back to article Confirmed: Oracle laid off 964 people from former Sun building

Oracle has filed a notice under California's Worker Adjustment and Retraining Notification (WARN) regulations that reveals it has recently made 1,008 permanent layoffs in its Santa Clara and San Diego facilities. The WARN notice (PDF) says the 964 Santa Clara sackings were made on September 1st, the day before the Labour Day …

  1. HCV

    It's a bit more than a "building"

    The Santa Clara campus was Sun's headquarters of record at the time of the acquisition. It's something like 60-80 acres, IIRC, and has 20+ buildings.

  2. Mark 85
    Facepalm

    I need new glasses..

    I read the last line as: “Lift and Shit Your Oracle Solaris Workloads to the Oracle Cloud.”

    1. FrankAlphaXII

      Re: I need new glasses..

      Glad I'm not the only one who thought they saw that. Though to be fair, shit and slowlaris tend to go together in my mind.

    2. a_yank_lurker

      Re: I need new glasses..

      I don't think you need new glasses, your version is more accurate.

      1. Mad Mike

        Re: I need new glasses..

        Whilst each persion will have their favourite(s) operating system(s), each has its pros and cons. Depending on the use case, some may be more suitable or better equipped to deal with things than others. Certainly Solaris had it's pros as well as some cons.

        The issue here is where do we want to end up. In the hardware world, we're pretty much in a monopoly position. Intel has the majority of the market and can largely drive it. Other manufacturers (IBM, Oracle etc.) are very niche. It's lucky that ARM has come along to drive some competition in areas, but this is limited and each company is to some extent staying in its own territory, albeit with occasional raiding parties into the other.

        Do we want the world of operating systems to become the same? With all these operating systems being stopped, are we likely to end up in a world with only RedHat and Windows as the operating systems for companies? Is this a good thing? Personally, I'd like to a healthy ecosystem of say 6 choices in each of hardware and software to choose from to ensure competition is strong and drive innovation. Once you're in a monopoly position, the need to innovate and enhance your software drops to some extent and you can just sit tight and harvest all the money.

        I really don't like this choice shrinkage we're getting more and more at the moment.

        1. Nate Amsden

          Re: I need new glasses..

          Solaris and other unix have been niche for a long time maybe a decade or more? A high value niche. Doesn't mean there is no money to be made still though. I'd say the same applies/applied for very high end storage arrays as well.

          Last place I was at that ran unix was 2006(HPUX on itanium and PA RISC before that). Though I work for smaller companies generally.

          In the linux space there are quite a lot of options depending on your business model. Ubuntu and CentOS remain very popular, and obviously lots of folks out there running other things that may have less formal support available. I haven't worked for a company that has been willing to pay for Linux support since that company in 2006 either.

          1. Anonymous Coward
            Anonymous Coward

            Re: I need new glasses..

            The niche Solaris / AX / HP-UX and all other Unix has, is large scale-up systems. Fact is that x86 does not scale well. Only recently the first 16-socket x86 servers has arrived to the market. Meanwhile we had 64-sockets Unix RISC servers for a long time.

            Sure, x86 has market dominance, but it is only for small servers. Small servers is enough today for many workloads. x86 has gotten better, so a 2-socket x86 can tackle larger workloads than before. But if you need large workloads, x86 is not an option as there are no x86 servers that can handle the largest scale-up workloads. Look at all the SAP benchmarks, the top list belongs exclusively to Unix. So the market for 16- or 32-socket servers are diminishing as x86 gets more powerful. And that is the problem.

            Typical large scale-up worklodas that can not be run on clusters such as SGI UV3000 servers with 10.000 of cores - are enterprise business software such as SAP, databases, etc. Clusters can only run embarassingly parallel workloads such as number crunching, but they can not run business workloads. Just look at SAP benchmarks, or TPC benchmarks - it is all Unix.

            But in the age of Big Data, is there not a need for large scale-up 16- 32-socket Risc servers? No, you can run it on clusters of x86.

            So large RISC servers need a killer app to stay alive. And that killer app were earlier, large business workloads. But x86 can almost do that today. Not many companies have a need for the largest business workloads. So RISC are not profitable anymore.

            Mainframes are also scale-up business servers, but Mainframes have carved out a (slowly shrinking) niche because of lock-in. There are numerous stories of companies migrating from Mainframes to x86 and saving loads of money and increasing performance many times, as Mainframes have much slower cpus than x86.

            RISC can not lock in, because of you can always easily recompile your software to cheap Linux/x86 and escape lock in. Databases are not easy to migrate, they have a huge lock-in effect, which is Oracle why can thrive (because of database lock-in).

            Mainframes are profitable and have huge lock-in. Oracle database are profitable and have huge lock-in. Both thrives. RISC has no lock-in effects, it is easy to migrate to Linux/x86 - so RISC will diminish.

            1. Mad Mike

              Re: I need new glasses..

              @AC

              "The niche Solaris / AX / HP-UX and all other Unix has, is large scale-up systems. Fact is that x86 does not scale well. Only recently the first 16-socket x86 servers has arrived to the market. Meanwhile we had 64-sockets Unix RISC servers for a long time."

              I don't disagree with much of what you say, but 16 socket x86 has been around for ages. IBM (when they made servers) produced them (from memory) 10 years ago at least. However, I do agree with your principle, as x86 generally doesn't have the interconnects required for effective scaling.

              "There are numerous stories of companies migrating from Mainframes to x86 and saving loads of money and increasing performance many times, as Mainframes have much slower cpus than x86."

              This isn't true. Mainframes have higher clock speeds than x86, but more importantly, they do more per clock tick than x86. When I was working on mainframe (some years ago now), I saw so many applications migrated onto x86 that then performed like a dog. The straight line performance of a x86 thread is nowhere near that of a mainframe thread.

              1. Down not across

                Re: I need new glasses..

                This isn't true. Mainframes have higher clock speeds than x86, but more importantly, they do more per clock tick than x86. When I was working on mainframe (some years ago now), I saw so many applications migrated onto x86 that then performed like a dog. The straight line performance of a x86 thread is nowhere near that of a mainframe thread.

                And here is a good lesson for young 'uns. It is not all about clock speed.

                I remember people frothing about how much faster x86 was compared to SPARC. When you actually built a x86 testbed to compare (yes, tested with both Solaris x86 and Linux) it ran like a dog. The x86 platform (yes this was long ago so even so called server platforms were not that much more advanced than desktops) just had nowhere near the I/O capability of the Sun box which run fine no matter (within reason) what you threw at it.

                1. elan

                  Re: I need new glasses..

                  ....and here I am curious about the upcoming nvme storage/mem tech. x86 arch will at best be suboptimal compared to other processor architectures (sparc).

                  Unfortunately intel has such market power in the meantime - they will deliver optimised/specialised processors as well.

                  Still running in circles: what was j.schwartz role in those days ?

                2. Anonymous Coward
                  Stop

                  Re: I need new glasses..

                  > When you actually built a x86 testbed to compare (yes, tested with both Solaris x86 and Linux) it ran like a dog.

                  Bullshit.

                  Either you post numbers or you can stop lying now.

                  "I remember when ..." doesn't count.

              2. Anonymous Coward
                Anonymous Coward

                Re: I need new glasses..

                @Mad Mike

                "...but 16 socket x86 has been around for ages. IBM (when they made servers) produced them (from memory) 10 years ago at least. ...However, I do agree with your principle, as x86 generally doesn't have the interconnects required for effective scaling...."

                These old 16-socket x86 servers have never been successfull because otherwise we would have heard of them when they took over the crown from Unix RISC servers for a fraction of the price. For isntance, the Bullion server with 16-sockets has an awful topology which makes it unfit for large workloads. It scales bad. Look at this picture how 16 cpus are connected to each other. In worst case, going from one cpu to the farthest cpu will take three hops going via BCS. This topology does not scale well to more cpus.

                https://deinoscloud.files.wordpress.com/2012/10/bullions-bcs.png

                Here we have the topology of the brand new Intel Xeon cpu. Look at the 8-socket topology. It looks contrived, unsymmetrical and it scales bad. What is the worst case, going from two cpus farthest away?

                https://www.servethehome.com/wp-content/uploads/2017/07/Intel-Skylake-SP-Platform-Architecture-Topologies.jpg

                Imagine these bad topologies for 8- and 16-sockets for x86 trying to handle 256 cpus as in SGI UV3000 Linux server. If you connect every other cpu to each other to miminize the number of hops, you would need 65.000 connections or so. That is not feasible. But if you arrange these 256 cpus in clusters, then it is doable. And that is why a large Linux server with 100s of cpus are always a cluster - because the topology sets limits. You see how difficult it is already to scale well on 8- or 16-sockets, try to do that on 256 cpus. There is no way each 256-cpu is connected to each other. They are arranged in large clusters, so going from one cpu to another has a very bad worst case scenario. And that is why you can not run large business workloads on large 256-socket servers. Because business workloads frequently talk to each other cpu all over the place, and you have to lock the data before accessing, etc. It becomes a nightmare for performance. However, if you run parallell number crunching workloads on each cpu separately from another, and does not have to lock nor communicate, then a cluster such as SGI UV3000 works fine.

                Compare these x86 topologies to a SPARC topology with 32-sockets. This RISC scales well, and worst case is two hops no matter how far away the cpus are. This topology scales to 96-sockets.

                https://regmedia.co.uk/2013/08/28/oracle_sparc_m6_bixby_interconnect.jpg

                .

                Regarding the Mainframes. Mainframes have excellent I/O, much faster than x86. But the Mainframe cpus are dog slow. I would never recommend to port a I/O-bounded workload from Mainframe to x86. But cpu bounded workloads should be ported from Mainframes to x86, because x86 cpus are much faster than Mainframe cpus.

                Dont mix up Mainframe I/O (excellent) and Mainframe CPU (worst in class). There is a reason IBM never releases benchmarks comparing Mainframes to x86. There are none. Nowhere. That is because IBM is trying to hide the fact that Mainframe cpus are dog slow. Try to google for SPECint2006 benchmarks for Mainframes. Impossible to find. Or any other benchmark is also impossible to find.

                Unless there is proof for a claim, you should not believe the claim. Otherwise you are superstitious as you believe in the words of a high priest. Be a scientist instead, and demand proof and benchmarks.

            2. Korev Silver badge

              Re: I need new glasses..

              So the market for 16- or 32-socket servers are diminishing as x86 gets more powerful. And that is the problem.

              No long ago if you needed 32 cores then you'd have to buy a 32 CPU server; these days a dual socket Intel server can give you the same core count.

              1. David Halko

                Re: I need new glasses..

                Anon>> So the market for 16- or 32-socket servers are diminishing as x86 gets more powerful. And that is the problem.

                Korev> No long ago if you needed 32 cores then you'd have to buy a 32 CPU server; these days a dual socket Intel server can give you the same core count.

                2005 - Sun released a 32 vCPU processor... this was truly the beginning.

        2. Down not across

          Re: I need new glasses..

          Do we want the world of operating systems to become the same? With all these operating systems being stopped, are we likely to end up in a world with only RedHat and Windows as the operating systems for companies?

          No.

          At least Poettering has not implemented some auto update, that cannot be disabled, into systemd...

          On a more serious note, I agree. We need the variety and competition.

        3. JBowler

          Re: I need new glasses..

          Solaris is SPARC, SPARC is dead.

          I toiled several years trying to deal with porting SUN's shyte to ARM (2 and 3 - this was the end of the '80s). SUN did not want to produce software that was portable, they wanted to sell SPARC; the supposedly portable aspects of stuff like NeWS was badly thought out and laughable. Well, Acorn was, in that sense, laughable too given that they bought it (as in paid money).

          The total misunderstanding that Intel have any hegemony is also laughable.

          I remember the words of one of the guys in Acorn who had a microphone hanging from the ceiling of his office (It wasn't connected to anything; he liked to screw with the idiots); something like "one in every washing machine". That stuck with me until I worked out the repost; "three on every desktop".

          Think about it.

          There are not just three any more, there are probably at least 10 in every Intel workstation. (10 ARM cores; it's clear some people haven't worked it out so my apologies.) At least 2 in every cellphone (never forget the graphics engine) more in every camera (remember you have to put an SD card in those; one in every SD card).

          This has been true for 10 years, "3" has been true for 20. The fools who design washing machines in the US still don't always put one in, sad (but the guy who said it didn't understand how incredibly stupid we can be in this country.)

          As for the original article, so what? Who gives a damn how Oracle repurpose their very valuable office space?

          John Bowler

        4. TVU Silver badge

          Re: I need new glasses..

          " Personally, I'd like to a healthy ecosystem of say 6 choices in each of hardware and software to choose from to ensure competition is strong and drive innovation. Once you're in a monopoly position, the need to innovate and enhance your software drops to some extent and you can just sit tight and harvest all the money".

          ^ I fully agree with this although all the signs so far coming out of Oracle this past couple of years indicate a regrettable path of decline and neglect for Solaris. If they are are going to ultimately phase it out then the very least they ought to do is open source it.

      2. Anonymous Coward
        Anonymous Coward

        Re: I need new glasses..

        “Lift and Shift Your Oracle Solaris Workloads to the Oracle Cloud. What could possibly go wrong?”

  3. Anonymous Coward
    Anonymous Coward

    The last hurrah of the 90s

    It is meet and fitting that Solaris has reached the end of the road, but also a little sad: capitalism's creative destruction in action. Feels like the last battle in the very long Unix war.

    I personally find it worrisome that OS selection is becoming a Linux monoculture. Yes, it is spectacularly good. But it also has spectacular baggage, and a monoculture with a weakness is a dangerous thing. How long before the kernel edifice collapses under its own weight?

    1. Hans 1
      Facepalm

      Re: The last hurrah of the 90s

      How long before the kernel edifice collapses under its own weight?

      What in Feynman's name are you mumbling about ?

      /* HINT: kernel modules */

      Besides, when it comes to choice ... BSD's are as good a choice as well ...

      1. Anonymous Coward
        Anonymous Coward

        Re: The last hurrah of the 90s

        Just playin' my bongos, @Hans 1 :)

        I'm not referring to the number of lines of code -- that is irrelevant, and as you point out, modularizing the kernel makes the actual number of used lines smaller. My question is around the internal APIs that the kernel uses, and whether, as Linux continues to take over the world, these APIs will multiply without bound, remain stable, etc.

        Side note: really interesting old post from Brendan Gregg on using Linux after Solaris here. https://news.ycombinator.com/item?id=13081465.

  4. IGnatius T Foobar
    Linux

    Solaris is the new Netware

    Solaris is the new NetWare. Snoracle will move all of their value added stuff to Linux so no one has to use their unwanted proprietary operating system anymore.

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Other stories you might like