And Dtrace, NFS, ZFS… truly a shame.

Larry Ellison made fun of cloud computing. Now he tries to sell his own cloud after giving competitors decades of headstart.

> They were very far ahead.

They weren't, though, not really. The tech industry just moves too fast for that kind of pronouncement. It's absolutely true that a Sun-3/75 in 1986 was a revelation[1], but every one of those features was cloned as free software and running on sub-$100k hardware by 1994. They likewise had a brief moment with SPARC in the early 90's where their machines were the fastest you could buy, but Intel demolished those hopes too. Their swan song was the early dot com boom, where they still held onto the reputation as "The Best Unix" and sold a ton of web server hardware to the more conservative outfits. But we could all see the writing on the wall even then.

I mean, it's true they had some great stuff toward the end. But not that great, and generally not marketed in a way that would help them. Containers are actually a good example. Solaris Zones were... just a weird circus act[2]. I don't know any major players who made bets on the technology. And, again, within 2-3 years Linux had cgroups and LXC running in a more performant (and much cheaper to deploy) environment and that's what Docker picked.

[1] Faster than a VAX, which was still the reference Unix box. Megapixel framebuffer with windowing and competent terminal emulation. Every box was on the Internet. Every box could see the rest of its network over NFS (and most were deployed diskless). FWIW: anyone know where I can get one? Would love one for the collection, but most seem to have disappeared (government and university equipment tends not to end up on eBay...)

[2] Edit because this was misunderstood: I'm saying it was an oddity with minimal uptake, not that it was chaotic or bad as a matter of design.

Solaris Zones are still far superior to LXC. It's hardly a circus act and SmartOS is being heavily used in many places.

Solaris 10 vs contemporary Linux 2.6-era kernels handled memory-pressure situations so much better, that you essentially needed more RAM on the Linux system just to avoid low-memory situations to begin with. Been there, done that :)

Linux won because it was a) free and b) aligned (simply because so many had and were using Linux) with Intel x86, and Intel's fabrication technology is/was world-leading. SPARC CPUs on the same fabrication node would still be amazing - but it isn't on the same node...

> every one of those features was cloned as free software and running on sub-$100k hardware by 1994

?

Zones weren’t released until nearly 10 years after 1994. Same for dtrace and zfs. Solaris was embryonic in 1994. Linux existed but wouldn’t take over servers until c2000-1 (as evidenced by Sun’s massive profits through the end of 1999).

What features are you referring to?

When you say:

> Solaris Zones were... just a weird circus act. I don't know any major players who made bets on the technology.

Joyent [1] readily comes to mind. IIRC, that kind of isolation was one of the selling points of their hosting service before they eventually shut it down.

(Of course I'm fully aware that one of Joyent's founders was ex-Sun and comments around here too :) )

1: https://en.wikipedia.org/wiki/Joyent

Presuming that you are referring to me, I am definitely not a founder of Joyent! That said, my rationale for coming to Joyent was not at all unrelated to some of the parallels I saw in Sun[0] -- and it is probably worth reading my own farewell to Sun as well[1]. (Also, because I can't resist completing the arc into the present, see becoming an ex-Joyeur[2] and starting Oxide[3].)

[0] http://dtrace.org/blogs/bmc/2010/07/30/hello-joyent/

[1] http://dtrace.org/blogs/bmc/2010/07/25/good-bye-sun/

[2] http://dtrace.org/blogs/bmc/2019/07/31/ex-joyeur/

[3] http://dtrace.org/blogs/bmc/2019/12/02/the-soul-of-a-new-com...

My bad! I must have equated your high profile association with Joyent with being a co-founder.

Thanks for swinging by to make the correction!

Ha, no worries! You're not the first to make the assumption, so I wanted to be sure to clarify. ;) And thank you for pointing out Joyent's use of zones in production. Indeed, it was Joyent's use of zones that attracted me to the company, and not the other way around...

I really would like to know which Linux or BSD distribution did containers before Tru64, HP-UX and Solaris.

FreeBSD had jails way before Linux had containers, but they did come after Solaris Zones.

The point discussed here is how FOSS did it first.

FreeBSD jails were released in 2000 in FreeBSD 4.0. Solaris Zones in 2004/2005. In fact, the original Solaris Zones paper cites jails as inspiration.

Jails: Confining the omnipotent root (2000): <https://people.freebsd.org/~bapt/pdfdocs/papers/jail.pdf>

Solaris Zones: Operating System Support for Consolidating Commercial Workloads (2004): <https://www.usenix.org/legacy/event/lisa04/tech/full_papers/...>

Originally, jails didn't virtualise the network stack. That was added later, in FreeBSD 8.0 with VNETs, which were buggy for a long time, but went through extensive testing and are enabled by default since FreeBSD 12.0 (quite recently). Solaris Zones went a step further than the original design, and run a separate network stack for each zone from the start, years before FreeBSD did.

BTW, recommended talk about those topics:

Papers We Love: Jails and Zones: <https://www.youtube.com/watch?v=hgN8pCMLI2U>

Meanwhile in 1996 with HP-UX 10.24 update.

https://en.wikipedia.org/wiki/HP-UX

https://community.hpe.com/t5/security/hp-virtual-vault/td-p/...

OpenVZ and Solaris Zones both came out in 2005; however OpenVZ was never accepted into the kernel.

HP-UX containers came out in 1998, if I still recall it correctly, and Tru64 already had some stuff into that sense with its microkernel and secure processes design.

> They likewise had a brief moment with SPARC in the early 90's where their machines were the fastest you could buy, but Intel demolished those hopes too.

Intel was slow in the early 90s. Intel (or more specifically amd64) started to become competitive on the lower end in the early 2000s when the Itanium hype reached its maximum. If you look at SpecInt 2000 an 1.4 GHz Intel is equivalent with a 500 MHz PA-RISC.

I think conceptually they were really good and innovative. They just had a hard time making money off it.

By the way what are you looking for? A VAX or a SPARCstation? VAXes are indeed very hard to come by, and I didn't realise they were ever the "reference unix box" - weren't they more common for VMS? But somehow I never came across them back in the day. I think part of the reason is that they mainly sold servers, they weren't that big on the workstation market.

Sun, SGI and HP-UX I used a lot, HP-UX the most. I still have two HP-UX workstations at home.

I have a few SPARCs, actually. Those are fairly easy to find. I want a Sun-3, because to my mind that's the truly transformative device. In principle I could sit at a 38-year-old Sun 3 today and do my actual job productively, with pretty minimal workflow changes. It's one of just a handful of Devices That Changed Computing Forever, and in principle one that should be accessible to a collector. But they're rarer than I'd have expected, especially given how many I saw sitting around at school in the 90'd.

And yeah, the desktop VAXen were almost exclusively for VMS. But the standard Unix box in the early-mid 80's that everyone talked about when they discussed/benchmarked/whatever the platform was a 11/780 running 4.2BSD. Those aren't meaningfully collectible for an individual, obviously.

I love collecting old machines also <3 I have a PDP-8 and 11 replica (from obsolescence guaranteed), some of my first computers, a G3 iMac of which I just love the design, HP-UX boxes because they were really formative for me with Unix. I also owned some Sun hardware (a SPARCstation 5 and an X-Terminal) but sold it when I moved to a small apartment. I still regret this.

But indeed the servers are really hard for collectors because of the size. Similarly with the real PDP's of the day. We had a real PDP-11 at the computer museum where we volunteered though we never turned it on. Too afraid of a cap blowing and damaging more.

HP had some nice small servers later on, but they were not very influential so like you I don't really see a benefit in collecting them.

There is also the MicroVAX. We got one (a MicroVAX II) at my place of work shortly before I left in 1986. It ran Ultrix, a BSD derivative. As I recall it, the box would not have been too big for personal use, though the price tag probably was on the steep side.

From before, we had a VAX 11/750. It, too, was a freestanding box in the computer room, possibly close to a cubic meter in volume. Not something I would want at home! It ran VMS, which I found clunky and difficult to work with. But we got a Unix emulation layer called Eunice installed on it, which made it more bearable.

I'm curious; I know the Sun workstations were generally well-regarded, but what was it about the Sun-3 specifically, that made it such an important quantum jump?1

The joke (at Sun) back in the day was that a Vax MIP was constant, like an electron volt.

They were bought by Oracle for salvage as an imploding shambles. If not Oracle, they would have ended up as SGI - merely a name for some white box vendor.

The problem here is the confusion between the wonder of late 80s workstation computing and the harsh realities after the Internet Boom.

Oracle's primary interest was suing Google over Java in that acquisition

Regardless, Android Java should never had existed in first place, now we have Kotlin as Google's C#.

Sun did not sue as they were already out of cash after Google screwed them over.

Not only did Google contribute to Sun's downfall, they didn't even bother to buy Sun afterwards.

They only have themselves to thank for the lawsuit.

What innovation was still happening at Sun by the time Oracle stepped in?

GraalVM was known at the time as MaximeVM on Sun Research Labs, for example.

What would have happened is the company would have folded.

Sun's problem's happened long before they were purchased. Their number one problem is that they were a hardware company and never figured out how to be a software company. It was Linux and cheap Intel hardware combined Sun's lackluster support for x86 Solaris that doomed them.

Yeah. Sometimes I wonder how everything would have turned out had they gone all-in on x86 starting with the Sun386i in 1988 and never developed SPARC.

Or someone like Motorola could have done RISC sooner. (They did eventually with the 88K but Sun, IBM, etc. had all gone with their own RISC chips by then.)

Someone did do RISC sooner; for example, MIPS shipped R2000 in 1985 and Clipper shipped C100 in 1986. SPARC and 88k were introduced within a year of each other as I recall (87/88).

Yeah, MIPS is probably the best example and did get some uptake. But the big vendors mostly went their own way. Although I was involved in that history to a degree, I'm not entirely sure why history played out the way it did.

Well...any history will attract a torrent of nitpicking, so this is my extremely reductist take. Motorola, IBM & Apple decided to join up on a common HW platform called "PReP" which was POWER/PPC based, so Motorola dumped 88k to play ball with that. PReP was a flash in the pan, but PPC briefly shone as Apples processor, and is still around in various forms (mostly embedded now, but POWER still ships in volume). Similar trajectory with MIPS. Clipper was never going to be a big player; it was a niche within a niche. SPARC, HPPA & Alpha got steamrolled by much cheaper, performance competitive x86 derived processors. In the end, most volume customers said "can it run my software fast and cheap" and by sometime in the 90s that was clearly an x86 (unless your name was "Apple", but that's a different history).

SGI was on MIPS, and if anything they fared worse than Sun.

They were one of the few companies in the 2000s that could have utterly owned both the browser space and the cloud space if they had worked at it and persisted.

They would have gone under, Java would have been fossilized at version 6, GraalVM would never have happened, CLR would be left as the only high end polyglot runtime with top GC and JIT compilation algorithms.

Sun was a innovator throughout the 90's. Many colleges, universities, and startups used Sun hardware. By the mid 2000's, they were on the way out. The market was flooded with barely used Sun hardware after the dot-com collapse. At the same time, Linux was now "good enough" that less-and-less people needed Solaris. Of course, some enterprises still needed it. The only thing Sun was hanging on to - Java - was freely available.

They were already doomed at that point. I was involved in several major purchase decisions during that period and Sun were totally out of touch. If they were quoting hardware that had an x86 equivalent, then they were overpriced. If it was something that didn't have an x86 equivalent, like they had for a while with 64 bit, the prices were atrocious. Then Lintel moved to 64 bit and it was all over. The price/performance equation was broken, but Sun, for whatever reason, kept on like nothing had happened. I, and I'm sure many others, tried to show them that they were uncompetitive, but you were dealing with reps working from a price sheet and citing the same old mantra that Sun was inherently superior. For years when I would tell peers that the Sun equipment was just throwing away money they wouldn't believe me because they hadn't done the benchmarking. Really, if people did proper benchmarking and didn't just 'buy what they know' without questioning, it all would have unwound even more quickly. They certainly had some good tech, and were not wrong about the advantages of containerization, which came full circle with linux containerization and docker, but IBM mainframes had similar virtualization capabilities in place long before Solaris zones, so it wasn't something that was a game changer. Ironically, it worked against them because even though they were right, it was seen by some as Sun touting their way of doing it because they didn't have a viable solution for the prevailing VM direction of hardware virtualization. Basically, they began by offering the best price/performance and innovation, but then died trying to be a 'premium provider' without the goods to back it up, and market forces then do what market forces do.

I agree that they were doomed by that point as well primarily due to Linux on faster and cheaper x86 hardware. My UltraSPARC-III-based Blade 1500 that was an incredibly overpriced workstation that only ran Solaris well, and wasn't on par with the performance of x86 at the time.

However, I do remember buying an Opteron-based Ultra 20 because it was cheaper from Sun than anyone else and had 100% Linux support, which everyone seemed to be migrating to from Solaris. By that time, I'm sure there was no clear market direction for Sun software- or hardware-wise, and everything was too little too late.

Sun's sales guys were killing it right up until their last breath. I can clearly remember thinking to myself "why does this Sun/Solaris node cost $32k when this equivalent Intel costs $10K?" One day it was the in-thing, the next pooof Solaris was GONE.

Before Docker came along though, containers were mostly just a more lightweight partitioning techniques than VMs which they had largely lost out to. See also Virtuozzo, BSD jails, and some of the early container work on Linux. But Docker followed by Kubernetes really made containers into a different cloud-native development and deployment model rather than just a lighter form of VM..