Submitted by This paper tries to compare Linux vs. Solaris. Its author comes to many conclusions, among which this is one of the more interesting: “All-in-all Solaris is a powerful, stable, conformant-to-standards OS that can run many open source applications as well as Linux, and some (mainly multithreaded applications) better than Linux. Like in the cases of Red Hat and Suse, the cost of support is extra, but it is more reasonably priced. Security patches are free which makes Solaris similar to Windows.”
So, as to not lead people in the wrong direction, could you please correct the title and put the real one instead! It’s a framework base for further comparison. Didn’t you read the article before posting it?
Nice article. The author, IMHO, makes some well deserved comments about the fact that Unix code was very available at the time Torvalds did his “hobby/vanity fair” project. It also points out that Sun made a substantial amount of contribution to the *nix based code that is being used today. RFC, NFS, and most recently DTrace all come from Sun.
It’s too bad for Sun that the most obvious of truths always comes to the forefront, which is that being good at computer technology does not necessarily make you good at “getting rich.”
So I’ve skipped ahead to the second page, and he’s going on an on about how rubbish Linux is because it’s not a micro-kernel, and therefore conservative, and therefore behind the times.
And then he goes on to talk about all Sun’s inventiveness, while failing to omit that Solaris is also monolithic.
And then he goes on about how much faster BSD development is, despite the fact that it started before Linux (albeit with a mid-stream re-write just as Linux was started in 1994), and omits the fact that in terms of performance and hardware support BSD is inferior to Linux.
Lastly there is the fact that Linux has seen innovation: the best example of which is HAL, but which also includes things like Reiser (version 3 of which is in the kernel). While HAL is not core kernel innovation, neither are things like PAM or NIS. They involve the kernel, but they are user-space tools in the end.
I have to say, after just one page, I’m not inclined to read the rest. The article seems a bit biased, and jumping on the microkernel bandwagon seems like flamebait. I could be wrong, but life’s too short to bother finding out.
Edited 2006-06-14 17:19
Well you do not have to read the rest.
I think it is an fair comparision and i know that every Linux geek thinks that such an article is biased because Linux is not the overall winner.
It is just such an shame that Linux users cannot be that openminded that they realize that linux is not always the best.
I actually think Solaris is a better kernel, I just think the author is talking nonsense. I am not a “Linux geek” and I’m frankly offended by the suggestion that I’d be so childish and biased.. Read my second comment below: as it turned out I couldn’t stop myself, and wasted some more of my time reading this trash.
I actually think Solaris is a better kernel, I just think the author is talking nonsense. I am not a “Linux geek” and I’m frankly offended by the suggestion that I’d be so childish and biased.. Read my second comment below: as it turned out I couldn’t stop myself, and wasted some more of my time reading this trash.
I haven’t even looked at the article yet, but given by the amount quoted here, the original author of the article doesn’t have the slightest clue about holding an adult level discussion about Linux vs. Solaris, even if suc war actually existed.
Red Hat vs. Sun – sure, Novell vs. Sun, sure, but Linux vs. Solaris? Please.
He specifically made the point that the innovative ideas are coming out in userland. You say that Linux has been innovating, and then point out userland tools, one of which is actually based on a creation by Sun, and in fact still compatible with Sun’s version, I’ll let you figure that out.
I’d also like to say that while the article may be rubbish, you yourself speaking rubbish about BSD is unnecessary. Where has it been proven BSD performs poorly in comparison to Linux, and in what aspect does BSD perform worse in? I highly doubt that you can make the broad generalization that BSD performance is inferior to Linux. If you wish to speak of inferior performance I suggest you begin by looking at your own comment writting abilities and your hard headed egotisitcal patronizing comment. I sir, am tired of seeing your rubbish plastered all over the web.
1) NetBSD supports more platforms than Linux – so stop using the generic term ‘BSD’ as it encompasses FreeBSD/NetBSD and OpenBSD.
2) There is nothing innovative about HAL; its an abstraction layer with policies – not to demean from the effort, but its hardly what I would call earth shattering when compared to NFS when it first arrived, or X11, or dtrace.
HAL was also designed as a multi-platform solution to get around the differences between UNIX so that there is a single underlying layer on which GNOME (and others) can be based, thus, simply requiring HAL to be made available, and everything should be able to compile ontop of that.
3) There is nothing innovative about ReiserFS; having a look at the structure, and the ideas are nothing original – just like ext3fs is nothin original.
4) I’ll give you that Solaris is inferior on the desktop, but so is Linux and BSD when compared to commercial offerings – commercial companies like Apple and Microsoft have the luxary of having the money and the man power to spend, Sun, Red Hat, and the FreeBSD don’t have that luxary, so they’ll always be behind the eighth ball.
UNIX, by enlarge, is a server operating system from the ground up – very few (Apple) have tammed it enough for the desktop – Solaris as a server OS, quite frankly, is top notch when compared to the commercial offerings offered by Red Hat; and once OpenSolaris is up and running, you’ll see updates being offered by the community, not only for security issues, but general bug fixes as well.
2) There is nothing innovative about HAL; its an abstraction layer with policies – not to demean from the effort, but its hardly what I would call earth shattering when compared to NFS when it first arrived, or X11, or dtrace.
Quite right. HAL or Hardware Abstraction Layer, is actually no innovation at all nor is it new. Windows NT has has a HAL since day one, think 1988 (18 years).
Quite right. HAL or Hardware Abstraction Layer, is actually no innovation at all nor is it new. Windows NT has has a HAL since day one, think 1988 (18 years).
Well, the HAL won’t be as comprehensive as the Windows NT one, but it should allow a common interface to hardware, api’s etc. to be exposed to desktop developers and third party application writers; and thus, for example, gnome-cd for example, hardwires itself right to, IIRC linux/cdrom.h – which is completely rediculous when it comes to developing a desktop for multiple UNIX’s.
Hopefully in the next few years, that terrible eyesore to portability will be replaced with hal/hardware/storage/cdrom.h which will provide a uniform cdrom.h to all platforms on which HAL has been supported.
3) There is nothing innovative about ReiserFS; having a look at the structure, and the ideas are nothing original – just like ext3fs is nothin original.
Are we talking about ReiserFS or Reiser4? Reiser4 is definitely innovative. Have you looked at it yet?
Are we talking about ReiserFS or Reiser4? Reiser4 is definitely innovative. Have you looked at it yet?
Yes, and it is built on the same ideas in ReiserFS v3, and improving and expanding; nothing wrong with that, but the idea of dancing tree’s and the like, again, are nothing new; HFS+ has dancing tree’s, but we don’t see Apple jump around?
Oh, it’ll be interesting to see if ZFS is offered on MacOS 10.5, give its support for massive files and massive volume sizes.
Yes, and it is built on the same ideas in ReiserFS v3, and improving and expanding; nothing wrong with that, but the idea of dancing tree’s and the like, again, are nothing new; HFS+ has dancing tree’s, but we don’t see Apple jump around?
What about the plugin architecture, the automatic repacker, and atomic transactions? It’s really very different from ReiserFSv3.
Atomic transactions – nothing new; its just the concept of either its been written or not, rather than a merky middle ground; ZFS has this, I’m sure VxFS had it as well.
Like I said, I’m not trying to demean the very hardwork in which these developers do, and bring some of these awsome technologies to underlyings like me, but at the same time, it is prudent to keep ones feet on the ground, and realise that not everything out there is a revolution of innovation, but rather, ian evolution of previous ideas – take an existing one, extend the good parts, fix the no-so good parts, and come out with something even better than before.
Atomic transactions – nothing new; its just the concept of either its been written or not, rather than a merky middle ground; ZFS has this, I’m sure VxFS had it as well.
That’s not the point. Who cares what other filesystems have had atomic transactions. Did those filesystems also have a plugin architecture and an automatic repacker. Are those filesystems faster than Reiser4?
Like I said, I’m not trying to demean the very hardwork in which these developers do, and bring some of these awsome technologies to underlyings like me, but at the same time, it is prudent to keep ones feet on the ground, and realise that not everything out there is a revolution of innovation, but rather, ian evolution of previous ideas – take an existing one, extend the good parts, fix the no-so good parts, and come out with something even better than before.
I do realize that. You have to realize that this applies to everyone, including SUN. Everything in computer science is evolutionary. You need necessary building blocks to build upon. My issue in this thread began when people started claiming that innovation was exclusive to SUN. They aren’t the only software company to innovate, not anymore than any other software company. If you want to call ZFS of NFS innovation then something like Reiser4 or GCC or Linux itself would have to qualify as innovative. They are all repackaged ideas in one way or another.
So I couldn’t stop myself, and I read some more, and I saw this
IDE
While KDE is a nice, flexible desktop, Solaris beats linux hands down if you use workstation with the SunPCi IIIpro card installed:
So when discussing Integrated Development Environments, the author suggests KDE, which as a desktop enviornment is an entirely separate thing, and then goes on to say Sun’s IDE support is so much better because a Sun box can have a computer on a card that lets you run windows.
W-T-F!!!!
Both Linux and Solaris share the entire software stack. Indeed, most Linux distros provide a better out of the box desktop experience then the Solaris equivalents. Anything Linux has, Solaris has, which invalidates a good chunk of this paper.
There’s also the question of why he’s talking so much about Windows in a comparison of Solaris and Linux.
What about this:
Debugging
UltraSparc has cleaner architecture and debugging for Solaris is generally easier due to big Endean CPU with RISC instruction set. x86 is a way too cumbersome architecture for debugging 😉
For tracking system calls Solaris has an excellent truss utility. For Linux, the strace utility provides a similar but weaker function. For more complex situations Solaris has Dtrace. Currently Linux has no analogs for Dtrace.
Who debugs assuembler?! And what does endianness have to do with anything?! And what does Sparc vs x86 have to do with Linux versus Solaris, given that both operating systems can be run on both architectures?!
As regards DTrace, Linux has an inferior analogue in the form of kprobes.
I’m sorry, but OSNews just got duped: this guy hasn’t got a clue, and the piece is total nonsense. Sun have contributed more innovation, and Solaris is, I feel, a more solid server kernel (though I think Debian is a better operating system for servers), but this article is just sillyness.
Edited 2006-06-14 17:42
Anything Linux has, Solaris has
Not entirely true.
Quite a few proprietary programs that Linux based OSes enjoy don’t exist on Solaris just yet. VMWare and modern versions of the Acrobat reader come to mind.
What are you talking about? Acrobat reader existed on Solaris before it did on Linux. Hell, version 7.0.8 is available for download. (http://www.adobe.com/products/acrobat/readstep2_allversions.html) Hit the drop down and it is there clear as day! As for VMWare, well their low end offerings are only available on Linux and Windows. The only thing that Linux has on Solaris, is that there are far more Linux trolls than there are Solaris trolls.
Find me a recent version of Acrobat Reader for Solaris x86. I’ll be a happy camper, seeing as how that’s what I’m posting this message from.
As for VMWare, as far as I’m aware none of their offerings support Solaris as a host OS.
And don’t even get me started on driver support.
Nevada’s getting better but it doesn’t hold a candle to Linux on that front.
From the otherside, I’d kill to see things like Dtrace and ZFS on Linux, Ditto for the presence of a stable ABI.
Edited 2006-06-15 01:18
Hmm, those Solaris technologies won’t appear in Linux due to licencing issues, but in regards to FreeBSD, Dtrace is already there, it’ll only be a matter of time before ZFS makes an appearance.
Solaris is also the most scalable OS on the market (with amazing scalability from one to more then 100 CPUs
That’s funny because Linux can scale from 1 to more than 1000 CPUs.
While we’re mostly a Linux/Intel shop here at InfoWorld, I don’t feel nearly as fired-up and evangelical about Linux as I did five years ago when I convinced the marketing department at my then-employer to issue a press release trumpeting our migration to Red Hat Linux (spawning a Slashdot posting, a Webmonkey article, a piece in PC World, and a mention in a News.com story ). Despite the flurry of attention that migration brought, the dirty little secret of this Linux proponent was the Sun E450 I had in the server room running our content management system (which was open-source based and ran great on Solaris) and our mission-critical Oracle database. Linux was a bit risky at that time, so I wasn’t foolish enough to migrate *everything* to Linux — we had a content business to run, so the core content-producing system stayed on Solaris/Sparc. An excellent choice, but no one issues press releases about rock-solid Sun boxes running Oracle. They were just too commonplace.
That comment was posted in 2004 which would mean that Linux wasn’t ready to handle the core content producing for this particular company in 1999. What does that have to do with Linux today? The author of this article trumps all the newest features of Solaris then compares them to seven year old Linux.
Fast forward to the present. The Red Hat Linux I run isn’t free (as in beer) anymore, and though the mission-critical Oracle I used to run on Solaris is now humming away on Red Hat Advanced Server on HP hardware, it’s not officially supported on Debian, our Linux distro of choice for our front-end web servers, mail servers, etc., so I’m stuck working with two Linux distributions.
Hmm. Ever hear of CentOS? It is free as in beer. You can even run web servers on it! I don’t think it is the fault of Linux that your particular distro of choice is Debian and Oracle doesn’t support it. The easy thing to do would be to switch all your servers over to supported distributions like Red Hat or Suse.
It’s obvious from the introduction that the author has a bias towards Solaris. Just look at some of his bullet points if you haven’t read the article yet:
Uniformity. The major problem here is multiple personalities of linux. Sun is a single vendor of Solaris, and Solaris exists as a single OS with optional open source components that you can install or can ignore. At the same time linux consists of Linux kernel plus environment, created by multiple groups. That creates many subtle differences in security space (for example please compare SELinux and Apparmor: the letter is different and more user-freindly)
What utter crap. The author fails to mention that Solaris isn’t the only UNIX.
Commonality. Common, popular OSes are the most favorite target of attacks (“Microsoft effect”). Linux status is Unix space is pretty much similar to status of Windows in general OS space: this is the most hacked flavour of Unix. Period. Also usage of GCC compiler and precompiled kernels are major disadvantages from the point of view of security as this greatly simplify exploiting buffer overflows (some Linux distributions like Gentoo are free from this limitation; also some recent work in gcc might lead to randomizing of code that can help to prevent such attacks).
That’s odd considering what the author says on the previous page:
Only recently linux start winning big financial services and retail accounts, especially as a Web front-end server park OS of choice, but it is important to understand that is remains pretty much niche OS in enterprise space with total market share probably less then 10%.
Nuff said.
Protection from buffer overflow. Solaris rules and Linux sucks here ;-). There is simply no comparison in this important area as Solaris has hardware-based protection both on the Opteron and UltraSparc chips.
Linux also has this protection when used on hardware that supports it, when the hardare doesn’t support it Linux has PAX that will do the same thing in software.
Virtualization.
I won’t repost the entire paragraph but basically the author complains about lack of virtualization before admitting that Xen is now included with current and upcoming distributions.
Role based access control. Solaris rules here. Red Hat is limited to SELinux functionality (incompatible with AppArmore solution used in Suse) that looks inferior and too over-engineered in comparison to Solaris 10 RBAC capabilities.
I really don’t see how SELinux is “limited and incompatible” just because it doesn’t interact with AppArmor. It doesn’t need to. The author also fails to mention RSBAC.
http://www.rsbac.org/
Alright this post is getting long enough. It would take me pages to dispute everything the author claims. This article is FUD, plain and simple.
Edited 2006-06-14 17:36
That’s funny because Linux can scale from 1 to more than 1000 CPUs.
The biggest linux server you can get supports 512 CPU compared to 128 for sparc/solaris. So while Linux might in theory support 1000s of CPUs no one has ever tried it.
Actually Solaris can support 144 CPU’s, the maximum configuration of the SunFire E25K.
IIRC, the ‘512 cpus’ which the parent poster (who you replied to) points out, is for a very specialised, customised server. The said poster also thinks that the only important thing in an OS is the kernel, completely forgetting the important components like libc, libpthreads etc. etc. which also dictate speed and scalability etc.
> is for a very specialised, customised server
Care to explain? asside from big//expensive its probably less so than the sparc box, the ibm p5xx, or a PS3
“Linux might in theory support 1000s of CPUs no one has ever tried it.”
more than 1000 = 1024 and yes they have.
Whom? Where? Do you have a link?
SG, wisconsin, http://www.ci.chippewa-falls.wi.us/
thanks
Maybe it’s just me, but I can’t find any links to any 1024CPU computer on that website. Do you have a direct link to the tech specs?
We were discussing testing, not product. I don’t know whether or SGI/Novell will qualify it, you will have to wait and see. As for specs its a modular platform so take the 512P Altix 4000 specs and x2 them
Some sites do apparently run 2048 CPU SSIs, NASA
Ames’s 2048 BX2 CPUs, and I think the Japanese atomic
energy agency’s 2048 CPUs have (and may still) run in
that configuration in production.
I think the issue is that they’ll probably not be on
standard support contracts.
I have heard from an SGI engineer that their newest
hardware theoretically supports 16384 CPUs, and they
have tested up to 4096 internally (although that may
be on a simulator rather than real hardware).
SGI’s Altix supports up to 1024 processors and it runs on Linux.
So you are saying I can download any Linux distribution off the internet and just install it on an Altix box with 512 cpus without modifying the kernel or recompiling it and expect it to work without batting an eyelid?? Have you used it in such a situation yourself?
I know I can download Solaris 10 from the Sun website and use it on my one processor workstation and my 64 processor server out of the box (maximum I have used it on) No modifications necessary. This is what they meant by scalability.
Edited 2006-06-14 22:29
“So you are saying I can download any Linux distribution off the internet and just install it on an Altix box with 512 cpus without modifying the kernel or recompiling it and expect it to work without batting an eyelid?”
I remember reading a post from an SGI engineer on fedora-devel where he explicitly stated that they were able to boot the stock (Itanium) Fedora kernel on a 512-CPU box. This was around FC2, BTW.
But the best results are probably only obtained by patching the kernel, which is common practice on Linux-land where no distribution ships the vanilla kernel.
”
So you are saying I can download any Linux distribution off the internet and just install it on an Altix box with 512 cpus without modifying the kernel or recompiling it and expect it to work without batting an eyelid??”
I believe debian for x86 doesn’t work as well as SLES9 for Itanium.
So you are saying I can download any Linux distribution off the internet and just install it on an Altix box with 512 cpus without modifying the kernel or recompiling it and expect it to work without batting an eyelid?? Have you used it in such a situation yourself?
So do you have a 512 CPU machine lying around? Are you going to install an OS you downloaded from the internet to install on a supercomputer? Are you crazy? Linux scales just fine thank you. If you want a 1024 CPU machine running Linux you can have one (you can even get 2048 if you want!). If you want an 8 way machine you can have one of those too. If you want a laptop you can have that, even a wristwatch running Linux if you would like. I call that scalability.
I know I can download Solaris 10 from the Sun website and use it on my one processor workstation and my 64 processor server out of the box (maximum I have used it on) No modifications necessary. This is what they meant by scalability.
Oh did I forget to mention that you can download Linux off the net and put it on a 64 CPU machine if you would like. Take a look at this article:
http://news.zdnet.co.uk/0,39020330,39184546,00.htm
Pay special attention to this statement:
The HPL benchmark, which is used to measure performance when solving large linear equations, produced similar results, rising from 18 gigaflops with one cell of four processors to 277 gigaflops with all 16 cells, or 64 processors, running.
“This was a standard Linux distribution,” said Cabaniols. “The kernel was able to discover the topology of the system and discover the memory in a NUMA pattern.”
And this one:
“The 2.6 kernel is NUMA aware,” said Cabaniols. Some patching was necessary, he said, but “all patches developed for the BigTux project are going into the mainstream Linux kernel and are included in standard distributions.”
That’s funny because Linux can scale from 1 to more than 1000 CPUs.
as far as i know the 100cpu solaris is smp and the 1000cpu linux is numa (or should i say was numa, as sgi will fade away). i don’t think linux scales as well as solaris in a smp-setup
as far as i know the 100cpu solaris is smp and the 1000cpu linux is numa (or should i say was numa, as sgi will fade away). i don’t think linux scales as well as solaris in a smp-setup
Why does NUMA get discounted? It’s a single image running on up to 2048 processors. If scalability sucks so much then why does Linux run on supercomputers and Solaris doesn’t?
Because supercompter codes are (mainly) userspace and enterprise code are (more) kernelly.
Sun built a bus that you can load 100 CPU onto? Man that is one helluva bus, can you pass that smoke this way please
>That’s funny because Linux can scale from 1 to more than 1000 CPUs.
Who will sell you a single system image Linux system with that many CPUs?
Sun have a reasonably good story here, since they practice what they preach and sell and support systems with many CPUs. Now, if only those CPUs where fast ones. 😉
some comments:
Protection from buffer overflow. Solaris rules and Linux sucks here ;-). There is simply no comparison in this important area as Solaris has hardware-based protection both on the Opteron and UltraSparc chips.
WTF. Linux supports “hardware-based protection” just as Solaris, Windows and others – this is hardware-dependent, not OS-specific (the OS just needs to setup the page tables). In fact, AFAIK Linux may have supported this in other architectures (PPC, or even sparc?) for a some time.
Red Hat is limited to SELinux functionality (incompatible with AppArmore solution used in Suse) that looks inferior and too over-engineered in comparison to Solaris 10 RBAC capabilities.
“Looks”? It looks different for me, I love SELinux design, but then it’s my opinion…like the one on that paper?
Linux generally badly miss virtualization bandwagon. Currently only linux distributions with Xen support can complete with Solaris 10 in the area of application security. All others fall far behind.
“Only distributions that include virtualization support can compete with Solaris virtualization and those who don’t include it can’t”. ???
Not to mention that it fails to mention UML and/or vserver.
So adding linux to the usually large enterprise mix evaporate all real saving form lower cost of hardware and actually makes running it more expensive in comparison with existing OSes.
Yes, adding Linux to your enterprise means Yet Another Operative System that you have to support. But I fail to see how this is “Linux fault” and a “solaris advantage”. This reasing should be applied for all OSes, not just everyone but Solaris.
FreeBSD, NetBSD and OpenBSD provide a free OS platforms with different emphasis (FreeBSD with emphasis on PC compatibility, NetBSD with the emphasis on portability and OpenBSD with the emphasis on security and simplicity). All of them are competitive to linux on low end and in several aspects are architecturally superior
“$FOO is better than $BAR”. This same argument can be reversed – just change BSD for Solaris, or Linux by Windows. I happen to think (and testing proves it) that open BSDs are seriously falling behind in some critical features like good SMP scalability, WRT to Linux and Solaris.
As for SMP support linux is limited by its tuning for low-end as well as Intel hardware
1) That was years ago, time passes. 2.6 scales just well to computers with lots of ram and lots of CPUS, like the 512-cpu SGI beasts, and IBM power machines with more than 100 cpus
2) Linux supports other architectures, not just Intel. In fact, Linux supports way more architectures than Solaris (including sparc, although not aswell as Solaris, obviously), it’s the number 2 operative system in that field, behind netBSD
Here Solaris has huge advantage over linux. 64 CPU Sun servers are reality rather then curiosity (the upper limit of SMP scalability for Solaris exceeds 100).
Again, Linux runs (and scales well, according to noumerous benchmarks available across internet) in machines with more than 100 CPUs
That achievement was matched by Linux only 7 years later with the integration of XEN.
UML has existed for a long time and ISPs have been using it for years. Solaris 10 started releasing XEN support really recently BTW, on February, well behind Linux.
One of the big differences between Solaris and the other two OSes is the capability to support multiple “scheduling classes” on the system at the same time.
Linux developers find “pluggable CPU schedulers” ugly. Patches have been proposed and posted for years, the way of thinking of Linux developers is: “either you have a single, good CPU scheduler, or your system is broken. If you need to implement pluggable schedulers it’s becuase your main scheduler is broken and sould be fixed”. So no, Linux cpu scheduler is not like the one in Solaris, but it’s a design decision (you mamy like it, or not), not a “misfeature”.
Solaris kernel is compiled using Sun proprietary compiler that produces a reasonably optimized code
So does Linux and GCC. GCC is not the best compiler ever, but it’s not the worst one. It certainly supports way more hardware architectures than any other compiler on earth.
Sun Studio 11 complier beats GCC on SPARC in all major tests.
And why Sun compilers wouldn’t be the best in their own architecture?
Conclusion: It’s just me, or this article is slightly biased towards Solaris? Solaris is a great operative system, but that doesn’t means Linux is not the crap that articles tries to make you to think.
And why Sun compilers wouldn’t be the best in their own architecture?
I think the bigger question; why didn’t the guy test Solaris x86 using Studio 11 and Linux x86 using Studio 11?
…and IBM power machines with more than 100 cpus
which machine is that?
Solaris scales better than Linux in all cases. The Solaris kernel has been highly tuned for better SMP. This is exemplified in some of the benchmarks that Sun released with the Solaris optimized version of MySQL. Not a pure SMP benmchmarks but it certainly gives you a taste.
There should be more SMP benchmarks: Solaris Vs Linux Vs BSD Vs Windows Vs OSX. 1 CPU, 2 CPUs, 4 CPUs, 8 CPUs and 16 CPUs. Lets choose some applications with lots of message passing and shared memory and do some benchmarks.
Right. We’ve heard it all before: Linux is for geeks, Solaris is better. bla bla.
And scalability? Yeah. Right. Linux sucks there too. That’s the reason Linux is market leader (above 70%) in HPC…
Let’s put this “scalability” discussion into perspective. First the article primarily discusses SMP implementations, not HPC so the “512 CPU scalability” argument doesn’t apply since we are not talking about HPC.
I was in a meeting with serveral sales engineers from HP and of course they wasted no time in trashing Sun at every turn (at this point exchange RedHat with HP and you get pretty much the same thing). HP was touting their x86-64 machines and I brought up the question about support of Linux on their Integrity servers, particularly support for nPars and vPars (hardware and software virtualization) if an x86-64 solution would not meet our requirements. Currently there is no support for either feature in RedHat or SuSe Linux (the two distributions supported by HP). One of the sales engineers said it would be available “in about three months”. One could look at this as “vaporware” (cue shaman).
If I was going to propose spending the kind of money an Integrity server, I would like to know that I can utilize all of the functionality that machine can provide, so why would I want to use Linux on this machine since it cannot support virtualization (but HP-UX does).
It is not just about scalability it is also about flexibility, and in this case buying big iron and running Linux on it does not necessarily get “most bang for the buck” that Solaris does. And that is what I am looking at, just because Linux runs on the hardware doesn’t mean I can leverage it the way I can with another OS (AIX, Solaris or HP-UX).
That benchmark was not about scalability. Solaris was being much better than Linux even on single-thread tests, ie no SMP scalability was involved.
…I agree with his general conclusion:
Linux has the advantage on workstations and “low level” servers.
It’s a wash on “midrange” servers.
Solaris has the advantage on “high end” servers.
Sounds about right to me. That said, I do question how important the high end server market will prove to be over time. The old guard Unix crowd were brought up to regard vertical scalability as a kind of holy grail, and much of the competition in the Unix market of the late eighties and nineties was staged in terms of a battle over SMP support and the support for threaded applications able to capitalize on a high number of processors.
But then something truly disruptive happened to the market, commodity x86 hardware became very powerful and, more importantly, extraordinarily cheap. While it’s true that x86 hardware is lacking many of the high end features of big iron, which allow for better vertical scalability and greater reliability, this has been mitigated to a large extent by the use of clustering and horizontal scalability (replacing one big iron box with numerous x86 commodity ones).
I know a lot of the old guard view this as an inelegant and hacky cludge; cheating almost. And, in the purest sense, I agree. It is a more inelegant solution, but–and this has always been the key to Linux’s success–it performs “good enough” at a fraction of the cost.
There will probably be a market for big iron for a long time to come–it is the only viable achitecture for some workloads–but look at the market data for last five years. All philosophical questions aside, the market for big iron boxes has remained completely flat, while the market for 4-way and 2-way commodity x86 servers has exploded.
The author makes some good points while the article is overall biased. Mysql does run faster on Solaris than it does on Linux and this is due to the fact that Solaris handles highly multi-threaded applications better.
I really don’t see how the author can say that Solaris’s mandatory access control is better than SELinux and *actually* be serious!? It’s true that most of the features of the older Trusted Solaris have been rolled into the latest greatest version of Solaris. I really see that SELinux has the edge over the Solaris implimentation of MAC though.
He says that the Solaris RBAC is more mature than Linux’s SELinux. How so? Sun had to modify and rewrite parts of it to go into standard Solaris and not in it’s own seperate OS (Trusted Solaris). This means that it isn’t nearly as tested or as battle hardened as SELinux.
The article is good, yet the author is biased to the point of placing very unfounded assumptions between facts. Solaris beats Linux in multithreading speed and builtin virtualization. Saying that SELinux in weak and “cuts corners” is a blatant lie where the real fact is that Solaris’s RBAC is still very untested since it was rewritten.
I really see that SELinux has the edge over the Solaris implimentation of MAC though.
Why? and How so? please explain.
He says that the Solaris RBAC is more mature than Linux’s SELinux. How so? Sun had to modify and rewrite parts of it to go into standard Solaris and not in it’s own seperate OS (Trusted Solaris). This means that it isn’t nearly as tested or as battle hardened as SELinux.
Solaris had RBAC since Solaris 8 which was released in january of 2000. SELinux started as a project in 2001.
Saying that SELinux in weak and “cuts corners” is a blatant lie where the real fact is that Solaris’s RBAC is still very untested since it was rewritten.
Please explain where you got the impression that RBAC in Solaris which has been there for more than 6 years is not tested enough.
Solaris had RBAC since Solaris 8 which was released in january of 2000. SELinux started as a project in 2001.
Just a minor correction. SELinux in a very primitive form existed in Decemeber 22 2000. It didn’t become very useable until 2001.
So speaking of unstable code. SELinux started out in 2.2.x and then was ported to 2.4.x and now 2.6. So RBAC in Solaris is far more stable and tested if you go by the new port philosophy.
Pardon my ignorance. I knew that Trusted Solaris was based upon Solaris 8 but I didn’t know that the rbac extensions were rolled into Solaris 8 at a later date.
I wrongly assumed that trusted Solaris was Solaris + MAC
(Mandatory Access Control). I know that Sun just phased
out Trusted Solaris in favor of Solaris 10 with the
Trusted Extensions.
When I think of SELinux and Solaris, I think of
Information Assurance and MLS (Multi-Level Security).
Solaris 9 didn’t provide that, but Solaris 10 with the
Trusted Extensions will. Redhat Enterprise Linux 5 also
will. Neither RHEL or Solaris are EAL 4+ RBACPP certified yet. RHEL is also going for LSPP which
even Trusted Solaris or the TBA Trusted Extensions
don’t have. LSPP = Labeled Security Protection Profiles.
http://www.sun.com/software/security/securitycert/trustedsolaris.xm…
Notice that LSPP is not on there, only RBACPP.
Since NIAP just transitioned away from NIST, they don’t have everything up, but here is a google cached copy of their old site:
http://tinyurl.com/lvg6n (CTRL F “Red Hat”)
http://h10018.www1.hp.com/wwsolutions/linux/solutions/cchsa/index.h…
http://www.sun.com/solutions/documents/articles/go_solaris10_common…
I guess I was thinking of more than just RBAC when I
made those statements.
Pardon my ignorance. I knew that Trusted Solaris was based upon Solaris 8 but I didn’t know that the rbac extensions were rolled into Solaris 8 at a later date.
Trusted Solaris versions of Solaris 2.5, 2.5.1, 2.6, 7, 8 have always existed.
http://docs.sun.com/app/docs?q=trusted+solaris&p=
Sun later with 9 realised that creating a trusted Solaris version was too expensive and with Solaris 10 merged the features into the base OS.
I wrongly assumed that trusted Solaris was Solaris + MAC
(Mandatory Access Control). I know that Sun just phased
out Trusted Solaris in favor of Solaris 10 with the
Trusted Extensions.
I don’t think your assumption was wrong.
When I think of SELinux and Solaris, I think of
Information Assurance and MLS (Multi-Level Security).
Solaris 9 didn’t provide that, but Solaris 10 with the
Trusted Extensions will.
Yes there was no Trusted version of Solaris 9 so I think you are correct.
Neither RHEL or Solaris are EAL 4+ RBACPP certified yet. RHEL is also going for LSPP which
even Trusted Solaris or the TBA Trusted Extensions
don’t have. LSPP = Labeled Security Protection Profiles.
http://www.sun.com/software/security/securitycert/trustedsolaris.xm…..
Notice that LSPP is not on there, only RBACPP.
From the website you linked.
Trusted Solaris 8
….
Protection Profiles
Labeled Security (LSPP); Role Based Access Control (RBACPP); Controlled Access (CAPP); plus Trusted Desktop and Trusted Networking
Assurance Level
EAL4+
Trusted Solaris 8 supports LSPP and RBACPP and is EAL4+ certified.
I guess I was thinking of more than just RBAC when I
made those statements.
That is fine. Except Solaris 9, Solaris in general (i.e trusted Solaris) covers security pretty well. And since Solaris has had a trusted version for 11 years I would say it has had more test coverage and is a more mature code base than SELinux.
Trusted Solaris 8 (a totally different product and fairly difficult from normal Solaris at the time) used label based security. Solaris 9 and Solaris 10 without the Trusted Extensions does not use Label based security.
Also, I read this article, but didn’t link it. Trusted
Solaris was LSPP certified, but it isn’t being rolled
into the Trusted Extensions 100%, it was simply too
difficult to manage. Notice here that Solaris 10 with
Trusted Extensions is not going for LSPP.
http://www.sun.com/smi/Press/sunflash/2005-10/sunflash.20051026.4.x…
Once the niac website is done moving their website
completely away from nist, you can see what products
are undergoing common criteria certification and for
exactly what levels of certification.
Thanks for debating this intelligently and not making
it a Linux vs Solaris flamewar. I just modded you up
one for the ability to hold an intelligent debate.
Notice here that Solaris 10 with
Trusted Extensions is not going for LSPP.
http://www.sun.com/smi/Press/sunflash/2005-10/sunflash.20051026.4.x…..
May be I am not reading the article you linked correctly. Here is a passage from the above link:
B>Furthermore, the Solaris Trusted Extensions layered technology plans to bring multi-level security to the Solaris 10 OS by the first half of 2006. Sun plans to submit this product for CAPP, RBACPP as well as Labeled Security Protection Profile (LSPP) at EAL 4+. Currently, Sun offers the Trusted Solaris 8 OS, which carries CAPP, RBACPP and LSPP at EAL 4+.
It seems to say LSPP will be in Solaris Trusted Extensions and Sun will submit it for EAL4+. No?
This link I posted above gives a more comprehensive feature list on the Trusted extensions fo Solaris10
http://www.sun.com/smi/Press/sunflash/2006-02/sunflash.20060214.3.x…
Any way the point is moot. IBM and Redhat have an OS they can sell to governments and so does Sun. Suffice it to say most users will never use these features.
Edited 2006-06-15 03:00
<me> I really see that SELinux has the edge over the Solaris implimentation of MAC though.</me>
<Arun>Why? and How so? please explain.
Well the Solaris version of RBAC uses path based
security where SELinux uses Label based security. If 1
file with a privileged path is compromised, the whole
system can potentially also be compromised on a Solaris
system. Labeled based security and label based
networking will be in the Solaris 10 Trusted Extensions
rollout and I believe might be in Solaris 10 on a
limited basis. SELinux had label based security from
day 1.
Notice how in my other post that Solaris 10 is
being evaluated for EAL 4+ RBACPP and RHEL 5 is being
evaluated for EAL 4+ RBACPP/LSPP. It is ahead of
Solaris in that regard. You can google about those
acronymns and what they mean, you are a very
intelligent individual.
Notice how in my other post that Solaris 10 is
being evaluated for EAL 4+ RBACPP and RHEL 5 is being
evaluated for EAL 4+ RBACPP/LSPP. It is ahead of
Solaris in that regard. You can google about those
acronymns and what they mean, you are a very
intelligent individual.
Solaris 10 is being evaluated for RBACPP and CAPP.
http://www.sun.com/smi/Press/sunflash/2005-10/sunflash.20051026.4.x…
Solaris 10 tursted extension will include LSPP.
http://www.sun.com/smi/Press/sunflash/2006-02/sunflash.20060214.3.x…
So RHEL is not ahead of Solaris because Trusted Solaris 8 is already EAL4+ ceritified in all the profiles IBM and RedHat are submitting RHEL 5 for certification.
Where did all the Sun astroturfers come from? Not just here, but everywhere… there were none of them two years ago and suddenly all over the Internet there is “BUT SOLARIS SCALES BETTER BLAH BLAH APIS BLAH BLAH DTRACE BLAH BLAH HUG ME” It definitely isn’t a result of Solaris 10 release.
Linux scales. It leads in HPC circles.
Linux is highly innovative.
Linux supports many more hardware architectures than anything but NetBSD.
Linux is stable.
Dtrace is a tool among many. Nice, but hardly earthshaking.
Sun made NFS and until version 3.x it was utter pants.
YP/NIS/NIS+ are horrid hacks. Sun didn’t create PAM.
Nothing wrong with Solaris. Nothing wrong with Linux. Sun makes great use of Java and probably has a nicer all-in-one administrative solution, but that doesn’t make it faster, more stable, more flexible or more lightweight. Both are good, but Linux is the obvious leader at the moment.
On another note, Sun has failed to release a completed, stable version of some of its next-gen features which are being shouted from mountain-top to mountain-top by the astroturfers.
As for the latest MySQL results, many of those who are in the know have looked at those results with a very speculative eye towards marketing bias.
Seriously, enough.
Edited 2006-06-14 19:10
“Where did all the Sun astroturfers come from? Not just here, but everywhere… there were none of them two years ago and suddenly all over the Internet there is “BUT SOLARIS SCALES BETTER BLAH BLAH APIS BLAH BLAH DTRACE BLAH BLAH HUG ME” It definitely isn’t a result of Solaris 10 release. ”
Why is anybody who likes Solaris get branded an astroturfer by you? Why are 90% of your posts to OSNews accusing somebody of being an astroturfer, or attempting to start an argument? Why do you never back up your statements with factual information?
“As for the latest MySQL results, many of those who are in the know have looked at those results with a very speculative eye towards marketing bias. ”
Yeah? Odd. Are you “in the know”? Are you looking at it with a speculative eye? What have you seen? Oddly enough, the mySQL AB guys say it’s realistic. Also, strangely enough, I see about 15% better performance on my widely varied workload. This is documentable factual information. Where are your facts?
To return fire along your seemingly ingrained ways, where did you come from? I never saw anything from you until the past year or two, then all of a sudden there was a Shaman explosion, and OSNews readers everywhere sobbed. It’s funny, some of the posts in response to Sun/Solaris articles even mention you by name *before* you post. So, if you want to call a bunch of people astroturfers, then I suppose a bunch of people can call you a troll. I’m inclined to go along with the troll theory.
Ass.
>Why is anybody who likes Solaris get branded an
>astroturfer by you?
Because the posts are OBVIOUS fanboy/astroturf posts. They present no facts, ideas or objectivity. Just praise at the expense of any other point of view.
>Why are 90% of your posts to OSNews accusing somebody of
>being an astroturfer, or attempting to start an argument?
Is that really my job? And pulling figures out of the air like 90% of my posts, I’d expect the same courtesy from you…
> Oddly enough, the mySQL AB guys say it’s realistic.
Oh, you think? I’d say it’s possible to write assembly/code that specifically works better with Sun’s compiler, too. What I wouldn’t say is that Solaris itself was the major factor. This is borne out in your own argument: single-thread performance. Perhaps sir is not a programmer…
> where did you come from?
15 years as an independent in the industry. Former Sun dealer, before Sun ran us over the coals repeatedly with software that was utter junk while they walked away from the problems. Former Sun dealer who was brushed aside more than once when Sun realized how big the opportunity we brought to them was (ala Cisco). I’ve partied with Ramsey at Keystone Resort, CO. My company was the first one to sell Javastations into the school system in Canada (and I think the largest actual deployment that ever actually happened), which Sun walked away from after weeks of trying to make their software work and failed at – but still shouted about in the papers. I still have a ton of Javastations that I made into various device containers. I presently have several UltraSPARCs, LX50s and old SPARCStations kicking around doing various menial tasks, running on a mix of operating systems. There are various other things on my resume, but that’s deep enough. I’ve been on the bad side of half-baked software and under-performing hardware that “the Dot in Dot.Com” was delivering in the late 90s. Who are you?
> some of the posts in response to Sun/Solaris articles
>even mention you by name *before* you post
I’d like to think the truth hurts.
Go with the troll theory, I’ll go with the theory that you’ve permanently got sand in your vagina.
Edited 2006-06-14 21:44
I bought a book on solaris 10. At the start the author makes a similar claim, that Linux is no longer free, and that Solaris is now free, and that support is cheaper from Sun.
What rubbish. SuSE Linux and RedHat is what Microsoft and Sun like to talk about because, yes, they are commercial organisations. They very rarely mention the huge range and number of free and open source distributions that are as good as if not better than SuSE or Red Hat Linux. Having said that, Sun is changing this, see recent announcements relating to Ubuntu on SPARC.
This article is incredibly biased. It very much seems to me to be a reactionary old-time Solaris user attacking Linux rather than trying to embrace it (like Sun appears to be doing, finally!).
Solaris is a great operating system, it has many advantages over Linux-based operating systems, – but lets have some objective reviews here, not one that is so terribly biased and based on flawed statements.
At the start the author makes a similar claim, that Linux is no longer free, and that Solaris is now free, and that support is cheaper from Sun.
What rubbish. SuSE Linux and RedHat is what Microsoft and Sun like to talk about because, yes, they are commercial organisations.
In addition, the author failed to mention the source code of Red Hat Enterprise (not sure about SUSE) are available for download under GPL license allowing the existance of other distros such as Scientific Linux, CentOS to name a few.
They very rarely mention the huge range and number of free and open source distributions that are as good as if not better than SuSE or Red Hat Linux. Having said that, Sun is changing this, see recent announcements relating to Ubuntu on SPARC.
Oh, if you want fairness; why don’t you talk about the distributions starting up, based on OpenSolaris?
“I bought a book on solaris 10. At the start the author makes a similar claim, that Linux is no longer free, and that Solaris is now free, and that support is cheaper from Sun.
What rubbish.”
Yes, there are many, many free Linux distros.
The major supported and certified distro’s aren’t free unfortunately. You can’t use Red Hat Enterprise without paying for a support contract. This is not the case for Solaris 10. Download Solaris 10 for free. Fedora is free, but most major apps aren’t certified to run on it. CentOS is free, but don’t expect support from application vendors. With S10 I can use a fully certified OS without paying for anything. It’s Oracle’s x64 operating system of choice too, so that works out nicely too.
Last time I checked Solaris support is cheaper (check sun.com and redhat to see for yourself).
You’re confusing two different kinds of scalability here. The systems you’re referring to in the HPC market are generally giant clusters of small boxes. What’s being referred to here is scalability in a single running instance of the OS. Apples to oranges.
People no longer seem capable of differentiating between the two. Most people here wouldn’t know what NUMA was if it bit them in the tail.
Horizontal scalability is all people that have grown up in the late 90s/early 00s know. Vertical scalability is a thing of a past to them (actually, most of them have seem to have no idea what it is..)
So, to them, they likely aren’t comparing apples and oranges. They just don’t know the difference between apples and oranges, and haven’t had a taste of either, so they just talk out their tail-end about what they *think* based on their obviously highly-educated guesses.
So, to them, they likely aren’t comparing apples and oranges. They just don’t know the difference between apples and oranges, and haven’t had a taste of either, so they just talk out their tail-end about what they *think* based on their obviously highly-educated guesses.
Babe, its known as what I like to call, ‘the google effect’ where by they scream, “hey! google can do that!” when ever they wish to diss Solaris, big iron midframes/mainframes.
Bascule had a good post explaining the differences, comparing the Google approach to a big iron running Oracle – too bad the Linux fanboys (and girls) can’t be bothered getting tuned into it.
“Babe”? Please, out of respect for me, don’t call me that.
Other than that, yes – concur, that’s generally what happens. Latency doesn’t mean a thing for some parallel jobs, to others it does. The jobs cited for examples don’t require extremely low latency between processors.
Something that requires in-order parallel processing wouldn’t be well suited to the “Google” style of clustering cheap boxes. When a reply might take an extra 1ms to get where it’s going, it can mean the difference between working and not working. Oh well. There are things somewhere in the middle (SGI), and then there are highly vertical scalable monster boxes from Sun. Of course you can find examples of each from other manufacturers, just using these as well known examples.
“Babe”? Please, out of respect for me, don’t call me that.
Sorry, force of habbit; hanging around with my gay friends for too long <blush>
Other than that, yes – concur, that’s generally what happens. Latency doesn’t mean a thing for some parallel jobs, to others it does. The jobs cited for examples don’t require extremely low latency between processors.
Hence the interesting thing when people look at POWER5, and the benchmarks, but as a Sun engineer said, thats all very fine and dandy, but the latency is terrible when compared to Sun’s bigirons.
Something that requires in-order parallel processing wouldn’t be well suited to the “Google” style of clustering cheap boxes. When a reply might take an extra 1ms to get where it’s going, it can mean the difference between working and not working. Oh well. There are things somewhere in the middle (SGI), and then there are highly vertical scalable monster boxes from Sun. Of course you can find examples of each from other manufacturers, just using these as well known examples.
Also, the accuracy of those results are not critical in the case of google; if a db falls over, who cares? it isn’t as though the results are as critical as say a large bank which needs to have accurate, to the letter, information being retrieved and displayed in a speedy time frame.
“People no longer seem capable of differentiating between the two. Most people here wouldn’t know what NUMA was if it bit them in the tail.”
This pops up every single time one talks about Linux scalability to large number of CPUs, and it’s annoying…
SMP and ccNUMA (“cc” – cache coherent – being the important thing here) are basically the same thing, a bunch of CPUs running a *single* system image, which means the same locking contention problems appear on both architectures, albeit with differences.
The difference between the two is the memory access, which (like the name says) doesn’t have uniform latencies/speeds for every address. ccNUMA schedulers are basically SMP schedulers that have to take memory locality into consideration.
SMP and ccNUMA are hardware architecture choices: on the SMP side the designer wants to make things simpler for the OS and optimizes for processes accessing mostly the same data, on the ccNUMA side the designer wants to maximize memory access bandwidth and optimizes for processes accessing their own data.
But in any case, all processes will have to make system calls, do IPC, and that where kernel scalability comes into play (together with making the best choices for memory allocations, CPU affinity, etc).
As to actual hardware…
A 512-CPU SGI Altix physically is a bunch of boxes with a couple of CPUs and memory, connected by an high-speed bus (NUMAflex on the Altix, if I recall correctly), but a big classic SMP box is also a bunch of cards with CPUs connected by an high-speed bus (but without any local memory).
The Opteron is ccNUMA, the Xeon isn’t. What’s the difference between scaling well to an n-way Xeon machine or an n-way Opteron? Get the picture?…
Moral of the story: 512-CPU SGI beasts AREN’T CLUSTERS.
http://oss.sgi.com/projects/numa/Linux_Scalability_for_Large_NUMA_O…
The big problem here is Solaris fanboys are so used to their “nothing scales like Solaris” idea that they mix hardware issues with software issues.
Nikolai Bezroukov is a well-known kook with an ‘alternative’ view on a lot of things.
http://www.grsecurity.net/features.php
Just reading the conclusions one knows this is written by someone who’s trying to sell you Solaris.
I would agree with some of what he has said but when comparing two operating systems you should be biased. If
I was comparing 2 OS’s I wouldn’t attack one of the creators of the operating systems.
I support HP-UX, Solaris, AIX and many Linux distros and find them all to have their benefits. Out of all of the *NIXes I am generally less impressed with Solaris. Just my opinion and I won’t be writing a 50 page paper secretly bashing one operating system calling it a comparison;)
Sun did create PAM. Here it is the RFC:
http://www.opengroup.org/tech/rfc/rfc86.0.html
You can see that V Samar and R. Schemers were the two SunSoft engineers that wrote the RFC. SunSoft was the software arm of Sun back in the ’90.
So Sharman, please do your research before posting. FUD is not the way Linux is better. That is how Microsoft Windows is supposed to play the game.
Linux and Solaris are both better OSes than Windows. Solaris being older (about 10 years older compare to Linux) is more stable and “scrutinized” in more depth, but because of its age, it carries more baggage than Linux. A lot of what makes Linux great is that it takes something that works and improve on it.
Linux and Solaris (aka OpenSolaris) camps should join together and fight the real enemy: Microsoft Windows and not fight one another for little things that a reviewer states in an article.
What a lot of people doesn´t seems to know is that Solaris is 100% backwards compatible with binaries for older solaris versions (i think it is back to version 2.5 or 2.6), that is something that is NOT true for linux.
So if you have an old software running on solaris 7 or 8 or sunOS 2.6, then it will run on Solaris 10 also, for sure.
So if you have an old software running on solaris 7 or 8 or sunOS 2.6, then it will run on Solaris 10 also, for sure.
Open source doesn’t have to worry about that. Software is just a recompile away most of the time.
On the negative side Solaris suffered from Sun’s preoccupation from Java and in a default installation has limited support for scripting languages: Perl is included with the Solaris 9 and 10 distributions but the number of packages in default installation is limited. TCL, while for many years was developed by Sun, needs to be installed separately. For PHP support you need to use Apache. Even ksh is still old ksh88 (POSIX shell), not ksh93. But company is slowly recovering from the Java fad and might put more resources into OS development including integration of scripting language.
This is very right. Sun pushed Java inside your troat when you had bought their hardware (IT Departements).
When you wanted to use another script language, even PHP (C Like) or Python (like for intranet portals)… Bzzz… not really supported ! Use Java, “it’s better”.
Linux isn’t bad; but it will always lack final polishment, it’s in its nature, it’s not a commercial standards OS.
(I hope you don’t mod me down for criticizing Linux again.)
Even Matthew Szulik chief executive of Linux vendor Red Hat, said:
http://news.zdnet.com/2100-3513_22-5101690.html
And that was 3 years ago. I don’t think Linux has changed on the workstation/desktop. Judging form the latest Fedora.
When you wanted to use another script language, even PHP (C Like) or Python (like for intranet portals)… Bzzz… not really supported ! Use Java, “it’s better”.
Hence the reason I think Sun should support mono, and given the have a comprehensive patent sharing agreement with Microsoft, they can easily shield their customers from any Microsoft litigating if Mono becomes too compatible.
Ultimately, Sun shouldn’t give a shit what is running ontop of their operating system, as long as the customer is running Solaris on Sun hardware, and getting Sun services – what they do with the equipment should not be the concern of Sun.
arguing about benefits of a free kitten vs free puppy.
You still have to feed both of them until they grow up
I’ll try to clear a few things up.
Firstly, Linux does run on the 512 CPU systems out of
the box on SUSE, and as far as I know also kernel.org
kernels.
Secondly, yes as others have pointed out, they are
single image systems (like the large Sun systems), and
I think they have tested up to 2048 CPUs in a single
system. SGI’s up coming next-gen hardware I think
expands this upper limit to 8192 (though Linux would
likely require a bit of work to go that high).
SGI’s systems are HPC focused, however Linux also runs
on HP superdomes and IBM POWER5+ pSeries systems, which
go up to 128 and 64 cores, respectively. And the
pSeries pretty much blows everything else out of the
water in enterprise computing.
Also, if you make the distinction between SMP and
NUMA, then Sun’s large 144 CPU systems (E25K)
are most definitely classed as NUMA. The data
interconnect topology goes:
– CPU local memory
– same board (over first level (2 CPU) board switch)
– same board (over second level (4 CPU) board switch)
– remote board (over FirePlane crossbar)
Maximum bandwidth decreases by a factor of 4, and
latency increases by a factor of 2.5 as you go out
the topology.
While this may not be as large a factor as something
like a 512 CPU SGI system, it is comparable to smaller
SGI systems, and “more NUMA” than IBM’s pSeries.
Finally, I’ve never seen a benchmark that proves
which is more scalable. Toy MySQL benchmarks on 2
CPUs have shown neither has a scalability advantage.
While benchmarks showing Linux scales better at
something like specjbb on big pSeries boxes than
Solaris on E25K boxes are invalid because the pSeries
hardware is so much more scalable.
So everyone arguing about that point may as well not
waste their time until they have some proof one way
or the other.
on HP superdomes and IBM POWER5+ pSeries systems, which
go up to 128 and 64 cores
—
Do HP support *linux* in a 128P partition or just HPUX?
Linux also runs on HP superdomes and IBM POWER5+ pSeries systems, which go up to 128 and 64 cores, respectively.
According to HP:
http://docs.hp.com/en/5991-1247/ch03s01.html
Linux is only supported on Superdome for a maximum of 16 CPUs.
Windows (surprisingly) supports up to 64, and
HP-UX up to 128.
Linux only supports a maximum of 128 GB of memory (which is much less than the hardware max).
Obviously these are only for Itanium superdome (Integrity), as Linux isn’t supported on their PA-RISC systems IIRC.
As for pSeries, AIX is by far more popular than Linux on this platform. I don’t think you’ll see many p5-595’s running Linux.
Besides, AIX provides more features when using LPARs (live memory reallocation etc. compared to Linux on POWER), among other benefits. IBM will probably try to sell you AIX over Linux anyway for pSeries.
According to HP:
http://docs.hp.com/en/5991-1247/ch03s01.html
Linux is only supported on Superdome for a maximum of 16 CPUs.
Windows (surprisingly) supports up to 64, and
HP-UX up to 128.
It will run on at least 64 (HP published some tests a
while back).
Linux only supports a maximum of 128 GB of memory (which is much less than the hardware max).
Linux has run with at least 4TB on SGI Altix systems.
Probably the most memory of any single system image
ever built. Far more than Solaris would ever have
run with.
As for pSeries, AIX is by far more popular than Linux on this platform. I don’t think you’ll see many p5-595’s running Linux.
Besides, AIX provides more features when using LPARs (live memory reallocation etc. compared to Linux on POWER), among other benefits. IBM will probably try to sell you AIX over Linux anyway for pSeries.
Sure, as the OS designed and built around that
platform, it is definitely a bit faster and more
scalable on it. Linux on pSeries still destroys
Solaris on anything that it can run on.
So HP can run Linux on an Intergrity server in the lab, point to a production instance. And Linux on POWER beating Solaris is based on what, show us that as well if you don’t mind.
I think the bigger question is this; if everything is purchased based on performance, then by going by the latest TPC benchmarks, wouldn’t it be true to assume that everyone is moving to Itanium based machines either running HP-UX or Windows 2003?
According to reality, as we all know, that isn’t the case – Sun’s server sales, especially their x86 ones, have been going gang busters in the last quarter; so if that rumour is anything to go by, there is more to life than just raw performance.
“Sun’s server sales, especially their x86 ones, have been going gang busters in the last quarter”
And those servers are running what? Mostly Linux and Windows 2003.
Actually no – care to back up your rectum pluck with some evidence?
Oh, hey I didn’t realise you were actually interested
in having a constructive conversation backed up by
real facts and numbers. In that case, can you please
back up your rectum pluck that NetBSD runs on more
platforms than Linux with some numbers?
I can’t count the number of platforms Linux supports,
because the ARM port alone contains over 50 (and how
do you classify the hundreds of x86 PC configs over
the years).
Counting by CPU architecture, Linux definitely supports
more (by about 50%). But I’d be very interested to see
your definition of platform, and the comparitive
numbers for each. Thanks.
Oh, hey I didn’t realise you were actually interested
in having a constructive conversation backed up by
real facts and numbers. In that case, can you please
back up your rectum pluck that NetBSD runs on more
platforms than Linux with some numbers?
You made a clear statement that BSD ran on more platforms than Linux – you failed to mention which BSD, as there are three flavours, and NetBSD is actually available with full support on 37 different architectures, 97 in total. Obtained off the http://www.netbsd.org site.
I can’t count the number of platforms Linux supports, because the ARM port alone contains over 50 (and how do you classify the hundreds of x86 PC configs over the years).
ARM is one architecture – whether it is made by Samsung, Intel, or some other company, its still one architecture.
x86 is pretty much the same architecture over and over and over again, with some tweaks here and there.
SPARC, PA-RISC, POWER, x86, x86(incl. x86-64), they’re different architectures, whether its made by Intel, AMD, SUN, Fujitsu or Via, it makes no difference to the software running ontop of it.
Edited 2006-06-16 10:18
You made a clear statement that BSD ran on more platforms than Linux – you failed to mention which BSD,
No that wasn’t me. I’m just wondering how you
conclude that NetBSD runs on more architectures than
Linux.
as there are three flavours, and NetBSD is actually available with full support on 37 different architectures, 97 in total. Obtained off the http://www.netbsd.org site.
No, you’re counting what NetBSD calls “platforms”.
Have a look under “Ports by CPU architecture” on
http://www.netbsd.org/Ports/ <– that page.
Now, even within this list, NetBSD has some funny
ways of counting architectures. For example, they
count big and little endian twice for mips and sh,
they count sparc and sparc64 seperately, counts
m68k twice due to different MMUs, etc.
alpha, arm, hppa, i386, m68010, m68k, mipseb, mipsel
ns32k, powerpc, sh3eb, sh3el, sh5, sparc, sparc64,
vax, x86_64 (17)
Even if we are generous to NetBSD and not to Linux
(eg. Linux supports multiple types of mmus, bit-sizes
and endianness in several architectures but only
counts them once), Linux supports *at least* as many
CPU architectures as NetBSD.
alpha, arm, cris, frv, h8300, i386, ia64, m68k, mips, parisc, powerpc, s390, sh, sparc, v850, x86_64, xtensa (17)
Actaully, the way you count NetBSD, you could also
probably add arm26, arm-nommu, m68k-nommu, sparc64,
sparc T1 (different MMU), mips64, sh64el, sh64eb, mipsbe, sh3be, off the top of my head.
ARM is one architecture – whether it is made by Samsung, Intel, or some other company, its still one architecture.
NetBSD counts it as multiple “platforms”.
x86 is pretty much the same architecture over and over and over again, with some tweaks here and there.
SPARC, PA-RISC, POWER, x86, x86(incl. x86-64), they’re different architectures, whether its made by Intel, AMD, SUN, Fujitsu or Via, it makes no difference to the software running ontop of it.
NetBSD count many of them as multiple platforms, even
though the usermode ISAs are the same.
That’s fair enough, but I just need you to come up
with the exact definition of portability, and numbers
you use to conject that NetBSD is more portable than
Linux. Thanks.
That’s fair enough, but I just need you to come up
with the exact definition of portability, and numbers
you use to conject that NetBSD is more portable than
Linux. Thanks.
But I never said NetBSD was more portable; you claimed that Linux was available on more platforms, to which I said, “what about NetBSD”?
I’m advocating neither because both do a good job, but you seem to be determined, like some sort of child, to scream, “look, look, I’m right, and you’re wrong’, in which case, I won’t participate in such a junvinile grand pissing competition.
But I never said NetBSD was more portable; you claimed that Linux was available on more platforms, to which I said, “what about NetBSD”?
For the last time, it was not me asking that question
before. However, you said:
“1) NetBSD supports more platforms than Linux”
So due to your utter inability to produce a single
scrap of evidence except for some vague references
to the NetBSD website, I’ll consider this is a
“rectum pluck” (to use your terminology).
I’m advocating neither because both do a good job, but you seem to be determined, like some sort of child, to scream, “look, look, I’m right, and you’re wrong’, in which case, I won’t participate in such a junvinile grand pissing competition.
You made an unqualified assertion that NetBSD supports
more platforms than Linux, so I think you have to
back it up when called to task. That is what a mature
discussion is all about.
If you don’t want to participate because you are wrong,
or you can’t produce any evidence, then just
gracefully admit it and move on.
So HP can run Linux on an Intergrity server in the lab, point to a production instance.
Well they sell and support Linux on 64 way superdomes,
like IBM sell and support Linux on 64 way pSeries.
http://h20341.www2.hp.com/integrity/cache/342254-0-0-0-121.html
And Linux on POWER beating Solaris is based on what, show us that as well if you don’t mind.
Almost anything you’d care to look at. The E25K with
its FirePlane interconnect and US CPUs is a dinosaur.
Everybody except Sun zealots will readily tell you
that.
On SPECjbb 2000 (a benchmark of Java performance),
Linux on a 32 core POWER5 scored 1,076,309. A year
later, Solaris on an 88 core E25K managed only 732,742.
Results submitted at the same time of a 64 core
POWER5 1.9GHz running AIX gets 2,505,245. More than 3
times faster with fewer cores. Assuming linear
scaling, the POWER5 is nearly 5 times faster than the
US, core for core. The number for a 2.2GHz POWER5+
would easily exceed a factor of 5.
SPECjbb 2005 shows a 144 core E25K with a score of 1,164,995 (not comparable to 2000 numbers). OK,
Fujitsu PRIMEPOWER is a little better, at 128
cores it gets 1,251,024 running Solaris.
However an SGI Altix with 128 cores running Linux gets
1,828,349.
Sun’s best SPECint/fp results for SPARC are 845/1353
IBM’s best for POWER are 1765/3513.
Sun doesn’t submit many benchmarks that can be easily
compared with other systems or software, but whenever
you can fairly compare a POWER or Itanium with an
ultrasparc, the ultrasparc is going to lose.
Throw something back at me.
“Well they sell and support Linux on 64 way superdomes,
like IBM sell and support Linux on 64 way pSeries.
” rel=”nofollow”>http://h20341.www2.hp.com/integrity/cache/342254-0-0-0-121.html”
The link you provide merely lists Linux as one of the operating systems supported on the superdome. It does NOT say that a *single* Linux instance can use 64 cores. (HP-UX supports 128). You will need to run multiple paritions (and therefore Linux instances) to fully use the available CPUs. You should study their docs better:
http://h10018.www1.hp.com/wwsolutions/linux/products/servers/integr…
16 Cores for Linux
64 Cores for Windows
128 Cores for HP-UX
First of all, I hope a “large enterprise” selects the best application that meets their specific business requirements *before* the OS/hardware platform. That in itself may select said OS/platform for them. (Don’t know why so many people want to select an OS/platform first, then shoehorn applications into that platform?)
Second, if your chosen application will run on either a commercial Unix or GNU/Linux, it probably boils down to corporate politics and/or corporate preference in the data center, very seldom is it truly based on technical superiority (if it actually exists). In any case, all sides can come up with enough marketing propaganda to slow down the implementation process for at least 3-6 months.
Third, after testing, tuning and documenting your selected application and selected OS/platform it will probably make little difference if it’s a Solaris, AIX, HP-UX or GNU/Linux (or even, *gasp* Windows) implementation, when in the hands of competent system designers and administrators.
Fourth, now let’s all join hands and sing kumbaya (but RMS needs a shower before we start hugging anybody
!
PS — Apparently there are many that cannot discern between SMP and NUMA architectures. They are *not* the same! Sun BTW *does not* sell NUMA systems! Doesn’t mean it’s good or bad, go back to my “First” point and start again. If your application works better with NUMA great, then pick an OS/platform, probably won’t be Sun. If you need a really big ass “large enterprise” RDBMS system, then Sun will most likely be in the mix (ie. NUMA not required or desired). This CPU-envy thing about my system has more CPUs than your’s is quite annoying and frankly alot of BS. If that was the end all of computing, I’d be running a Cray XT3 supporting up to 30K CPUs, so I guess standard Windows, Linux and Unix all suck….NOT!
PS — Apparently there are many that cannot discern between SMP and NUMA architectures.
OK, how about you enlighten us.
They are *not* the same! Sun BTW *does not* sell NUMA systems!
What is your definition of NUMA? Using the normal,
accepted definition (non uniform access to memory),
Sun most definitely does sell NUMA systems. Look at
any Sun whitepaper on their big interconnect
topologies.
Doesn’t mean it’s good or bad, go back to my “First” point and start again. If your application works better with NUMA great, then pick an OS/platform, probably won’t be Sun.
Actually, it is pretty much impossible to scale a
modern system above about 4 CPUs because of the
difficulty of signal distribution, not to mention
contention on shared resources (eg. memory).
“NUMA slows things down” is a common misconception.
NUMA allows things to be greatly sped up and scaled up.
Some corner cases may be almost as slow as a similar
SMP would be, which might make it appear to be fragile.
If you need a really big ass “large enterprise” RDBMS system, then Sun will most likely be in the mix (ie. NUMA not required or desired).
Any enterprise DB worth its salt (Oracle or DB2 come
to mind) are highly NUMA aware. They’ll spend literally
years of programmer hours to get 1% performance
improvement. So I don’t know what point you are trying
to make here.
This CPU-envy thing about my system has more CPUs than your’s is quite annoying and frankly alot of BS.
Apparently the Sun zealots weren’t at all annoyed when
they had the biggest systems on the block (and/or most
people didn’t know about SGI’s Origin+IRIX with 1024
CPUs).
But it isn’t BS, it happens to be one of the topics
being discussed. If you don’t like it, no need to
whine about it… What’s more, the Sun camp started
it by asserting that Solaris scales better than Linux
without any evidence to back it up with.
If that was the end all of computing, I’d be running a Cray XT3 supporting up to 30K CPUs,
We’re talking about scaling up in a single system
image (ie. cache coherent shared memory), so this is
nonsense.