next few years (up through 1991 or so) would see the launch of SPARC[1] and all the Unix goodness we all still work on: shared libraries, NFS, RPC, pervasive IPv4 networking, basically everything about the modern datacenter software environment dates from these few years at Sun.
And then, sort of out of nowhere in the mid 90's, Linux distros running on P6 boards had essentially cloned it all on hardware 1/10th the price and the end had begun. Sun would continue to make a lot of money through the doc com boom, but their status as the thought and innovation center of Unix hit a brick wall.
The story of the end was all about Java and Oracle and datacenter markets. And IMHO it's not that interesting. What the hell happened to Unix?
[1] In hindsight it was just a flash in the pan, but the RISC arrival in the Unix world was shocking at the time. Even though in hindsight the workstation vendors had at most a 3-4 year lead on Intel at the peak and would rapidly fall behind.