Not logged in
Log in now
Create an account
Subscribe to LWN
LWN.net Weekly Edition for June 20, 2013
Pencil, Pencil, and Pencil
Dividing the Linux desktop
LWN.net Weekly Edition for June 13, 2013
A report from pgCon 2013
OpenIndiana lead Alasdair Lumsden resigns
Posted Sep 8, 2012 10:33 UTC (Sat) by mabshoff (guest, #86444)
Yep, it is.
> However, the must funny part was about slowlaris beating Linux. Keep dreaming till you wake up and realize slowlaris is dead.
Meh, I hate it when people use derogatory terms for OSes since it weakens their argument. While on average system call time for example seems to be faster on Linux than Solaris 10 or 11 on the same hardware, it is a long way from Solaris 7 and 8 days where it could be rather bad.
Last week AIX come up and some comments called it dead, too. IBM sells over 5 billion power hardware for AIX a year (compared to about 8 billion a year for the Linux hardware market overall) and in the second quarter 2012 it gained 6 points market share in the Unix server sector. Sure, it fell 10% in revenue compared to the second quarter 2011, but every other Unix vendor fell more. This has in part to do with upgrade cycle, i.e. Power 7 -> Power 7+, T4 -> T5 for Oracle and the latest Itanium for HP. The Unix market overall is shrinking but calling it dead is far from the truth. There are also still plenty of main frames around simple because there are workloads where they excel and people called the main frame dead in the 90s, too. And I am sure any hardware vendor would love to have a 5 billion a year hardware product range with juicy margins on top of which it could sell quite a bit of services, too.
Large SMP Unix boxen will just simply not go away. Savings for using Linux vs. Unix seem to matter rather little once you figure in for example the cost of an Oracle DB on that box. On top of that add risk avoidance and potentially a little FUD and who in their right might would want to migrated a number of large DBs for potentially little gain? It is different for green field deployments, but decisions are done very differently in a conservative enterprise setup. So AIX, Solaris as well as HP/UX will be with us for some time to come ;)
Posted Sep 8, 2012 12:27 UTC (Sat) by khim (subscriber, #9252)
The Unix market overall is shrinking but calling it dead is far from the truth.
The proper term is "zombie", I believe. This is lock-in schemes at work. Think War of Currents: AC distribution networks won back in XIX century yet some customers still bought elecricity from DC networks in XXI century!
UNIX may linger for similar amount of time, but it's relevance is the same: existing customers keep the thing alive, but there are few (if any) new users.
Posted Sep 8, 2012 13:36 UTC (Sat) by mabshoff (guest, #86444)
Yeah, inertia is a powerful force :)
> UNIX may linger for similar amount of time, but it's relevance is the same: existing customers keep the thing alive, but there are few (if any) new users.
I don't think it will last quite as long as DC current, i.e. even 20 year old technology is woefully out of date, but still lurks in some dark corners of data centers. I have never seen one, but I heard stories that some places keep some PDP hardware alive to run certain jobs. My own experience closest to these stories was some BS 2000 mainframe sitting in the corner of a data center to do certain jobs, no idea what exactly, but I counted myself lucky I had nothing to do with it.
As long as Unix vendors make money, they will sell and support it. There are really only three vendors left that matter (IBM, Oracle, HP) since Irix went into really deep maintenance mode and Tru64 is officially no longer supported. The BSDs do not really matter commercially in that market, but they are important for the custom storage and routing operating systems for companies like Netapp, Isilon (bought by EMC recently) and Juniper to name a few. Illumos is competition to the 'storage BSDs' and I would personally like to see it prosper since it is at least open source while the filesystems developed by Netapp and Isilon are not. Linux is certainly also used more and more in storage applications, so it will be interesting to see how commoditization will affect that market. James Bottomley has been giving talks on the economic forces of open source, i.e. look at LinuxCon Japan 2012 iirc for a recent example, and in those he argues that products that face commoditization tend to have the aspects that aren't value added parts open sourced. And the filesystem and routing function in operating systems seem to be at this point, i.e. five or ten years ago a storage box was big bucks, these days you can build one with open source without much trouble. So there has been price pressure on commercial vendors since you really start to have trouble to charge extra for a dedup feature in your array if you can just get an open source one. There is still plenty of gear to be sold in the high end, but I am certain places like Netapp fell the pressure.
To get back to Unix: I personally think that HPUX will be the first to go of the three left since Itanium's future looks more than a little bleak if you have read the discovery documents Oracle published . HP has attempted to port HPUX to x86-64 twice ( and ), but killed it again and again, so I think it will be the first to go as Itanium hardware support fades. Solaris has x86-64 and IBM seems to be able to sell enough Aix on Power despite the lack of workstation models which traditionally has indicated that a platform is in trouble.
And in the end what matters most is applications, which I suspect is the reason that the x86-64 port of HPUX was killed, i.e. why port to HPUX on x86-64 if you can run the software on Linux? If you look at the articles you will see that HP wanted to offer RHEL 6 compatibility so that Linux apps can run. I have no idea how they wanted to do it, i.e. visualization, kernel run time support via personalities or whatever. And this is really the issue with Illumos, to get back on topic, i.e. will software vendor support commercial software on it? I doubt that it will happen for a significant number of software packages and Oracle will certainly be hostile to them since the uname is still identical to Solaris. I am sure plenty of that unsupported software will run, but I do not think many companies will take a gamble to run their Oracle DB on a Illumos derivative :). Even free software project have started to have issues, i.e. there was a patch on the Mesa list to disable some features on OI/Illumos due to missing headers that worked perfectly fine on Solaris 11. It didn't go in. I have seen similar conflicts with patch sets to the user space code to OFED for example, too.
In the very end the weakness of the Illumos/OI community can be IMNSHO demonstrated via two issues: (1) no support for Sparc (2) the Illumos port of KVM does not have AMD support, i.e. it is Intel x86[-64] only, roughly two years after it was released. I wish them luck since I believe they will need it.
Posted Sep 8, 2012 18:26 UTC (Sat) by khim (subscriber, #9252)
I have never seen one, but I heard stories that some places keep some PDP hardware alive to run certain jobs.
It's even worse then that: PDP hardware is too expensive and rare today thus they run that stuff on emulators. I know someone who helped big customer to speedup their PDP-11 emulator for Itanium. Wonder what they'll do now when Itanium goes the way of dodo itself...
Posted Sep 9, 2012 7:34 UTC (Sun) by mabshoff (guest, #86444)
Hehe, I have been around the block a couple times, so those PDP stories are not something I heard about last week and I am sure are still told to this day as yarn to frighten the young sysadmins. Emulation makes sense given the space and cooling requirements and it probably runs much faster in emulation, too.
> Wonder what they'll do now when Itanium goes the way of dodo itself...
I am sure you will be able to buy Itanium hardware from Intel until the end of the decade with the last CPU we know about (Kittson) speculated to appear in 2016 according to . Throw in an additional die shrink like it happened for PA-RISC [but not for Alpha :(] and 2022 or 2024 does not seem unrealistic for a date until you will no longer be able to buy Itaniums. That will give time for the people running OpenVMS or NonStop and the strange main frame OS vendors that ported to it to get onto something else ;).
Posted Sep 9, 2012 2:46 UTC (Sun) by dlang (✭ supporter ✭, #313)
I'm sure that many IBM shops have so much invested in AIX software that they don't take much advantage of it, but it's an option.
Posted Sep 9, 2012 7:22 UTC (Sun) by mabshoff (guest, #86444)
Yep, but I suspect that the number is relatively small. I have never seen any estimates, but given that RHEL as well as SLES is available on Power it must be worth it. There certainly is Linux on Power, i.e. Watson, the system that beat the reigning Jeopardy champion ran on 10 racks of Linux/PPC . There is also a lot of Linux in HPC, i.e. Blue Gene with the third generation Blue Waters coming  (aka Blue Gene/Q in marketing terms). Even though none of the three generations is not really a 100% Linux system since only the IO nodes run Linux while the compute nodes have some minimalistic kernel .
IDC gives estimates for operating system market share in the server market (see  for the second quarter 2012) and while all those reports should be taken with a pinch of salt (or discarded like the insane Itanium projections from the late 90s :), these do seem to make sense.
I quoted my numbers above from memory and it seems AIX is probably slightly less than five billion and the Linux hardware market clocks in a little higher at around ten billion a year. Ironically systems like the Sparc based Fujitsu K (HPC 500 current second most powerful system) also count there since it does run Linux and not Solaris. All the other deployments of those Fujitsu HPC systems I have read about also run Linux even though it is also offered with Solaris. That should tell us something about the HPC market and the demand for Solaris in that sector ;).
> I'm sure that many IBM shops have so much invested in AIX software that they don't take much advantage of it, but it's an option.
Yep, with virtualization there is the possibility to run mixed workloads and given how much effort IBM spends on KVM/PPC it seems that there is demand. Linux on S390 certainly seemed to have helped IBM in the main frame business.
Copyright © 2013, Eklektix, Inc.
Comments and public postings are copyrighted by their creators.
Linux is a registered trademark of Linus Torvalds