It’s Either That Or Move To Linux
I suppose if you’ve found your own little pool of the “More Money Than Sense” brigade, you might as well soak them for as much as you can get, for as long as you can.
Oracle has quietly extended paid support and upgrades for Solaris 11.4 to 2037 – three years past its previous deadline – and did the same for earlier versions of the OS last year. Solaris is a proprietary Unix that Oracle acquired along with Sun Microsystems in 2010. Teamed with Sun's own SPARC silicon, the OS was an …
When you're running truly big workloads, there are situations where Solaris is still far more capable than Linux.
And, of course, even if you could use Linux, when you've spent a few millions on hardware and application licenses why would you rip them all up and start again just to get effectively the same thing if you don't have to?
Linux scales to millions of CPU cores in the world’s most powerful supercomputers. No other OS can compete in that league.
> when you've spent a few millions on hardware and application licenses why would you
>rip them all up and start again just to get effectively the same thing if you don't have to?
Because Oracle is continuing to soak you for ongoing support and licensing fees.
"Linux scales to millions of CPU cores in the world’s most powerful supercomputers. No other OS can compete in that league."
But that's not a single instance of Linux. In large clusters of servers used as supercomputers, scalability is due to the application more than the operating system.
Linux scales to millions of CPU cores in the world’s most powerful supercomputers. No other OS can compete in that league.
Ah, someone who thinks "big" means just a few PB of data and lots of parallel processing.
There's a world of a difference between an HPC application running on a million-core supercomputer, and a suite of business apps running on a very large mainframe-class business system. There are many sorts of scalability, Linux is excellent at some of them, but Solaris is still better for some of the others.
Because Oracle is continuing to soak you for ongoing support and licensing fees.
The cost of hardware maintenance and application licenses, which are much the same no matter what OS is running, dwarfs any difference between OS costs. The cost of throwing away working stuff just to replace it with other working stuff would be uneconomic.
Linux does scale to at least 768 cores and 48TB in the HPE Superdome Flex servers though. And IBM is moving its POWER servers from AIX to Linux. So single system image scaling of Linux is now quite competitive with the Sys V Unices. SGI IRIX had the largest SSI scaling with its Origin MIPS systems, to over a thousand cores, but its NUMA latencies made programming it problematic.
There'sa big difference between "running" and "scaling". Running means not crapping your pants with a lot of cores. Scaling means leveraging these cores to do useful work. There is no universe - this one, or a parallel one, where Windows scales to 2000+ cores...
I know you meant well, trying to use that phrase “mainframe-class”, but remember that “mainframe-class” represents exactly the sort of batch mentality Sun and Unix were trying to get away from. All those brilliant, yet retired, Sun engineers, would be horrified to learn from you that they had failed.
You obviously have never seen a real mainframe setup. 200 online transactions per second is quite normal when I last worked on one twenty years ago peaking at 350. I am pretty sure a modern Z series can at least double that. Plus at least 4 machines spread over two sites for 99.999 % availability ( the o.oo1% was unavoidable application upgrades ).
There are just some problems you cannot solve by running lots of processes in parallel so you really need a powerful single instance. As of 2022 commodity intel chips can finally compete with OS/360 hardware in this space.
Having said that the Solaris HA setups were equally impressive most of the ones I worked on outlived the companies they were built for.
Open Solaris was demonstrated on System z in the mid-late 2000s. IIRC, Oracle pulled support for development when it acquired Sun.
But returning to the original point, you don't seem to understand the nuance and complexity of different kinds of scalability, and while Linux is also a great OS, it doesn't need misguided fanboys to promote it.
Gee, I wonder why. Something about not scaling beyond vapourware, to production environments, perhaps?
No doubt a commercial decision. But perhaps you'd care to explain why Linux would scale better on zSystems.
Remember, Linux is being used in “mainframe-class” production environments, Solaris is not.
And also remember that the "mainframe-class" goodness — resource management, fault tolerance, etc. — comes from IBM's deep stack of proprietary software and hardware providing a virtual machine under the hood. Linux doesn't manage the hardware directly.
To Linux, the mainframe can appear as a tightly-coupled cluster of machines, so many instances of Linux can run efficiently without having to scale across a big system in one big system image.
Because it actually runs on zSystems, Solaris does not. Remember we were talking about “mainframe-class” deployments, were we not?
> Linux doesn't manage the hardware directly.
It seems to manage that better than Solaris ever could. That’s what counts.
“Anthropic principle” ... now there’s a “desperate appeal” to some pretty abstruse philosophical rationalization as to why one product failed in the marketplace and the other did not.
Remember, I was not the one who introduced the phrase “mainframe-class” into this discussion. I just ran with it.
The big thing people were trying to get away from was a systems design paradigm established decades previously, where hardware was slow and expensive and thus batch was the only way to tackle big jobs. With the risk of fast, large (by the standards of the day) and cheap disks, faster and more capable processors and software. A new system design paradigm was possible - replacing those tape based databases and batch systems with an online disk-based relational database and transactional systems. The savings to be had were significant, for one client the overnight batch run changed from having to process the entire 5m customer (tape based) database every night to only having to access and process less than 200k customers assigned to that nights billing cycle. Obviously, once that customer data was accessible online, businesses were able to both mine it and to deliver new services to customers.
Wow - you worked on Solaris that recently? What a waste of time.
Solaris as an OS any one would want to use peaked in 1995 after which NT4 blew it out of the water as a Workstation. Even back in the early 2000s Solaris release notes were all "new in Solaris xxx introduced (copied from) in Linux in yyyy". By 2002 Linux had been bashed into a better shape than Solaris for most workloads. By 2006 there was basically nothing that Solaris was better than Linux for.
Now I have a vague understanding that it's still available for licensing, in the same way as AIX still technically exists. Certainly we ended sales for the Solaris platform in 2010 (and that was after 5 years of no feature development).
“You've got the occasional box running some crap that doesn't matter in the a corner of a room.”
You haven’t got a clue. If that was true, Oracle wouldn’t be extending support. It’s huge in Finance and Telecoms industries. Finance in particular where the pace of change is glacial.
I agree, but in my experience most Solaris holdouts are applications that were originally built in the mid to late 90's when you really did need the specialised hardware and software to get the performance for your particular problem.
Commodity hardware has caught up for anything in that space well and truly. Only the most niche and volume applications really need the levels you're suggesting. I have to say they are damn tasty machines, they fit that gap where an IBM Z is too much but commodity not enough.
Oh look, it's the mainframe idiots. They always pop up when they're least wanted.
No, your mainframe is not faster than a rack of servers.
No, "transactions" is not a measure of computer speed, nor is it in any way relevant. There's a reason why IBM never quotes actual benchmarks.
No, your mainframe isn't faster at IO than a rack of servers either.
Just stop. The fastest computers in the world are supercomputers running Linux.
2037, I see. Not going beyond that date, possibly because : https://en.wikipedia.org/wiki/Year_2038_problem
Don't see many being able to jump on the bandwagon and cash out with this one. All I would hope is to be on the correct side of the grass at that point.
I don't disagree.
But the Lindy Effect is currently suggesting that Solaris 10 will likely be the last remaining OS on earth, outliving all of us.
I mean, seriously, if I knew the lifespan that Solaris 10 was going to have in 2005, I would have absolutely prioritised it over even Linux. 20 years is almost an entire career length.
That's where Sun really shot itself in the foot. That "upgrade" requires a complete reinstall, and is one of the main reasons that so many people are still on S10 (and why not, it's still solid). After the SunOS→Solaris 2 move Scott McNealy promised that they'd never do that to customers again, but...
A cynical person might wonder if McNealy knew he wasn't going to be around Sun forever, and wasn't terribly worried about 'promises'".
"What? Oh, right -- sure, tell them we'll never require upgrade-reinstalls ever again. Schwartz can deal with that later...."
Turns out it's Larry rather than Jonathan Schwartz, but whatever. Larry doesn't care either.
The lengthening of system lifespans as technology got better and matured is inevitable, just that some, particularly in sales and marketing are obsessed with “new” .
The challenge is finding a way for a company to survive such timescales to provide on-going support and maintenance.
I'm guessing that money isn't a problem and you make a lot of bad decisions every day anyway.
Have you ever been in an environment where Solaris and Sun hardware has been in use? Yes, it may very well be legacy now but in my experience, Solaris and Sun hardware, especially when running as a cluster, is pretty bombproof with uptimes running into the years (decades?) rather than days and weeks for Windows.
It's not so much that "money isn't a problem" it's having the confidence and comfort of a having a rock solid platform to run your enterprise or mission critical applications, some of which may literally mean life or death to your customers/clients.
I admit it's been a long time since I've had to look after a Solaris box to that extent, but I reckon there's still a fair few of them running away quite happily, doing what was promised of them in the 2010s, 2000s and even 1990s, hence why support has been extended.
"having a rock solid platform to run your enterprise or mission critical applications, some of which may literally mean life or death to your customers/clients."
Which, to my mind, means something you can easily upgrade, replace, seek alternate support and don't have just one avenue / company for that who are notoriously for killing acquired products or stinging you into bankruptcy just to keep doing what you've been doing for decades.
Sorry, but "uptimes running into the years (decades?)" is a BAD THING. It means you're entirely reliant on that system just keeping working and have never actually thought about what happens when it goes wrong, is no longer supported, needs to be moved to another city, etc. etc.
Yes, you're right - many organisations are entirely reliant on their mission critical systems. But I'm pretty sure if that's the case, then I would hope the majority have a DR solution in place that is regularly tested, the plan is kept up-to-date etc. and the hardware may even be geographically spread out for load balancing, DR purposes, etc.
And with all due respect, I'm unsure if one of my theoretical mission critical systems could be upgraded "easily" or replaced - rightly or wrongly - as the organisation would have made a major investment in it over the years. And the kind of organisation that is happy to spend millions on building, running and maintaining a mission critical system will happily pay for proper vendor (in this case Oracle) support rather than some kind of third-party service provider whose SLAs include the term "best endeavours".
One of the organisations that I worked for in the past had a vendor's parts bank on-site with common spare parts, despite being promised a one hour delivery of any part, as their system really couldn't be down. And, yes, this system was fully clustered, fully redundant, full DR capability (with DR site being updated in real time), so shouldn't have needed any of that, but such was the value of the support contract and the clout of the organisation, the vendor happily agreed. (This was a shared parts bank. The vendor also had an engineer based on-site 24/7/365.)
So, yes, maybe relying on a single vendor is bad from a financial point of view, but in some cases, it's not only the right choice, but the most prudent choice.
Yes, but Sys V Unix, on which Solaris 2+ is based, was released in 1983. The truth is that all the Unices and Linux are antiques, but in the way horseshoe crabs, dragonflies, and cockroaches are biological antiques. Primitive but they get the job done and are flexible enough to adapt to change.
>(snip) just for an antique OS, (snip)
Looks at Wikipedia's "Comparison of Operating Systems" page and struggles to find an OS that is 1) new, 2) cheaply supported, 3) mainstream. Finds.... nothing!
It's quite an interesting list to look at, sorted by "Initial Public Release". The newest, truly fresh and potentially very significant OS going forward is Redox - a from-scratch OS development written in Rust. In case you've not seen it, it's work taking a look - it might be "The Future".
What has impressed me the most about that project is just how quickly it went from nothing to a working graphical desktop, applications. It really hasn't taken them very long at all; I reckon they've written an awful lot of code really quite rapidly. That probably means Rust is pretty efficient in terms of developer time, which in turn probably indicates a future direction for all on-the-metal software projects; C/C++ are slower and riskier to develop for.
Considering that several existing OSes (Windows, MacOS, even Linux) look to be in the process of getting Rustier, in time it's possible that several antique but very mature and widespread OSes could eventually wind up being majorly reimplemented for the modern age.
Hmmm while working for a 300-person division of Time Inc at the beginning of the current millennia, I needed to reboot one of the $800K Sun 6800s due to a clock bug that showed up after (oddly) 520 days (not 512) on a beast that had uptime of almost 500 days...Per usual, at the weekly IT & biz team lead meetings, all the Windows team had to say to justify an outage was 2 words "Patch Tuesday". I on the other hand as the head of Solaris support had to justify to the Nth degree why I needed an outage (and produce a 50+ step MS Project plan detailing exactly how said reboot / maintenance window would be handled.)
But hey, Solaris put my 2 daughters through college.
Sounds like an lbolt bug, if I've done my sums correctly that would be after 497 days (assuming HZ is 100 which was pretty much universal on anything except Windows 9x back then). It shouldn't have affected Solaris at all but third party drivers may have had an issue.
It one of those things well known to the grey beards of the time or those that spent proper time with them to soak up their knowledge. Most of the grey beards I knew are dead now.
During to 90s and 00s a small group of second user SparcStation 5s ran management applications for a travelling set of private networks. They travelled all over the world for some 13 years and never batted an eyelid. They would have run longer still if they hadn’t been forcibly retired. I dread to think of the care and feeding that a bunch of latter day Windows boxes would need to do that.
And in 2018 Oracle just about confirmed that status by halting development of a major upgrade and freezing the OS at version 11.4, plus continuous patches and package updates.
Don't you wish more OS would stop monkeying with adding new/shiny/shitty stuff and just keep fixing the known/working/stable system?
I noticed no one mentioned IllumOS, an open source derivitive of Sun's Opensolaris (from Solaris-10.)
If you were running in house Sol-10 applications on amd64/x86_64 hardware then a IllumOS based distribution like OpenIndiana is likely to be binary compatible. Whether Oracle software would run under IllumOS is a $64 question.
Porting software from Sol-10/sparc to amd64 is more likely to be more successful than to another OS eg Linux, *bsd.
I know of an attempt to port scientific software to Linux from Solaris which after six months and 10s of $K was abandoned.
In any case at the price, worth looking at. :)
It's "illumos". No capitals. Some of the team get quite picky.
But yes, illumos continues the "just works without randomly breaking stuff on a regular basis" philosophy. Like Solaris, sits quietly in the background doing its job without requiring an army of squirrels constantly caring for it. Binary compatibility with Solaris 10 is excellent.
Used for new and very cutting edge stuff, like Oxide and Helios.
(Some of us are still running illumos on sparc. Now that is a bit niche, I'll admit.)
Marketing and subtly planned obsolescence convinced many people that everything has to constantly be upgraded and replaced.
IT people specially love to dance to the music from Big IT, since participating in this scam party pays part of the rent.
A company who bought in 1998 a Sun server with an app that does important tasks for them, can still use it today like they did in 1995.
The amount of money they saved by using the same configuration, not spending fortunes on upgrades and 4-5 years replacement cycles, is enormous.