“Nobody can deny that Sun has the technology and expertise to continue doing good things. But do they have the right direction? IT consultant Paul Murphy says what must be done if Sun is to correct weaknesses eroding its position. Here’s his analysis on where Solaris and SPARC are headed, and the kind of management focus that can place Sun on higher ground.” Read the article at Open Magzine.
Sun Swat Redux: A Strategic Roadmap
About The Author
Eugenia Loli
Ex-programmer, ex-editor in chief at OSNews.com, now a visual artist/filmmaker.
Follow me on Twitter @EugeniaLoli
“Those who want to believe will avoid thinking about how specious and self-serving the comparisons are.”
In the article he cites a TCO comparison of MySQL on a Dell system to Oracle on a Sun. I think this extends to many other comparisons of commercial to open source software that I see being made often, such as comparisons of the Gimp to Photoshop, or Linux’s SMP or I/O performance to Solaris’s. In the case of these comparisons it is clear from the outset that that the two things being compared aren’t in the same league with each other.
However, the article took some odd turns towards the end:
“Apple has a real problem right now. Its relationship with Motorola has been under pressure for years and they seem about to drift into IBM’s embrace with a deal on using the Power4 CPU in the next generation Macs.”
This is an excellent path for Apple to take because it requires absolutely no non-volitional changes on the part of the developer or end user.
The Power4 is nice machine, but the deal will give IBM far more control than it should have and position it for a quiet take-over of Apple’s executive suite. Sun does not need to buy Apple to head that off. Instead, all it has to do is bite the bullet and do whatever it takes to get Apple as a partner on Solaris/SPARC. The MacOS X shell already runs on SPARC and Solaris can as easily run a PowerPC board inside the box for full backward compatibility.
I’m not sure specifically what he’s suggesting here, but this deviates far from anything Apple would ever intend to do. Switching the underlying operating system from Darwin to Solaris simply would not work, for the following reasons alone, there are many more:
* OS X already has a large and continually increasing base of KEXTs which provide support for 3rd party hardware for which Solaris has no support for as the target market of such hardware lies far beyond Solaris’s key demographic (i.e. drawing tablets) Apple has stated many times their unwillingness to implement any features in the XNU kernel which would break compatibility with 3rd party KEXTs, such as poll(). Due to XNU’s modular driver architecture. Porting this driver architecture to Solaris would require a considerable abstraction layer, and it would be nearly impossible to make such an abstraction layer 100% compatible with existing KEXTs. If he’s actually suggesting Apple move to Solaris/SPARC, I think the compatibility issues regarding the disparate underlying ISAs and hardware architectures would be overwhelming. Bottom line, this issue alone is not possible for Sun to contend with, and it alone is a dealbreaker for Apple.
* All (native or carbonized) OS X applications use the Mach-O ABI, and are of course compiled for the PPC ISA. I don’t know if the author of the article wants Apple or Sun to write (yet another) emulator to run Mach-O PPC binaries on Solaris/SPARC or what…
* Process synchronization between the OS X Window Server and applications occurs using Mach message queues, which Solaris obviously doesn’t have. Of course, if Apple or Sun is already writing an emulator to execute Mach-O PPC binaries on SPARC, this would obviously be considered and could be implemented using any other IPC channel the implemenor pleased.
Bottom line: Apple is still trying to transition users from OS 9 to OS X. They aren’t going to be switching ISAs or hardware platforms any time soon. Furthermore, Apple’s market niche seems to be shrinking, and they fail to attract new users due to their already expensive hardware prices. Using SPARC in place of PPC would only add to the cost. Furthemore, hardware support doesn’t exist in Solaris for the majority of the hardware Apple is currently using. Finally, Steve Jobs is CEO and I don’t think selling out to Sun is part of his “creative vision”
“Strategically a deal like…adds Motorola to the SPARC architecture team”
Huh? How would Motorola even be involved? If Apple chose to support a different architecture besides PPC, I’m sure Motorola would care less, and let them go their merry way…
and gives both Apple and Sun a wonderfully coherent desktop-to-server story to tell their customers.
I would love nothing more than to see Quartz/Aqua/CoreServices/QuickTime/etc. ported to Solaris, and that to deployed in place of the existing OS X. The resulting OS would arguably be the best designed GUI platform in existance. However, this entire setup has less of a chance of happening than Apple porting OS X to Intel.
“The current Linux adventure is a mistake.”
That statement couldn’t be more correct. Who’s going to buy Linux servers/desktops from Sun, as opposed to the dozens of other companies offering high reliability Linux servers for less…
This was certainly a strange article. I’ve never seen so many good ideas intersperced with such wild and poorly thought out ones. I’d have to say though that overall it made a very interesting read.
>”The current Linux adventure is a mistake.”
>That statement couldn’t be more correct. Who’s going to buy Linux servers/desktops from Sun, as opposed to the dozens of other companies offering high reliability Linux servers for less…
Not so.
The Sun Linux machines are thin clients and simple PCs. Sun, AFAIK, are not going to sell servers based on Linux (except the LX50 which is just an entry level system and Sun doesn’t even try to push it – it’s a cheap solution to fill the product range). Here is their line of thought, as it was explained to me the last time I was eating chicken in skewers with Sun executives (about 2 months ago):
Right now, Sun has a big number of partners and clients, and these companies want to upgrade their desktops. Now, Sun tells them “get this cheap thin client box, which runs Linux, instead of going with Windows or our expensive-ish SPARC workstation boxes”. In the meantime, Sun partners with Red Hat and Red Hat develops this “corporate desktop”, a move which started with Red Hat Linux 8.0 and a new version is anticipated for version 8.1 on April. The version 8.1 will be on these boxes, which are expected to ship this summer. So, these “cheap” Linux corporate desktops are going to replace Win95 or old SPARC workstations mostly for **existing** Sun customers. This was a move by Sun in order to be able to keep their existing customers regarding their workstations. And they have contracted Red Hat about it.
So, as you can understand, Sun is not interested in Linux the same way HP or IBM or SGI are. They only need and want Linux for the *corporate desktop* in order to keep their current customers. And with this line of thought, it does make sense to do so, as SPARC boxes are more expensive than PCs, and Solaris is not focused on the secretery’s desktop. Sun’s offer on Linux thin clients (which interoperate with Solaris very well) are just an addition to the whole SOLUTION Sun is offering to very big companies. These companies prefer to have the whole IT solution given by a single company (e.g. Sun or MS or IBM), so Sun tries to fill the gap of the “cheap corporate PC” in their product line. It makes perfect sense if you think about it. Red Hat does all the work anyway and Sun is “branding” it afterwards (they have contracts and partnership the two companies).
I hope now this is more clear.
The author says that the shell of Mac OS X runs on Sparcs already. I assume he means darwin, the BSD Unix base of Mac OS X. He is wrong. I have never heard of Darwin running on anything other than PowerPC and Intel chips. Now if he means NeXTStep, then he’s correct but NeXT is a long way from Mac OS X.
… all it (Sun) has to do is bite the bullet and do whatever it takes to get Apple as a partner on Solaris/SPARC. The MacOS X shell already runs on SPARC and Solaris can as easily run a PowerPC board inside the box for full backward compatibility. Strategically a deal like this leaves IBM dependent on Microsoft and Intel for the desktop, adds Motorola to the SPARC architecture team, and gives both Apple and Sun a wonderfully coherent desktop-to-server story to tell their customers.
OpenStep was already ported to the SPARC architecture long ago, so ideally it would be an incremental leap to port OSX to the current SPARC (I bet Sun would even tell themm all the details on the UltraSPARC III :-))
The goal, I think, would be to use the SPARC chip as the New CPU for the Apple products, though not necessarily Solaris.
See, somebody like Apple woult move forward to take advantage of the SPARC chip, and scream to the world about it (like the Altivec on the G4 today). 64-BIT POWER! Uber Floating Point! Really Cool Logos!
And what that gives the market is two manufactures on the SPARC architecture. Apple can be in the research and development area for those wishing to leverage the architecture, and Sun on the larger end with servers and what not for those wishing to leverage the operating system.
Whether Apple could pull off selling machines with two cpus (SPARC and PPC) and intergrating that into the system would be a bit of a trick I think.
But folks have been yelling at Apple to switch CPUs, and think that Intel is the only way to go (actually, they want OSX on commodity PCs, but that ain’t gonna happen).
Many feel that the SPARC chip is a purty slick chip, and it’s being developed by a company that’s actually interested in it, so it might be a better fit for Apple.
Basically, the hit on Apple is the same whether they go Intel or something else that’s not binary compatible with the PPC. So, why go with the Intel architecture which is best known for its ubiquity rather than its performance?
Back In The Day, when OpenStep was running on 68k, SPARC, HPPA, and Intel, the source code was very portable across the platforms. This is why Apple is really pushing OS X so hard I think, as it will make it that much easier for applications once ported to OSX to be ported to another platform.
That’s why Apple is pushing on things like Java as well. Imagine what wonders Apple could come up with if it could mix and match processors on a whim. SPARC here, PPC there, StrongARM thataway. Neeto.
But folks have been yelling at Apple to switch CPUs,
Folks have been yelling at Apple because they want OS X but feel the price/performance ratio of a Mac is too low. Switching to SPARC definitely will not help this situation. Existing notebooks based off the UltraSPARC IIi run upwards of $5000 (such as the Tadpole SPARCbook 6500) and would not provide acceptable performance for users who are already complaining about Apple system performance. To my knowledge no one has made an UltraSPARC III based laptop, and no one in their right mind would buy one unless they have considerable need for a portable high performance SPARC system.
and think that Intel is the only way to go
Read my last post. Apple does NOT want to break backwards compatibility with any software released for OS X at this point. Their main focus right now is getting OS 9 users to transition to OS X. There is *ABSOLUTELY NO CHANCE* of them switching ISAs any time in the near future. They will stick with the PPC ISA for some time to come.
Whether Apple could pull off selling machines with two cpus (SPARC and PPC) and intergrating that into the system would be a bit of a trick I think.
This would require:
* Apple ports CoreFoundation/CoreServices/Quartz/Cocoa/Carbon/Aqua/QuickTime to Solaris.
* Hardware companies provide two drivers, one for XNU/PPC and one for Solaris/SPARC
* Software vendors supply two different versions of every program, one for Mach-O/PPC, and one for Solaris/SPARC. The only other possibility is adding support for the Mach-O ABI to Solaris, at which point the OS X developer tools can automatically cross-compile all programs to both platforms, and provide fat Mach-O binaries. However, this still means that all existing OS X applications will not run on the new platform.
* Apple provides some sort of incentive to use SPARC based Macs over PPC based ones, most likely by ceasing development of new PPC based Macs and only continuing to sell their current models.
Apple in its current form would never agree to the above. For it to happen, it would require a buyout of Apple by Sun, which in turn require a liquidation of a large portion of Sun’s assets, and a large amount of internal restructuring. The UltraSPARC III in its current form would probably be unacceptable due to its high manufacturing costs. A “lite” version would have to be released, probably with stripped cache and some of the execution units cut. Stripping the processor down should also focus on removing or redesigning any components which are hindering its clock speed. Sun would also need to shift to a newer manufacturing process, which would most likely require them partnering with some other company to produce their chips.
All of this said, this proposition is insane. This would not increase Apple’s business. It would not improve the price performance situation; if anything, it would be worsened by the pricey SPARC processors. It would not improve business for Apple, but would probably drive a number of existing Mac users back to Wintel.
Bottom line: Apple will stick to the PPC ISA. Don’t expect this fact to change for a long, long time.
My favorite part about all of this speculation is that anyone would still consider SPARC competitive. If Apple were to migrate from PPC, then I would hope that they would choose a new architecture with some additional benefits. SPARC also provides poor performance and more limited software support. Instead Apple should be considering the PPC for compatibility and product stability, x86-64 for performance and (hopefully) future cost benefits, and possibly ARM for great integration and power requirements.
Regarding Sun, I think that they will be forced to emulate SGI and start merging Solaris’ technologies into Linux. Not too many left, though, that Linux has not caught up with or surpassed; which just leaves stability.
>>>>The Power4 is nice machine, but the deal will give IBM far more control than it should have and position it for a quiet take-over of Apple’s executive suite.
The reason why Motorola and IBM are so slow on developing gigahtz powerpc cpu is that they make a lot more money on powerpc in the embedded and telecom industries. Cisco is a bigger powerpc customer than apple, that’s why all the IBM/Motorla development dollars go into embedded powerpc chips.
IBM doesn’t want control of apple- their powerpc development group doesn’t care about apple.
I think that any SUN/Apple merger would be a very LONG TERM type deal.
I would imagine it taking upwards of 5-10 years befor any type of integration that is being talked about here would even start.
The advantage that i see is that you have apple on the desktop, and sun on the server. Then you would work from there to merger things like your CPU and overall arcitecture.
and lets think the other way around about SPARC. what if sun decided to work on PowerPC? just a thought
Sun has a big number of partners and clients, and these companies want to upgrade their desktops. Now, Sun tells them “get this cheap thin client box, which runs Linux, instead of going with Windows or our expensive-ish SPARC workstation boxes”.
Actually no. Sun has its own hardware based thin clients. What they want want Linux for it thick clients that can act independently like their thin clients. In other words if the office runs off a StarOffice, Gnome, workplace setup and you want people working from home over a modem you can’t have them running X remotely; at least not for anything but the most exceptional circumstances. So instead you give them a Linux desktop system with a corporate configuration and the Linux box just sincs the data on an as needed basis without the network overhead of genuine thin client computing. The Java machine is running local, star office is running local, the desktop is running local… but all the data is passing back to the Sun servers. That way the desktop feels to the worker exactly like what they would get from Sun’s thin client environment.
The article seemed too advanced for me.
But i did get excited over the Apple/Sun deal….however wouldn’t this be the same situation as before. I know i wouldn’t buy an apple mac, cos its too expensive. Who cares about aqua. Surely Sun know what they’re doing.
I honestly don’t think Sun need to worry too much, everybody seems to wanna be in fashion and say yeah “me know Linux”.
In my opinion Linux makes far more sense as a philosophy.
Some random anonymous COWARD states:
My favorite part about all of this speculation is that anyone would still consider SPARC competitive. If Apple were to migrate from PPC, then I would hope that they would choose a new architecture with some additional benefits. SPARC also provides poor performance and more limited software support. Instead Apple should be considering the PPC for compatibility and product stability, x86-64 for performance and (hopefully) future cost benefits, and possibly ARM for great integration and power requirements.
Regarding Sun, I think that they will be forced to emulate SGI and start merging Solaris’ technologies into Linux. Not too many left, though, that Linux has not caught up with or surpassed; which just leaves stability.
Okay, let’s get some things straight… x86 is definitely not outperforming SPARC yet.
My traditional example is always the atmospheric modelling program we run at work. This is an extremely CPU bound program which is run in parallel via MPICH (over 100Mbps Ethernet, gigabit has no effect on performance) The timestep with a 4 * dual 1GHz PIII cluster was higher (indicating worse performance) than on our dual 900MHz Blade 1000s, using shared memory MPI. Considering the highly parallelizable nature of the model and its veritable independence from I/O restrictions, I think it’s safe to say that 2 900MHz UltraSPARC IIIs are beating 4 * 1GHz PIIIs. Considering that clock-for-clock a PIII is faster than a P4, this would place a 900MHz UltraSPARC III well above any x86 solution (i.e. well into the 4GHz P4 range)
As far as Linux being in the same league as Solaris, you are validating a claim the author made which I cited earlier:
“Those who want to believe will avoid thinking about how specious and self-serving the comparisons are.”
Obviously Linux is nowhere in the same league as Solaris, both in terms of features and in terms of overall design. Does Linux have a non-executable user stack? How well does it parallelize… especially above 64 processors? Does it have zero copy NFS? How’s the kernel’s I/O performance? I mean, jesus christ, I could go on and on…
The reason why Motorola and IBM are so slow on developing gigahtz powerpc cpu is that they make a lot more money on powerpc in the embedded and telecom industries.
Why are you listing IBM in this description? IBM’s interest in the PPC ISA comes from their work on the POWER architecture. They’ve made some lower end CPUs in the past,
such as the G3 used in Apple’s iBooks, and various other processors they’ve sold to Apple and other companies such as Nintendo. But IBM certainly isn’t making embedded/telcom oriented CPUs… that’s Motorola’s market.
Cisco is a bigger powerpc customer than apple, that’s why all the IBM/Motorla development dollars go into embedded powerpc chips.
Once again… IBM’s money is going into processors for their mainframe class servers. They’re also developing another processor for which Apple has been cited as a customer (the GPUL)
>>>But IBM certainly isn’t making embedded/telcom oriented CPUs… that’s Motorola’s market.
Then explained why the guy in charge of IBM’s powerpc development is Chekib Akrout — IBM’s vice president of PowerPC and Networking Development.
http://www.eetimes.com/sys/news/OEG20021112S0021
Of course POWER4 is still for big servers, but its simplified cousin (powerpc and GPUL) is for NETWORKING (primary purpose) and desktop pc (secondary purpose).
First, I’m sorry that my anonymity put you off.
I’m glad that you are getting better performance on your modelling application with the Sun Blade. But I don’t accept your extrapolation. Today’s Xeon or Althon systems have memory subsystems that bring them to the level of today’s UltraSPARC, and both surpass SPARC for peak performance, integer and fp. I would guess that the pc’s are level if not slightly ahead now in terms of sustained throughput. So it would be interesting to see a Dell’s latest Precision workstation, say, against a Sun Blade 2000 or comparable Sun server.
I also agree with you that Solaris has a technological edge over Linux today.
But we are looking to the future. And the “low-end” is marching upward more quickly than I think you are willing to accept.
SPARC has been slipping in high performance computing for some time. Compared to its HPC competition – Alpha, Power4, or Itanium2 – Sun’s systems are not competing on performance[1]. With x86-64 we should start to see 64-bit systems at commodity prices. Do you really think that SPARC looks to be competitive for much longer? If so, why? I haven’t paid much attention to Sun’s announcements on SPARC’s future for the last couple years. Are we expecting a big jump or breakthrough technology soon?
In the past I haven’t liked Linux very much because it has been very unstable. But its fantastic pace of development cannot be denied. There is work being done on a non-executable stack. SGI just announced a new system, Altix[2], that runs Linux on Itanium and scales well to 64 processors. I won’t be surprised if the next revision goes up to hundreds of CPUs. Significant Linux NFS development is ongoing at the University of Michigan [3]. More management funcationality is also coming down the development pipeline. Do you believe that Solaris will keep its technological edge and for how long?
I believe that Solaris’ greatest strength today and for some years to come will be its great stability and maturity. There is enough good software that will keep Solaris and SPARC around for a good long time as a legacy system. But I don’t see Sun keeping up with Linux’s pace of development, and as Linux matures Sun’s advantages will just decrease. Sun’s challenge (and HP’s, IBM’s, SGI’s) with Linux will be to harness Linux’ rapid pace of change while not losing the stability that its customers rely on and pay for.
I am sure that all of this progress will be interesting.
[1] Look at the SPEC submissions for 2002Q4 – http://www.spec.org/cpu2000/results/res2002q4/ – Sun just isn’t in the running anymore.
[2] http://www.sgi.com/servers/altix/index.html
[3] http://www.citi.umich.edu/projects/nfsv4/
First off, Sun has never been about making really fast processors. It’s been about making highly scalable systems with a high I/O bandwidth. The UltraSparc 3 line is a huge improvement to the Ultra Sparc line. Just look at the Sun Fire interconnect technology, something you won’t see in PC’s for a long time. Now SGI has been able to do some interesting things wth Linux, and it’s own hardware, but you won’t see that same technology on a Dell PC. While Linux is being developed rapidly, it still has about 10 years worth of maturing from a kernel perspective. There are tons of features in the Solaris kenrel, that are either not possible or years off from being remotely stable in Linux. Sun will be around for a long time, just as IBM, HP, and the rest will be.
Now for some interesting food for thought. Sun’s next big thing in the Ultra Sparc world will be the Ultra Sparc V, which will be an async processor. Which means that clock speeds will be a thing of the past. Now put that in your pipe and smoke it:)
You choose Sun Microsystems you pay for best quality, perfomance and stability!
Sorry people, but Sun Microsystems technology at now is the best.
If Apple were to consider switching its ISA, it would do much better to consider buying/licensing Alpha from HPC/Intel than to team up with Sun on a SPARC platform.
HPC already has very good processor interconnect technology (witness: Wildfire,) though that really isn’t Apple’s cup of tea. Should they decide to push into the server market, they could use that technology, but their niche is, and will continue to be for the foreseeable future, the desktop.
Using the Alpha would bring them stupendous FP and INT performance, a robust ISA, good bus technology, and Tru64’s Mach roots serve to illustrate that a Mach-based kernel will work very well on that platform, indeed. Teaming with Samsung/API/HPC/Intel/whomever to build a line of relatively inexpensive Alpha CPUs for their machines would put them ahead of the game in every way, should they decide to switch ISA.
Which they won’t.
The comments here are among the most cogent and articulate I’ve seen – and that includes email from Sun insiders- however there are several issues I’d like to clear up:
1- I am not suggesting that Sun buy Apple or Gateway; that would weaken everybody.
I am suggesting that Sun, Apple, and Gateway have complementary strengths that can best be brought to the market if they work together.
There’s an implicit issue here not mentioned in the article: not only is Altivec what SPARC needs, but, more importantly, I believe that the <EM>personal</em> computer is being replaced by handhelds combining phone and PDA functions. Thus bringing Motorola in as the original PPC developer will add its manufacturing and microelectronics strengths to SPARC while positioning all four companies to lead this change.
2- Yes, I am suggesting that MacOS X run on Solaris. I believe that it already does although not, of course, with the full set of device drivers needed. Bear in mind two things:
2.1 that MacOS X is made up of three main layers: a mostly Mach kernel; a mostly BSD Unix; and the MacOS X GUI. Of course the GUI includes code designed specifically for the PPC/Mach combination; but replacing those elements isn’t a big deal given the skills available in both companies; and,
2.2 that it is the device which has to meet the specification; not the other way around. When you try to engineer backwards compatibility at the hardware level this gets reversed because you can’t retroactively change the spec and the devices that meet it. But the Solaris answer to this is to use a PPC co-processor board for backwards compatibility and open standards for forward compatibility. That adds about $200 to the cost of the machine, but only for those who need it -and most big customers don’t