Server maker Sun Microsystems has a new Linux partner for its "Niagara" family of multicore processors and their related servers: Wind River Systems. Sun has plenty of sway with telecom companies and the network equipment providers that make the switching gear they use. But since the dot.com bust, Carrier-Grade Linux has been on …
Sun was probably waiting for a mult-million dollar order promise before they made the move. Sales people are useful but sometimes they have too much influence. Thankfully, WindRiver does a good job so it should at least be tested before it ships(I know its a novel concept but I can hope).
Correction and Answer
Writer says, "Sun and Ubuntu Linux distributor Canonical announced that Ubuntu would be supported on Niagara servers using the four-core, eight-threaded Sparc T1 variants of the multicore chips."
This is incorrect. The SUN UltraSPARC T1 is an octal-core, four-thread-per-core, 32 threaded CPU.
Writer asks, "It is interesting that Sun has had Ian Murdock, the co-creator of the Debian distribution of Linux, on the payroll for two years now and Sun has not figured out that it needs to have a Linux distro for all of the Niagara boxes"
SUN figured out that they need a new packaging system, that is different from System VR4 UNIX packaging - this was one of the primary goals of bringing on Ian Murdock... to figure out how to move Solaris to Debian packaging.
Ubuntu has a very strong and growing popularity base, so moving Solaris to have the same packaging as Ubuntu and other Debian packaging means that Solaris, Ubuntu, and other Debian based systems could grow together.
Since SUN sells Intel and AMD systems with Linux, the need for Linux on the SUN T platforms is not as critical. If it was, other Linux vendors would be falling all over themselves for the revenue, coming from a growing T base.
No common Hardware model for Sun systems?
What is it about the T1, T2 and Niagara that make porting Linux so hard? Do they not run Sparc code from older systems? Or is it that the memory model or I/O subsystems of the resultant systems do not match what has gone before.
I admit that I am not following Sparc developments that closely, because I work mainly on IBM systems, but as far as I am aware, and as a comparison, there is no real need to port the Power based Linux distributions onto Power6, the current ones just work. Of course, there is a quite high degree of commonality between Power5 and Power6 systems (and to older systems as well, as chrp is a a common model), and you have all of the built-in virtualization which may help to isolate the OS from the hardware. And you probably won't get the full benefit of new features, but it works.
Have SUN not got that degree of abstraction from the hardware?
Case in point
I have actually tried Ubuntu on a T1 machine and it served as a stellar illustration how poorly Linux scales over a comparatively large number of cores/threads compared with Solaris. On a 8core/32thread T1000 box, I got almost an *order of the magnitude* better performance running my Tomcat based benchmark using Solaris as opposed to Linux. Perhaps this is why Ubuntu never really caught on - Linux performance really sucks on that sort of hardware.
Re: No common Hardware model for Sun systems?
It's not a bad question really. Sun has tended to have a very solid record of consistency. In the case of The T1/T2 and Rock when it comes out, they changed to more of an abstracted level like IBM. The USIII/IV/Etc. and SPARC64 systems all show a sun4u interface to the OS, while the newer interface is sun4v with it's "built in virtualization". The future will be sun4v for Sun.
I'm not sure where you can say that Linux "just works" on Power though. There was much work to be done to get it to work on the Power VM. Once it was there, it was relatively easy to move to newer versions of Power. I expect the same to be true with the sun4v variants of Linux...
Re: No common Hardware model & Case in point
Sun has been fairly consistent at their hardware. Especially with backwards compatibility. In most cases, they back port OS patches for older Solaris releases with newer HW support. The ability to also run something compiled on SunOS 2.5, and have it run on Solaris 10 is a huge plus. Granted, it will run better if it's recompiled, but that isn't always an option.
Sun's x86 servers are certified to run Windows, Linux, and VMWare, as well as Solaris x86. Although the entry level Niagara boxes are cheaper then the midrange & high end Sun servers, the people that buy them are buying them for stability and to push data. Solaris will continue to win this on Sparc for a few years. The Niagara boxes do a damn good job at what they're meant to do -- push a lot of data on a single power efficient box.
I think the writer of "Case in Point" has hit the nail on the head. I love Linux, but it just can't scale across cpus and handle multiple processes like Solaris. Solaris has been doing this very well for 10+ years. Linux is just now starting to handle it with the 2.6 kernel. This is why Intel, AMD, and IBM partner with Sun. And also one of the reasons IBM may buy Sun.
Re: Correction and Answer -- Yes, the T1 can be bought with 8 cores, but the article implies Ubuntu was only supported on the 4 core version. I suspect it's because Linux doesn't scale well.
There have been a few benchmarks comparing Window vs Linux vs Solaris. On the larger boxes, Linux kills windows, but Solaris kills Linux. Linux will continue to do better, Windows will continue to suck, and perhaps IBM will buy Sun, advertise, and Solaris will grow outside of the datacenter where it works so well.
While not "commercial" gentoo works on the T2 and VF series, yes even on the quad socket 5440. It was disappointing to find Ubuntu had pulled support but I had a need to check out some Linux code on the 5440. The same image works on the 5220 and 5240 which has been very helpful in other debug and development work.
To "Case in point", Ubuntu was never a high performance distro, if I have a chance I'll run some benchmarks on Gentoo.
more interesting is the fact that sun4v firmware now looks to be based on a linux distro already and has been for a number of releases, so these guys who think that they are not running linux in their datacentre should look again !!!!
It used be a vxworks base but that costs $$$ and linux is eeerrr free ???
Re: No common Hardware model for Sun systems?
The simple answer would be "It's the applications, stupid!" The majority of Linux apps have been written for thick-threaded x64 CPUs, even recompiled they won't run as well on Niagara as they won't take advantage of all those cores and threads. Instead, the current Linux apps will try pushing those thick threads through those thin cores, a bit like trying to force custard through a sieve. Unless the code is completely rewritten to take advantage of all those wheiner cores - which means someone has to pay developers to redesign, rewrite and optimise the code - the performance will look pants compared to bog standard (and cheaper) x64, and the average buyer will wonder why they even bothered looking.
Sun's problem here is the same chicken-and-the-egg situation that faces any CPU vendor - getting those companies that make the popular core apps like (like Oracle) to stump up the cash to rewrite their apps. Unless Sun throws cash at them to encourage them to port their apps, they may hesitate to do so when there is so much uncertainty around the future of the SPARC range. And Sun throwing money at Linux isn't likely in the current environment, even if they could get round their own Slowaris bigotry. In areas such as HPTC, where they usually handcraft their own specialised apps, the idea of using Linux on Niagara will have appeal as they don't need to wait for a software vendor to do the rewrite. However, HPTC is also the realm of distributed apps on big x64 Linux grids, and I'm not convinced there is enough of a market to sustain a Linux on Niagara effort for long. And then there is the pain for Sun of knowing that any such success means the loss of Slowaris revenue as there is unlikely to be any support revenue from such projects.
In prediction of those Sunshiners that will immediately scream about how Niagara offers such unique advantages that the Linux community will kill each other stampeding to the Niagara-Linux bandwagon, people said the same about Linux on UltraSPARC, Linux on Power, Linux on Alpha, Linux on PA-RISC (yes, the latter did exist!), <insert Linux on your prefered flavour of CPU here>, etc, etc, etc.
Re: Re: No common Hardware model for Sun systems?
While I agree that a lot of Linux applications are written by relative novices that do not understand multithreading, making their apps not run as well in multithreaded environments, I have to disagree that this will be that big of a problem here. The LDOM's feature is probably much more interesting here. You can put 256 low end solutions on one Telecom ready box... The VM solutions on X64 are just not telecom ready yet. I'm not sure that the VM vendors will go there for a while and even if they do, will the Telecom companies trust it for a while? The cost savings will be HUGE on Linux on Niagara when compared to X64.
Telecom already knows, loves and trusts Niagara.
RE: Re: Re: No common Hardware model for Sun systems?
"While I agree that a lot of Linux applications are written by relative novices that do not understand multithreading..." Wrong target - for acceptance in a commercial role the Niagara Linux systems will have to run commercial apps, not hobbyist ones. People like Oracle and SAP know plenty about multithreading, it's whether they see a financial gain large enough to offset the cost of porting, testing and supporting their product on a new flavour of Linux. Linux on Niagara may be viable in niche solutions, but not at the volumes that would make it anything more than another BeOS.
"....The LDOM's feature is probably much more interesting here. You can put 256 low end solutions on one Telecom ready box...." <Yawn> OK, so we're back to wheiner threads again. If they aren't man enough to do the job with Slowaris (whch already has some application base), why is it better to have wheiner threads in Linux on Niagara?
"....The VM solutions on X64 are just not telecom ready yet....." I'm not sure which area you may be looking at, but I know for a fact that major telcos such as Vodafone, T-Mobile and BT use VMware (I've seen the case studies). If you are talking about very explicit situations like the actual cell switch servers, last time I looked it was written by professional companies mainly using commercial UNIX. I can't see them jumping to Linux on Niagara unless it offered some specific advantage over OSs such as Slowaris itself.
"....The cost savings will be HUGE on Linux on Niagara when compared to X64....." If that was true, then x64 wouldn't be eating up the UNIX business from the bottom up, and Niagara is the very bottom of the UNIX bizz.
"....Telecom already knows, loves and trusts Niagara." Maybe, but will IBM love it enough to put any money into it? Doubtful.
Re: RE: Re: Re: No common Hardware model for Sun systems?
"Wrong target - for acceptance in a commercial role the Niagara Linux systems will have to run commercial apps, not hobbyist ones."
But we're talking Telecom here Matt. Telecom seems more willing to find the best solution vs the biggest vendor to do their apps. Many times the best fit is a small company that will compile their application to whatever box you want. Also, many Telecom apps are homegrown. Finally, in Telecom, you are talking thousands, if not millions, of users. Niagara is definitely the sweet spot for this. It's not how fast they run, it's the number of users they can handle. For the cost, nothing comes even close to Niagara on number of threads (read concurrent users).
"OK, so we're back to wheiner threads again."
It doesn't matter if they're wiener threads. Most or many of the applications that need to go here are low end applications (well, not exactly low end, but they need throughput, not speed). The many thread solution for Telecom is very important, but being able to do it on a small solution that you can throw into a brick non-datacenter building somewhere is also important.
"I can't see them jumping to Linux on Niagara unless it offered some specific advantage over OSs such as Slowaris itself." I agree, but they are doing it... Telecoms tend to be first to bite on a lot of this stuff for some reason and I don't see Sun doing this unless they had customers requesting it.
"If that was true, then x64 wouldn't be eating up the UNIX business from the bottom up, and Niagara is the very bottom of the UNIX bizz."
I'm not sure what "very bottom of the UNIX bizz" means, but again, we're talking Telecom here. This is not general purpose computing. They need Telecom certified gear, and that is what the Netra/Niagara is.
"Maybe, but will IBM love it enough to put any money into it? Doubtful."
Sadly, I agree. You never know though. It's a better than 1B dollar business with nice margins and growing faster than most platforms. IBM likes it's disparate platforms that it can charge huge sums of money to integrate. I just don't see them wanting to keep two UNIX around... Again... you never know...
- Product round-up Six of the best gaming keyboard and mouse combos
- China building SUPERSONIC SUBMARINE that travels in a BUBBLE
- Boffins attempt to prove the UNIVERSE IS JUST A HOLOGRAM
- Review Raspberry Pi B+: PHWOAR, get a load of those pins
- 6 Obvious Reasons Why Facebook Will Ban This Article (Thank God)