| [an error occurred while processing this directive] |  Linux International Member |
| ★ wanayoo — archive 1999 http://linuxtoday.com/stories/15853_flat.html | Nouvelle recherche | Portail wanayoo |
| [an error occurred while processing this directive] | Linux International Member |
Linux Today Feature: IBM on 'Open Source' Solaris: Get Real!
Jan 29, 2000, 02:27 UTC
by John Matthews, Linux Today Silicon Valley correspondent
From a phone conversation with Miles Barel, IBM's program director for Unix marketing
SAN JOSE, California, January 28, 2000 - Several months after the first stories surfaced that Sun was planning to 'open up' Solaris 8, the details were finally announced this past Wednesday. Those details began leaking a week earlier, so reaction to Sun's move has been coming in steadily over the past week. There seems to be a fairly broad consensus that Sun is attempting to head off erosion of its server market share, from Windows 2000 at the high end and from Linux on the low end (CNET 1/00, ZDNet 1/00). Opinion as to the impact on Linux is divided, with Bruce Perens believing, "This will be a tremendous shot in the arm for Linux", and IBM calling it "an effort to distract the Linux community" (both quotes from Red Herring 1/00).
IBM is a Sun competitor that's going after Sun's market--high end, low end, and middle--with its RS6000 product line (Linux Today 9/99) and recently expanded support for Linux across its entire server product line (CNET 1/00, PC Week 1/00, Red Herring 1/00). We asked Miles Barel, IBM's program director for Unix marketing, to comment on the Sun's new licensing and pricing strategy for Solaris 8.
Sun's agenda is anything but open
As many others have noted since Sun's announcement (see Related stories below), Barel thinks Sun's Solaris 8 program is anything but open. "My guess is that if you order it [the source code], you won't get the whole system--you wouldn't be able to compile your own working operating system", said Barel.
Why?--the third-party code in Solaris that Sun has licensed but does not own. Sun may have gotten around this by negotiating agreements with the owners of those pieces of code, and that may be what has shaped the licensing terms that led Sun to avoid the use of the term 'open' in their big announcement (Red Herring 1/00)--the $75 'media charge' and the requirements that (a) changes made in the source must maintain compatibility with the 'standard' Solaris version and (b) can only be redistributed via an agreement with Sun that 'may involve royalties' (SRO 1/00).
In Barel's view, "open source is about the industry driving the development of industry standards". Elaborating on 'the industry', Barel indicated that he was using the term much the way that open source people use the term 'the open source community', or 'the Linux community', in the broadest sense: In addition to the open source developers, he's including users, commercial vendors, and anyone else who wants to actively participate.
What Sun is doing is more like riding on the popularity of the open source concept in order to promote what essentially remains its own proprietary, tightly controlled standard. Barel stressed that the Solaris source code is 'free', neither in the sense of 'free beer' nor in the sense of 'free code'.
IBM considered open sourcing AIX
Barel said that IBM had given serious consideration to the idea of open sourcing its proprietary Unix code, AIX. "IBM looked at all the trade-offs involved in making AIX code available and concluded that it would just confuse the market." Barel believes this is exactly what will happen as a result of Sun's opening up Solaris. He even goes so far as suspecting that this may be precisely Sun's objective, in hopes of slowing the growth of Linux at Sun's expense. The 'Lintel value proposition'--the combination of the license-free Linux OS with relatively cheap Intel PCs--is generally considered to be as much of a threat to Sun's low-end hardware market as it is to Microsoft's OS sales.
Instead IBM chose to give AIX code to Linux
IBM concluded that a far better strategy than open sourcing AIX would be to contribute key parts of the AIX code to the Linux development team (PC Week 1/00, VNUnet 1/00). Which parts? "We'll work with the Linux kernel development team to determine exactly which parts they can really use," said Barel. IBM will leave it to the Linux kernel team to make the announcements about which parts of IBM's offer they decide to accept and their timeframe for implementation. "We don't want to raise false expectations" by giving out more details now.
Where will IBM draw the line?
Assuming that IBM were offering the Linux kernel team anything they wanted from the AIX code (which isn't exactly what anyone at IBM has said)--wouldn't that let Linux leapfrog ahead to parity with AIX? Not in Barel's opinion. "The maturity of an operating system [like AIX] takes many years to develop." The Linux team wants to continue to develop Linux, not turn it into 'open AIX'. By allowing the Linux team to borrow pieces of AIX code that they can easily integrate into Linux, and by supporting the Trillian project to port Linux to Intel's 64-bit chips, IBM can promote the development of a more robust Linux that they can support on their server line. Linux development will be greatly accelerated, but meanwhile AIX development will continue as well.
Sun, Linux, and IBM
Barel observes that "Sun has consistently said over the past few years that they're only interested in one thing, and that's Solaris on Sparc."
According to Barel, IBM long ago learned that 'one size fits all' doesn't work, and sees that as even more true today. To IBM, 'e-business' is a huge collection of components running on different platforms. IBM's big strength, the major value that it offers its customers in the internet age, is their skills at "putting all the pieces together and making it work".
IBM is firmly committed to Linux as a cornerstone of their internet and e-business strategy, reiterated Barel. "We believe that Linux is the future of the internet. The internet has been based on open technology. Linux is one of the core technologies that will drive the future evolution of the 'net." From IBM's perspective, "It's in everyone's interest to promote Linux."
BTW: AIX had the 'new' Solaris features 2-3 years ago
While we were talking about the Linux aspects of the Solaris 8 announcement, Barel was quick to point out that most of the new features appearing in Solaris 8 had been added to AIX two or three years ago--for example:
Related stories--IBM supporting Linux, challenging Sun:
VNU NET: IBM making good progress on open source plan (Jan 26, 2000)
CNET News.com: IBM exec touts Linux as key to Net evolution (Jan 26, 2000)
Red Herring: IBM wants to kick Sun butt (Jan 25, 2000)
SRO: IBM Steals Sun's Java Thunder (Jan 24, 2000)
ComputerWorld: IBM has big plans for Linux (Jan 12, 2000)
CNET News.com: IBM boosts Linux on servers (Jan 10, 2000)
PC Week: IBM plans major expansion of Linux efforts (Jan 10, 2000)
IBM Leverages Linux in Sun Challenge (Sep 14, 1999)
| Brian Strand - Subject: hot upgrading of os!? (2000-01-29 03:13:00) |
| I've been thinking about this for a while. How do they do that, exactly? Does the new kernel start life out as a user process and then "inherit" the data structures of the old bit by bit? This sounds like an interesting feature to implement. If it is true that you can hot-upgrade the kernel without any user processes noticing, that is amazing... |
| Eduardo Ahumada - Subject: (Microsoft & Solaris) | (Linux & IBM) (2000-01-29 05:05:36) |
| =( .......................................................... |
| Lonnie Nunweiler - Subject: Sun has to do more than talk (2000-01-29 06:42:01) |
| Sun announced that Star Office was under their Community Licence in August 99, but we have yet to see anything except the claim it will be so. Now Solaris is to be the same. To modify a somewhat famous question "Where's the CODE?" Just what are they waiting for? Heck, if Microsoft was watching closely, they could open their whole bag of software goodies in exactly the same way. Pure marketing and PR. Either this source code is available or it is not. Time to get off the pot. |
| Robert Hampf - Subject: Re: hot upgrading of os!? (2000-01-29 08:04:15) |
| Wouldn't this open a door for viruses on Unix? |
| Kevin Forge - Subject: Re: hot upgrading of os!? (2000-01-29 12:38:43) |
| "Wouldn't this open a door for viruses on Unix?" In a word. No. |
| Martin Vermeer - Subject: Hot upgrade, my guess (2000-01-29 14:03:55) |
| I suppose this is only for multiprocessor installations. Shutdown part of your CPUs and bring them up running the new kernel. Then shut down the rest and do the same with them. I understand that's how discs in RAID drives are hot-swapped too. Or CPU's in SMP machines, for that matter. (Would work wouldn't it? Wouldn't it?? :-) |
| ac - Subject: W2K attacking SUN on the High End ? (2000-01-29 14:21:45) |
| Get Real, I agree that Linux may be attacking SUN on the low end. However, stating that W2K is attacking SUn on the high end is ridiculous. Even if you consider W2K a good OS, it doesn't run on High End Hardware, because it only runs on x86 PCs (No alpha, no PPC, no Sparc, no MIPS, not even the new Intel ia64 Architecture). However, Linux runs on all this architectures including in the IBM mainframes. If one of this systems is attacking SUN on the High End it's Linux. |
| Jamie - Subject: what does AIX have to offer? (2000-01-29 15:25:59) |
| What does AIX have that Linux kernel developers would want? |
| random static - Subject: hot upgrading - modules? (2000-01-29 15:31:59) |
| kernel modules let us at least modify the running kernel, if not replace it entirely. hypothetically, maybe, if more or less everything in a kernel was modularizable, one might replace the whole thing bit by bit, but (1) such modularization might be extremely hard to accomplish, and (2) by then you might be better off with some kind of microkernel anyway. think hurd, maybe? |
| Leen Besselink - Subject: Re: hot upgrading of os!? (2000-01-29 15:59:00) |
| Probably by modulizing most of it all, and just reloading modules... just my guess. |
| Anil Wang - Subject: Re: hot upgrading of os!? (2000-01-29 16:47:19) |
| > > Does the new kernel start life out as a user process and then "inherit" the > > data structures of the old bit by bit? This sounds like an interesting > > feature to implement. If it is true that you can hot-upgrade the kernel > > without any user processes noticing, that is amazing... > Wouldn't this open a door for viruses on Unix? Not if done properly (i.e. you can only do a hot upgrade if you log in as root). Most of the linux kernel can already be run in user space. Assuming that the whole thing could be run in user space, something like the following would need to be done: (1) Start up the new kernel (2) Mirror the state of the old kernel to the new kernel. (Believe it or not, there are some high availability algorithms that allow you to do this while the old kernel is running!) (3) Set a flag in the new kernel so that it can only recieve input, but not produce output (i.e. it can know what's going on in the system, but not change things). (4) Run both kernels in lockstep (i.e. both kernels do exactly the same thing). Both kernels will recieve the same input. When the user calls a system call in the main kernel, the system call in the new kernel is called. The main difference between the two kernels is that although their state will be the same, the only one affecting anything would be the old kernel. (5) Once step (3) has stabilized, change the flag on the old kernel so that it can read but not write, and change the flag on the new kernel so that it can read and write. Note, step (4) must be an atomic operation. (i.e you must do it all in one step) (6) Now get rid of the old kernel The process is a bit more complicated if userspace processes can have pointers to kernel data structures, but some fancy footwork can solve this problem too. This would make a great PhD Thesis, but I'm not sure how practical it would be to implement, since step (2) is very kernel dependent (unless you can store the kernel's state as XML:-] or some well defined backwards and forwards compatible format). Step (3) will likely require a change to each and every device in the kernel. I suspect that it would be much easier to do this in HURD (which has a microkernel) or the microkernel version of Linux on the Mac, since the microkernel can act as a mediator and do things that are difficult in user or kernel space. Of course, upgrading the microkernel is an entirely different issue:-) |
| Brian Strand - Subject: Thanks Anil (2000-01-30 01:46:11) |
| Interesting, I had similar thoughts in mind, up to step 2 :) . It seems like an enormous amount of work to make the kernel understand the in-memory structures of the previous version, as well as the (possibly structurally modified) in-memory structures of the current version. It makes module versioning seem positively blissful by comparison. I think the analogy I heard somewhere was "changing a tire while you're going down the highway at 160kph", or maybe s/a tire/an engine/. |
| Lin - Subject: Re: Linux commercial market will suffer. Only FreeBSD still have more attractive license for commercial customers. (2000-01-30 02:39:36) |
| I don't think the commercial market will suffer much. You have to realize that just about everyone connects Solaris with the SPARC architecture. Yes, Solaris works on x86 and indeed, the x86 version has always been available for a low cost ($100 - $200 for non-commercial use I think). However, adoption on the x86 platform has been minimal. The situation is this: Those already using SUN servers and workstations will willingly adopt Solaris 8 and probably be very happy. Those using the x86 (and in the future, Merced) architecture won't switch over from Linux (or even NT) because they simply can't see themselves running Solaris. |
| Evil Bifrost - Subject: Linux is a lame duck (2000-01-31 18:50:23) |
| Sun is putting out their source so that people can't complain that they don't have access to the source code anymore, thats pretty simple. Linux will not really be a commercial success for anyone, the only reason IBM is doing any porting work is so that they can save their dying architecture. Yes they will stay around for their mainframe and proprietary markets, but their Unix systems are already in the trash. If IBM wanted to work on a *truly* free OS, they would focus on NetBSD or OpenBSD, simply because you could sell it if you wanted to, and not tell anyone thats what it was, and the fact that *BSD is usually more stable. Instead, IBM is going for Linux market share, and when people buy IBM hardware to run Linux on, IBM's sales go up. IBM is trying to work into Sun's "Hardware only" market, and its not going to work. The only chance IBM has to make good is to port Linux or some other open source OS to their Sequent line of boxes. Blazing speed and really nicely scalable. |
Post your comments using the form below. [ Return to Today's Headlines | Top of Story ]
All times are recorded in UTC.
Copyright ©2000 internet.com.
(webmaster@linuxtoday.com)
Linux is a trademark of Linus Torvalds.
Powered by Linux 2.2.9 and Apache 1.3.6.
Linux Today is a corporate member of Linux International.