Oracle just killed Solaris/SPARC/ZFS teams ...

[vermaden]

Oracle, just layed off about 2500 people from Solaris, Storage (ZS1/ZFS), SPARC and Library teams, earlier this year 450 people has been layed off.

At December 2016 information about Solaris 12.0 being canceled:

The Layoff: Solaris being canned, at least 50% of teams to be RIF'd in short term
https://www.thelayoff.com/t/KBEVoB1

The Layoff: Never Mind the Bollocks, Here's The Update
https://www.thelayoff.com/t/KTCW4qz

After which Oracle announced that there will not be Solaris 12, and that Solaris 11.next would be delivered using 'Continous Delivery' model.

Next, in January 2017 about 450 has been layed off from their Hardware Division:

The Register: Oracle lied: Database giant is axing hundreds of staff – at least 450 in its hardware div
https://www.theregister.co.uk/2017/01/23/oracle_layoffs_in_hardware_div/

A month ago John F Fowler'a has been layed off from Executive Vice President Systems, and also information came out that Oracle 'Continous Delivery' stratedy was just a mockup:

The Register: Oracle's systems boss bails amid deafening silence over Solaris fate
https://www.theregister.co.uk/2017/08/02/oracle_john_fowler_bails/

Oracle has revealed that John Fowler, a Sun veteran who stayed to serve as Oracle's executive veep for systems, has left the company.

An SEC filing [PDF] < http://d18rn0p25nwr6d.cloudfront.net/CIK-0001341439/1f9ea8c6-95d8-4b92-9a64-114cf6beeed6.pdf >
dated July 27, 2017 says “John F Fowler resigned his position as Executive Vice President, Systems effective as of August 2, 2017.”
No reason is offered for Fowler's departure.

And a former Oracle staffer tipped us off that [Big Red's Github repo] < https://github.com/oracle/solaris-userland/commits/master >
for Solaris's user-land is hardly a hive of activity suggesting a thriving continuous delivery effort.

“For a 'continuous delivery' model, I would expect several GitHub branches with a high frequency of commits,” they said.

“I very strongly suspect the Solaris 'continuous delivery' story is not true. They made it up and went public with it so they
don't scare off the last few customers who still buy hardware and Solaris licenses from Oracle.”

This came out on Friday:

The Register: Oracle finally decides to stop prolonging the inevitable, begins hardware layoffs
https://www.theregister.co.uk/2017/08/31/oracle_stops_prolonging_inevitable_layoffs/

Oracle is laying off staff in its hardware division, The Register has learned.

Current and soon-to-be former staffers have whispered online and to El Reg that the database giant is shipping out packages of
paperwork for ending their employment. The workers learned of this by receiving alerts from FedEx that the parcels, which
need to be signed for, are en route for a September 1 delivery.

"One of my co-workers emailed that he received a notification from FedEx of a label created by Oracle America, Inc,"

"I just checked and a label has been created for my home address. This is in the US. Looks like Friday is it for Sparc MicroElectronics."

... and Today:

The Register: Oracle staff report big layoffs across Solaris, SPARC teams
https://www.theregister.co.uk/2017/09/04/oracle_layoffs_solaris_sparc_teams/

Threads on The Layoff < https://www.thelayoff.com/oracle > suggests that around 2,500 layoffs have been made, covering
Solaris, SPARC silicon development and storage hardware, including tape libraries, with one result being that development
work has ceased on the ZFS Storage Appliance. The fate of Solaris and SPARC silicon remains unclear.

Oracle's silence on the matter is true to form, as the company seldom discusses layoffs. This round was communicated to
workers on the Friday before a long weekend and on the first day of the month. That timing means that immediate media
scrutiny was less likely and that regulatory filings under US States' Worker Adjustment and Retraining Notification (WARN)
regulations will remain out of sight for extra days in jurisdictions that report them weekly, or extra weeks in States that
publish monthly.

Meshed Insights Ltd: Oracle Finally Killed Sun
https://meshedinsights.com/2017/09/03/oracle-finally-killed-sun/

The news from the ex-Sun community jungle drums is that the January rumours were true and Oracle laid off the core talent
of the Solaris and SPARC teams on Friday (perhaps hoping to get the news lost in the Labor Day weekend). That surely has
to mean a skeleton-staffed maintenance-only future for the product range, especially with Solaris 12 cancelled. A classic
Oracle “silent EOL”, no matter what they claim as they satisfy their contractual commitments to Fujitsu and others.

The Layoff: FedEx labels have been created, delivery September 1st
https://www.thelayoff.com/t/P23GpT5

... and on Twitter:

Very sad day for UNIX and Solaris community.

Fortunately Solaris heritage lives in free and open Illumos.

Regards,
vermaden

 

[Remington]

I'm not surprised. Anything Oracle acquires usually die. I'm going to bet MySql will be next in silent EOL.

I hope that laid off ZFS developers will join Illumos project to continue improving ZFS.

 

[ShelLuser]

Sad story indeed. I owe my whole Unix expertise to Solaris, it was the first Unix environment I learned to work with.

To be honest this was to be expected the very moment these guys took over. One of the first things they did was triple the subscription support costs while seriously axing the services you got in return. I used to pay approx. E 200 / year per CPU but after the takeover I was looking at an estimate of E 900. For the exact same environment. Needless to say I skipped.

Anyway, I consider myself very lucky that I discovered FreeBSD when I did. Back then it still used the Solaris package manager, I immediately implemented ipf (which is still running on most of my servers, but I am busy with a switch to pf) and to me it was like a Solaris 2.0. Of course FreeBSD isn't the same as Solaris, but it had plenty in common. I always considered it a sort of successor, also never looked back.

 

[sko]

I hope that laid off [...] developers will join Illumos project

Bryan Cantrill of Joyent commented on hackernews[1]

For us, with the exception of the human toil on former colleagues (which shouldn't be minimized), this is a total positive: it eliminates the Solaris confusion (which we've seen recently in Go[1]), and it gives us a new talent pool to draw into illumos. While I don't know how many will matriculate, I can say with confidence that it will be significant in terms of magnitude if not in terms of numbers. So yes, I expect it to affect the direction of SmartOS -- and very much for the better!

[1]https://news.ycombinator.com/item?id=15171697

And IIRC it was on twitter where he also stated that there have been "a lot of talks during the last few weeks".

His recent blog post about the event is also worth reading: http://dtrace.org/blogs/bmc/2017/09/04/the-sudden-death-and-eternal-life-of-solaris/


I'm sure this event will bring a lot of momentum to the illumos project and/or projects based on its more widely used technologies like ZFS and DTrace as lots of key developers will be welcomed with open arms at the various companies that have improved illumos during the last 7 years far beyond what Solaris under Oracle was or became.

I haven't been exposed (at least knowingly) to Solaris/illumos until we started using SmartOS on our virtualization hosts ~2 years ago, but I almost immediately fell in love with it and spent many hours with zones on our smartOS hosts or OmniOS and OpenIndiana at home.
Every solution specific to the OS is thoroughly thought through and carefully integrated into the system - ZFS, DTrace, Zones, Crossbow, MDB (and I even like some aspects of SMF, albeit it being a radically different beast compared to RC). And what still impresses me: even the most powerful technologies are relatively easy to grasp and use as they are not utterly complicated to manage from an admin perspective - you can really tell they have been developed by and for those who actually use it.

 

[ralphbsz]

I hope that laid off ZFS developers will join Illumos project to continue improving ZFS.

Well, I have a higher priority: I hope they find good jobs. These are people with husbands, wives, children, mortgages, and cats that need to be fed, and we wouldn't want their cats to starve.

If in their spare time they want to help out illumos, that's a fine hobby to have. But first and foremost they need a day job.

 

[Cthulhux]

Leaving the (well-earned) Oracle hatred and the important remark about how SPARC is the notably bigger loss than Solaris aside for a moment:

Even under the assumption that illumos and its still living distributions (SmartOS, OpenIndiana, OmniOS - missed any?) will see a lot of new development activity now, Solaris will not be replaced just that easy. It made a good desktop system from what I heard, something that OpenIndiana still can't. For servers, ZFS (especially the Oracle version) is still plain awesome. For desktops, Solaris is probably over now.

(Please correct me.)

 

[ralphbsz]

From a commercial standpoint ... (and in the end, the commercial standpoint is the only thing that matters for paid software development, since the money to pay developers has to come from somewhere):

Solaris and Sparc have been completely dead as desktops for several years, and dying beforehand. Same applies to other non-x86 architectures. There is just about nobody who uses Sparc/PowerPC/Itanium, or Solaris/AIX/HPUX as a desktop machine. (Yes, I know this statement is not 100% true; I actually had lunch this week with a person who did have a PowerPC under his desk and ran AIX, but he retired from IBM two years ago). Yes, I know that you can buy desktop-packaged (non-rackmount) machines of these architectures, but the sales numbers for those is minuscule; they are bought as onesies for sites that deploy hundreds or thousands of them in data centers.

There are hobbyists (amateurs who don't get paid), collectors, and enthusiasts who still use these systems (CPUs and OSes) as desktops. But they don't matter to the commercial market, because the manufacturers don't make money of them. I highly respect this crowd, being a part of it myself: I have two small VAXes at home (yes, real VAXes, not Alphas or Itaniums), a friend of mine has a Xerox Alto and a PDP-15 at home, and is also the guy who manages the team that keeps a 70-year old IBM 1401 running. Yet, none of DEC/Xerox/IBM make money when a collector runs a 30 or 70 year old computer.

For several years now, the only place that non-x86 architectures get sold (in non-negligible quantities) is data centers, and their market share is cratering rapidly. With two significant exceptions, whenever a customer goes through a major upgrade from an existing system using proprietary CPUs (and the non-open OSes that come on them), they switch to x86. Obviously, that upgrade process is slow, and older architectures often continue to be used and sold for a decade or two after they become moribund. That two exceptions are: (a) The "agencies" within the US federal government (and other goverments) like one particular non-x86 CPU architecture a lot, and orders from those customers has been keeping that product line alive (at least for now). That tends to be pretty hush-hush and not talked about much. (b) The super-large commercial customers (led by Google and Facebook) are worried about the Intel/AMD monopoly, they don't want to become blackmailable by those manufacturers, and artificially keep one non-x86 CPU architecture alive (by spending a lot of extra money on those few CPU boxes).

For a long time now (several years), the writing has been on the wall that Sparc, Itanium and PowerPC will go away. We didn't exactly know which order that will happen in. Here would have been my predictions from about a year ago: Itanium didn't sell well, and is owned/used by two rational and well-managed companies, which will make sensible decisions about when to EOL it, and that decision was reached about half a year ago. Sparc is now owned by a company that is run by an insane egomaniac who can never admit defeat, and has a cult following in the microprocessor engineering community, so it might live for a long time, while Oracle sinks billions into keeping it alive. Turns out that even Larry Ellison (whose hot tub costs more than most CPU engineering teams) isn't that crazy, and just dropped it. And IBM and PowerPC is a special case; IBM has a knack for keeping older architectures not only alive but also profitable, and PowerPC has two sets of special customers that are interested in it not going away, and have the deep pockets to fund that. But I don't know how much longer that will hold.

An interesting question is: When will ARM make significant inroads in the server market? About two years ago, we thought it would start happening any moment now. It seems to not have happened yet. It might never happen. Instead, AMD once again has become very competitive in the x86 server market; but AMD has the skill of snatching (commercial) defeat from the jaws of (engineering) victory.

 

[ekingston]

An interesting question is: When will ARM make significant inroads in the server market? About two years ago, we thought it would start happening any moment now. It seems to not have happened yet. It might never happen.

I don't see this happening any time soon. The x86 still far out performs the ARM architecture in the server world and the new massive multi-cores coming from AMD are putting pressure on Intel to make significant improvements to the x86 line.

With the way virtualization technology is going, if you need a cluster of small servers, virtualizing them on one moderate (or big) server is becoming the preferred solution.

I can see in 5 or 10 years ARM starting to take over the desktop market. For most desktop workloads the ARM is sufficient and at the low end both cheaper and require less power than x86. I don't see that changing. We are already seeing some inroads in this area with ARM based Chromebooks (admittedly still niche) and Microsoft's recent interest in getting Windows 10 on ARM (although that appears to have stopped or gone into hiding).

We've also seen a few examples of phone docks that turn your cell phone into a desktop (or laptop). I suspect this is where ARM will take over the end-user market.

Instead, AMD once again has become very competitive in the x86 server market; but AMD has the skill of snatching (commercial) defeat from the jaws of (engineering) victory.

I really hope AMD keeps up their recent success, but they do seam to have a bit of a history.

P.S. I used to run a cluster of Spark servers with Solaris (and PowerPCs with AIX). I liked them. Shame they will soon be no more.

 

[Cthulhux]

I would assume that someone (Fujitsu?) will take over SPARC. After all, it is open hardware. But that will still hurt.

 

[ralphbsz]

I would assume that someone (Fujitsu?) will take over SPARC. After all, it is open hardware. But that will still hurt.

Why does it matter that Sparc is open?

If Oracle (=former Sun) and its licensee Fujitsu couldn't make money by selling Sparc, why would Fujitsu alone be able to make money? The development costs for CPUs, systems and OS would be similar for Fujitsu as for Oracle+Fujitsu, and the market size even smaller.

By the way, it's interesting to study the history of how Fujitsu (=HAL) and Sun cooperated (and occasionally competed) in the history of Sparc. And how the first 64-bit Sparc came about: not from Sun, not even from Fujitsu proper, but from HAL, a small company run by (or against?) a renegade cowboy who had already left IBM, and who is now a rock musician.

 

[Cthulhux]

The thing is: People could take SPARC and make their own ones. I just wonder why they don't. It makes a more profitable platform than ARM.

 

[ralphbsz]

The x86 still far out performs the ARM architecture in the server world ...

Although rumor has it that server-class ARM does better in computrons per Watt (where I use the humorous "computron" as some metric of computing something). And that metric might be more important to certain data center customers than raw performance.

and the new massive multi-cores coming from AMD are putting pressure on Intel to make significant improvements to the x86 line.

It's not just number of cores. The new AMDs have really good memory interfaces (AMD has a long tradition of doing well there), and they finally have a sufficient number of PCIe lanes for server hardware with a single socket populated. For many markets where the server chips are IO bound, that might cut the cost of CPUs in half, because with Intel you had to artificially populate both sockets in order to get enough IO bandwidth (which then hurts you in memory bandwidth, welcome to the wonderfully contradictory world of NUMA).

Just the fact that AMD exists again in the server space means that Intel will have to work harder: both drop prices, and come up with better products, if they want to maintain their traditional 90% market share. For customers, that's a good thing.

I can see in 5 or 10 years ARM starting to take over the desktop market.

I don't know whether in 5 or 10 years there will be a desktop market with real OSes. What do I mean by a real OS? The one where you can log in and perform low-level operations (like system calls) as a human, for example by writing programs in compiled languages, or by using shells or scripting languages. In particular my definition of "real OS" includes that it is self hosting. It's quite possible that in a few years, the only devices that humans directly interface with (look at the pixels, or press buttons on) will be tablet-style OSes (like iOS or Android). At that point, all software development will take place on servers, with the front end being just a user interface. My personal development style is already 90% there; I do nearly all my coding by ssh'ing into headless servers. Although it's nice to have a file system, editor, and compiler suite on my portable machine; that means that I can be productive even when I'm stuck in a place with no network connectivity.

 

[ralphbsz]

The thing is: People could take SPARC and make their own ones. I just wonder why they don't. It makes a more profitable platform than ARM.

Take the known open Sparc netlist or Verilog, and make their own chip, which is competitive with the latest Xeon, Epyc and Power9? You know how much it costs to develop a chip like that? That number is in the hundreds of millions, perhaps approaching a billion now. Small companies or amateurs don't have that kind of money (unless their name is Paul Allen, and he's more interested in building ancient computers).

And making a Sparc that competes with a chip from 10 years ago is not a way to make money.

 

[vermaden]

From a commercial standpoint ... (and in the end, the commercial standpoint is the only thing that matters for paid software development, since the money to pay developers has to come from somewhere):

Solaris and Sparc have been completely dead as desktops for several years, and dying beforehand. Same applies to other non-x86 architectures. There is just about nobody who uses Sparc/PowerPC/Itanium, or Solaris/AIX/HPUX as a desktop machine. (Yes, I know this statement is not 100% true; I actually had lunch this week with a person who did have a PowerPC under his desk and ran AIX, but he retired from IBM two years ago). Yes, I know that you can buy desktop-packaged (non-rackmount) machines of these architectures, but the sales numbers for those is minuscule; they are bought as onesies for sites that deploy hundreds or thousands of them in data centers.

There are hobbyists (amateurs who don't get paid), collectors, and enthusiasts who still use these systems (CPUs and OSes) as desktops. But they don't matter to the commercial market, because the manufacturers don't make money of them. I highly respect this crowd, being a part of it myself: I have two small VAXes at home (yes, real VAXes, not Alphas or Itaniums), a friend of mine has a Xerox Alto and a PDP-15 at home, and is also the guy who manages the team that keeps a 70-year old IBM 1401 running. Yet, none of DEC/Xerox/IBM make money when a collector runs a 30 or 70 year old computer.

For several years now, the only place that non-x86 architectures get sold (in non-negligible quantities) is data centers, and their market share is cratering rapidly. With two significant exceptions, whenever a customer goes through a major upgrade from an existing system using proprietary CPUs (and the non-open OSes that come on them), they switch to x86. Obviously, that upgrade process is slow, and older architectures often continue to be used and sold for a decade or two after they become moribund. That two exceptions are: (a) The "agencies" within the US federal government (and other goverments) like one particular non-x86 CPU architecture a lot, and orders from those customers has been keeping that product line alive (at least for now). That tends to be pretty hush-hush and not talked about much. (b) The super-large commercial customers (led by Google and Facebook) are worried about the Intel/AMD monopoly, they don't want to become blackmailable by those manufacturers, and artificially keep one non-x86 CPU architecture alive (by spending a lot of extra money on those few CPU boxes).

For a long time now (several years), the writing has been on the wall that Sparc, Itanium and PowerPC will go away. We didn't exactly know which order that will happen in. Here would have been my predictions from about a year ago: Itanium didn't sell well, and is owned/used by two rational and well-managed companies, which will make sensible decisions about when to EOL it, and that decision was reached about half a year ago. Sparc is now owned by a company that is run by an insane egomaniac who can never admit defeat, and has a cult following in the microprocessor engineering community, so it might live for a long time, while Oracle sinks billions into keeping it alive. Turns out that even Larry Ellison (whose hot tub costs more than most CPU engineering teams) isn't that crazy, and just dropped it. And IBM and PowerPC is a special case; IBM has a knack for keeping older architectures not only alive but also profitable, and PowerPC has two sets of special customers that are interested in it not going away, and have the deep pockets to fund that. But I don't know how much longer that will hold.

An interesting question is: When will ARM make significant inroads in the server market? About two years ago, we thought it would start happening any moment now. It seems to not have happened yet. It might never happen. Instead, AMD once again has become very competitive in the x86 server market; but AMD has the skill of snatching (commercial) defeat from the jaws of (engineering) victory.

That would be true only if Solaris would run only on SPARC or if it will be limited in any way on X86. Its not, Solaris has the same features on X86 as it has on SPARC. Of course LDOMs or recently Oracle VM for SPARC uses SPARC only features but that is Oracle VM for SPARC, a separate product. Solaris has the same features on X86 and on SPARC.

Solaris is the most modern UNIX miles ahead of AIX and HP-UX. Last HP-UX release was on 2007, there are patches that add new features thou. HP-UX is so old, that with system tools (ls/df/bdf/du/...) You can only display units as bytes, kilobytes or sectors, image how 'great' it is to work with HP-UX on current multi terabyte data ... We also know that Itanium is dead, so HP-UX is also dead, they will not port it to X86. AIX also has 'dead end' that IBM shows already, there are more options for POWER servers for Linux then for AIX, for example POWER 1U servers are Linux only ... and all POWER servers for Linux are a lot cheaper then similar servers with the same POWER CPU for AIX. In 2014 IBM paid GlobalFoundries (current AMD and ARM owner) $1.5 billion to take POWER CPUs production from IBM. If something is your most profitable line, You do not pay $1.5 billion to your competition to take over your core business. Latest AIX 7.2 release could be 7.1 patch as well as changes/new features are very small. IBM POWER servers are often used for SAP deployments because of PowerVM flexibility and performance, but latest SAP HANA offerings are supported only on SLES/RHEL on POWER, no go with AIX.

Solaris 'had' bright future because it was fully working on X86, and also is modern, has reliable package management with repositories (Ever tried to add/remove packages with dependencies on HP-UX or AIX? Lots of useless 'fun'.), not to mention its core technologies like Zones, RBAC, ZFS with BE, Comstar, Network Virtualization ... it is almost complete. The only 'missing' bit is hypervisor that would allow You to run Linux or Windows on it. When SUN was alive there was Solaris xVM Server, which had builtin XEN for that purpose, but as Oracle made tookover they killer that product for Oracle VM for X86 (based on Oracle Linux which is RHEL clone like CentOS). SmartOS (Illumos) done a good job here with porting KVM to Illumos. Also LX Branded Zones came back to Illumos recently (possibility to run Linux inside Zones instead of Illumos/Solaris).

I would understand 'killing' the SPARC division, as the world would be left only with AMD64 and ARM64 CPUs, it is happening already, but killing currently the best UNIX on the marked is very strange to me.

Oracle also 'killed' Solaris from the beginning that You can not use Solaris without support fee, for example if You buy RHEL with 1Y subscription, and that year is over, You can still use RHEL, but without support, with Solaris, You have to pay for support ALWAYS, even if You do not need that support ... and its not cheap (like everything in Oracle).

Any technical person that knows BSDs, Solaris/Illumos and Linux systems always prefers to use BSDs and Solaris/Illumos instead of Linux, and especially in the recent pulseaudio/systemd/journald years ... but that is generally how the World works, the shit wins ...

 

[Oko]

I would understand 'killing' the SPARC division, as the world would be left only with AMD64 and ARM64 CPUs, it is happening already, but killing currently the best UNIX on the marked is very strange to me.

Listen as somebody who grow up in Eastern Europe like you I understand your confusion but here in U.S. the corporations exist for one purpose only to make money. If SPARC, Solaris or anything was making money Sun would be still alive. Anyone remembers SGI and Irix? In early to mid 90s Irix was better OS than Solaris and XFS which came our of SGI is still the best file system available on Linux. If Oracle could make money by selling any of the Sun product they would not have killed it. Personally I always thought that Oracle bought dying Sun to be able to sue other companies for Java royalties. Sun played that card by suing Microsoft over built in browser Java VM and lost it. So did Oracle.

What concerns me more as a ZFS consumer is that the last thing from former Sun portfolio worth a lawsuit is ZFS. To make things scarier Oracle is not the only company which holds the ZFS patent claims. It is also the IBM which holds ARC cache algorithm related patents. In the age when suing other people is far more profitable business than creating the value by in inventing or building things I see no other cause of action for Oracle but to sue Joyent (SmartOS) and FreeBSD foundation for patent infringement. I just hope that Matt Dillan wins the race and finishes HAMMER2 before I have to ditch ZFS.

 

[vermaden]

Listen as somebody who grow up in Eastern Europe like you I understand your confusion but here in U.S. the corporations exist for one purpose only to make money. If SPARC, Solaris or anything was making money Sun would be still alive. Anyone remembers SGI and Irix? In early to mid 90s Irix was better OS than Solaris and XFS which came our of SGI is still the best file system available on Linux. If Oracle could make money by selling any of the Sun product they would not have killed it. Personally I always thought that Oracle bought dying Sun to be able to sue other companies for Java royalties. Sun played that card by suing Microsoft over built in browser Java VM and lost it. So did Oracle.

What concerns me more as a ZFS consumer is that the last thing from former Sun portfolio worth a lawsuit is ZFS. To make things scarier Oracle is not the only company which holds the ZFS patent claims. It is also the IBM which holds ARC cache algorithm related patents. In the age when suing other people is far more profitable business than creating the value by in inventing or building things I see no other cause of action for Oracle but to sue Joyent (SmartOS) and FreeBSD foundation for patent infringement. I just hope that Matt Dillan wins the race and finishes HAMMER2 before I have to ditch ZFS.

I also 'wait' for HAMMER2 as when it will be mature to compare it to ZFS, after it was ported to FreeBSD (which probably wont happen that fast :> ).

 

[ralphbsz]

Vermaden: You make many technical arguments that today's Solaris is the best Unix operating system. Your technical arguments may very well be right, and I won't argue with them. Partly because I have very little recent Solaris experience (recent = newer than 10 years). You also argue that HP-UX and AIX are the worst Unixes; and from having more recent experience on them, I probably agree.

But then you make an incorrect leap: Because Solaris is *technically* so much better than Linux, it should dominate the market place, and therefore Oracle=Sun should be able to continue to produce it. What you forget there is two things. First, on the revenue side: People use operating systems for a variety of reasons, most of which are not technical. Many customers use Linux because it is free (don't have to pay license fees), or at least much cheaper (license fees for commercial distributions, in particular RHEL, are much lower than for the big three proprietary Unixes Solaris/HPUX/AIX). In particular, the presence of things like CentOS (which is de-facto RHEL but without license fees) reduces the cost (in particular because you can do mix and match: have a few RHEL machines to allow you to get paid support, and then have a lot of CentOS machines without the license fees). Many customers also use Linux for reasons of trying to be hip and with the trend: there is a reason it's called a "LAMP" stack, and many corporate users feel that being "modern" requires them to use Linux and FOSS. Example: I once had the CIO of a major european financial company tell me that he would be happy to use our very complex and expensive software product (the expensive part are the high annual support fees), but only if we first open-source it: he wanted the quality and features of our product, he was happy to pay a lot of money for our excellent technical support, but for purely religious reasons he insisted on being able to see the source code (which is so complex, he would never understand, much less be able to modify). So on the revenue side of selling operating systems: it's very hard to compete with the Linux juggernaut, therefore revenue will be low.

On the expense side, programming, improving and supporting an OS like Solaris is very expensive. It takes a lot of people. And in the case of a company like Oracle=Sun, those are very well-paid and competent people with oodles of experience. In comparison, Linux is maintained by a large mix of people, employed by a lot of different companies: RedHat is probably the largest employer of Linux staff, but various computer vendors (HP, IBM, ...) and hardware vendors also contribute. Just as an example: The device driver for the LSI SAS HBAs for Linux is written by LSI people, and then released for free; the device driver for the same HBA inside Solaris/AIX/HPUX is written by Sun/IBM/HP staff to LSI's specification. Some Linux developers are not paid at all (they are amateur hobbyists), although the contribution from hobbyists is getting pretty small these days. And finally, the average pay of Linux people (development, support and SG&A) is lower than for proprietary software, which is also correlated with the average age (=level of experience) being lower in the Linux (and in general FOSS) market.

So, there is little money in selling Solaris (or other non-free OSes), and it is expensive to produce them.

Which gets us to the core question: Why did Oracle drop Sparc and Solaris? Probably because they can't make a big enough profit by selling them. Matter-of-fact, I suspect they were losing money on it for a while, and only Larry's giant ego prevented the decision from being made earlier. Similar arguments apply to Itanium and HP-UX (and other proprietary HP offerings, like VMS, NonStop, MPE): HP decided that there is more money to be made in other markets, so it end-of-lifed them. Why is PowerPC and AIX still alive? Probably because IBM is still making a profit on them, which may be correlated with IBM having better skills in managing the production, sale and support of older systems. Perhaps the fact that with OpenPower they are moving part of the cost structure to other companies helps with that too. How much longer that will go on is a fascinating question, about which we can only speculate.

The technically better product doesn't always win. In the computer industry, you actually get the feeling that the technically worse product often ends up succeeding in the marketplace, because customers are more driven by emotion than by reason.

 

[gofer_touch]

IBM and PowerPC is a special case; IBM has a knack for keeping older architectures not only alive but also profitable, and PowerPC has two sets of special customers that are interested in it not going away, and have the deep pockets to fund that. But I don't know how much longer that will hold.

Hopefully things like the Talos II might be persuasive enough for some to consider PowerPC systems for high performance use cases.

 

[ralphbsz]

... consider PowerPC systems for high performance use cases.

Pardon me, "consider"? PowerPC systems are already used quite heavily in HPC (high performance computing). If you look at the top500 list, you find that while the majority of systems (both by the number of systems and by the fastest systems) use Intel, PowerPC systems are well represented, on the list, and the largest non-Intel CPU on there. Interestingly, there are still quite a few Fujitsu-built Sparc systems on the top500 list (maybe 1/3 as many as PowerPC).

(The history of CPU architecture market share in HPC is interesting; the non-Intel architectures have been declining, and indeed the new chips such as what is used in Talos II might be able to slow or halt that decline.)

But then, the top500 describe the highest end of the supercomputer market. Many of the systems on that list are not built to be cost-effective for general purpose HPC workloads, and many systems on that list are not efficiently utilized. Instead, there are several non-technical factors in play here: some of them are "trophy systems", where a site (such as Livermore) or a country decides to have the fastest computer in the world at any cost, even if they don't have a real use for it. A subset of that effect is that governments heavily fund development and purchase of supercomputers; as an example look at how DARPA in the US plows hundreds of millions into vendors such as Cray, HP and IBM. Some of the systems are loss leaders: sold at a deep discount by computer vendors, so they get the privilege of being able to advertise with their presence on the list of largest supercomputers. Quite a few supercomputer systems are carefully engineered to look really good in the speed test that's used for the placement on the top500 list (some variant of Linpack), and may not actually be very useful for real-world HPC applications with real-world communication and IO requirements, and with a variety of programs.

And a very large fraction of scientific computing doesn't show up on the top500 list. There are lots of medium-size compute clusters used for "scientific" or high-performance computing, for applications such as biochemical and genetics (protein folding is an example), seismic analysis and oil&gas, and intelligence gathering and analysis; I would bet that the aggregate sales volume of those medium-sized systems is much larger than the huge ones on the top500 list, and in particular makes the vendors higher profits.

Still: High-performance computing (in particular government use of compute clusters) is probably the single biggest reason PowerPC is still alive, and Sparc survived for as long as it did.

 

[Remington]

What are the chances of Oracle reverting Solaris back as open-source project as they apparently are not working on it anymore?

 

[phoenix]

What are the chances of Oracle reverting Solaris back as open-source project as they apparently are not working on it anymore?

Slim to none, but not completely impossible.

 

[vermaden]

Many customers use Linux because it is free (don't have to pay license fees), or at least much cheaper (license fees for commercial distributions, in particular RHEL, are much lower than for the big three proprietary Unixes Solaris/HPUX/AIX).

I also thought the same, that Solaris, as its real UNIX, will cost more then 'just' a Linux, that RHEL for sure must be cheaper, but the reality is the opposite. Solaris is cheaper, then RHEL with the same support plan. For example, to get Solaris on 4 socket machine, You have to pay $1000 [1] a year, with RHEL You have to buy TWO licenses because every RHEL Server license 'covers' only two sockets, so with RHEL, the same support as Solaris (24/7) costs $2600 [2]. So Solaris is 1600 cheaper per year then RHEL with 24/7 support. If You get RHEL with 'weaker' 5/8 support it will be $1600, still $600 more then Soalris which is still $1000. For 2 socket machine, the difference would be S:$1000 and R:$1300 or R:$800 with lower support plan.

The difference gets a lot bigger (Solaris being MUCH cheaper) when You take virtualization into the play, for example You have single VMware ESXi host with 4 sockets on which You want to run as much RHEL or Solaris as possible, lets say 100 for example. For Solaris You would still buy $1000 license, which allows You to have unlimited instances of Solaris on that box. For RHEL it gets complicated, because You either have to buy two RHEL for Virtual Datacenters licenses, which with the same support plan as Solaris would be $4000 x 2 = $8000, so Solaris is 8 times chaper then RHEL on that box, for 2 socket it would be S:$1000 vs R:$4000, still 4 times less for Solaris.

So the bigger the hardware (sockets) the cheaper Solaris support becomes.

Solaris can also can be get for free, just buy X86 or SPARC server from Oracle, its generally the same as Supermicro/IBM/HPE/Lenovo/... and You get lifetime both Solaris and Linux license for free.

In particular, the presence of things like CentOS (which is de-facto RHEL but without license fees) reduces the cost (in particular because you can do mix and match: have a few RHEL machines to allow you to get paid support, and then have a lot of CentOS machines without the license fees).

That is little on a RHEL side, while Solaris 'has' Illumos/OmniOS for servers and Illumos/OpenIndiana for desktop/laptop there is no 100% compatibility between them.

Many customers also use Linux for reasons of trying to be hip and with the trend (...)

Going that route, sorry for language, not a personal attack but just 'saying' - "Lets eat shit, millions of flies can not be wrong."

People should be using 'this' or 'that' because 'this' or 'that' is superior (or just better), not because its 'trendy', but we know they don't ...



I agree with most of what You said (no reason to quote it all), I would also prefer that Oracle would have the same business model as Red Hat with everything (or almost) being open source with possibilities for free clones.

This is also interesting read:
Cost comparison: Solaris/SPARC vs Linux/x86
http://www.zdnet.com/article/cost-comparison-solarissparc-vs-linuxx86/


[1] https://shop.oracle.com/apex/f?p=DSTORE:2:::NO:RIR,RP,2:pROD_HIER_ID:6916016290451192110906
[2] https://partnercenter.force.com/s/MSRP-pricing

 

[ralphbsz]

Interesting. I'm quite surprised that Oracle Solaris support is cheaper than RedHat RHEL support. You tought me something unexpected, which is always pleasant.

P.S. In my opinion the probability of Oracle open-sourcing Solaris is zero. First, it would require a lot of work to scrub the source code of stuff that is under NDA, patent, or copyright (I think Solaris still has licenses for the original AT&T -> SysV code). Second, there is always Larry's ego, which is probably saying: If I can't have it, nobody else can either. He's quite an unpleasant person that way, to put it mildly.

 

[ShelLuser]

What are the chances of Oracle reverting Solaris back as open-source project as they apparently are not working on it anymore?

Why should they? There was Open Solaris but they shot that down, but there's still OpenIndiana (or Illumos), which is basically an open source variant.

Solaris is cheaper, then RHEL with the same support plan. For example, to get Solaris on 4 socket machine, You have to pay $1000 [1] a year, with RHEL You have to buy TWO licenses because every RHEL Server license 'covers' only two sockets, so with RHEL, the same support as Solaris (24/7) costs $2600 [2].

Sorry but that analogy is flawed. First of all the price starts at $1000, it's not $1000 perse. If you dig further into that page (and the sub pages) you'll come across this:

1) The number of physical sockets in a server determines which product a customer buys

So: 1 - 4 sockets means you're buying Premier subscription. But then there's this:

2) The number of occupied sockets determines how many Solaris subscriptions must be purchased

You said you're on a 4 socket machine and all 4 sockets are being used? Congratulations, you're looking at a $4000,- expense.

See this page.

 

[vermaden]

Why should they? There was Open Solaris but they shot that down, but there's still OpenIndiana (or Illumos), which is basically an open source variant.

Oracle passed OpenOffice to Apache foundation, but they didn't do the same with OpenSolaris, I doubt that they will open source Solaris, but with Solaris this is not a problem as Illumos exists and a lot of people is working on it since 2010.

Sorry but that analogy is flawed. First of all the price starts at $1000, it's not $1000 perse. If you dig further into that page (and the sub pages) you'll come across this:


So: 1 - 4 sockets means you're buying Premier subscription. But then there's this:


You said you're on a 4 socket machine and all 4 sockets are being used? Congratulations, you're looking at a $4000,- expense.

See this page.

Sorry, but You got it wrong (which happens often wit various licensing models).

If You have server with 1 to 4 sockets, then ONE Solaris support license will do for $1000. If You have server with 5-8 sockets, then You have to but $2000 license. That is. I have talked about this with Oracle representatives myself, as we considered using Solaris for database environment.