Drue stated his position on operating systems in "Last of the Mohicans" last week.If he's right, things certainly look bleak for increasing marginalized operating systems, but I don't think he is right. Much of Drue's blog was a point by point response to the arguments I put forward in "Ticket to Ride" and I think his points are worthy of rebuttal. First, on hardware changes Drue contends that Microsoft can get by with service pack type upgrades. Leaving aside the issue of whether customers want service packs to contain new features (as opposed to concentrating on fixing what's broken), I think there are flaws to the argument...
- Accommodating new hardware - finding a way to make optimal use of a new generation of massively parallel servers isn't like supporting a new SCSI controller. Supporting a Serial Attached SCSI controller is trivial, to the OS it just looks like any other SCSI controller, the host doesn't care whether it's using a serial or parallel implementation of the SCSI protocol,. Same goes for MPIO, just another layer inserted inserted the I/O stack. I believe that making the most of the next generation of hardware will take much more than changing the "Number_CPUs" in a header file and recompiling the kernel. Drue also asked the question about whether there was a 1000 processor, exabyte memory system in our future anytime soon. I'd say no, but a 128-core, terabyte memory system is just around the corner (2010 would be my guess), and that's a substantially different class of system to anything Windows has ever run on before, and I expect the OS will have to evolve accordingly.
- Application support - application vendors are slow to transition to new hardware architectures and Drue challenged us to name native Windows 64-bit applications. OK, how about Exchange 2007 which only runs on 64-bit Windows? I'm sure there are other ISVs with native 64-bit Windows applications, after all 64-bit Windows is getting on for a decade old (I went to the NT64 "kickoff" meeting in Redmond in 1998). Traditionally the 16-32bit and 32-64bit transitions have been slow for every hardware architecture, but based on previous examples such as MIPs and SPARC, I'd say Windows is on track.
- Modularity and the rise of the hypervisor - the question here is whether server virtualization fundamentally changes the way we use operating systems and there role in the infrastructure. In his blog, Drue suggests that VMware would like operating systems to remain constant and subsume more of the functionality for managing hardware into the virtualization layer. That seems directly at odds with the way things are going. Today, the hypervisor is actually trying to get out of the way of the operating system when it comes to dealing with hardware, so much of the onus for adapting to new hardware will remain on the operating system.
On the subject of revenue streams and Windows releases, I'm going to have to take issue with the idea that Windows 2008 took five years to produce. The fact is that MS took at least two years out of that working on Windows Server 2003 R2 which had some substantial changes, so it's at least 1.5 operating systems in 5 years which is about par for the course in my opinion.
In addition to his responses to my points, Drue also raised a couple of other issues:
- The threat of Google - I think the industry makes too much of this, Google has done great job at building the world's best search engine, but it's forays into other areas haven't exactly set the world alight. Hands up anybody considering replacing Microsoft Word with one of Google's web applications for the enterprise? Online applications and "cloud computing" are mostly vapor at the moment so there's a lot of scope for an OS vendor to carve out a segment for best cloud computing OS, and I wouldn't count Microsoft out.
- Slow adoption for new operating systems - Drue points to the slow adoption of Vista as evidence that corporations won't embrace Windows Server 2008. To a degree he's right, I don't expect Windows Server 2008 to be a majority of Windows Server shipments for at least 18 months. But it will inevitably supplant Windows Server 2003 at some point, just as Vista will ultimately displace XP.
Interestingly, Microsoft's tipped its hand on the future of operating systems last week by releasing a research version of a next generation operating system called Singularity. Singularity is a radical departure from just about any operating system that has gone before, it's not a minor rethink of the Windows kernel, it's a ground-up re-design, and not before time. If you look at widely used commercial operating systems, most can trace their design lineage back to MULTICS in the mid-60s, that's more than 40 years! Even Windows NT which is less than 20-years old owes more to DEC's VMS (a design from the mid-70s) than to any modern design. So there's plenty of scope for improving performance, scalability, security, and modularity of operating systems. If you want to know more about Singularity, there's excellent overview whitepaper here. So suffice to say, I think there's still quite a few Mohicans left out there in the woods of Redmond (and elsewhere for that matter.)
Anyway, enough for now, I'll sign off by making some predictions, so y'all can come back in 2012 and marvel at my prescience (or laugh at me
)
- Business as usual - Windows Server 2012 enters beta test late next year and ships in late 2011, followed by Windows Server 2016 (I hope they think of a more imaginative naming scheme)
- Throwaway hypervisors - Hypervisors, not operating systems get commoditized over the next few years. I believe that the real scope for innovation in server virtualization resides in hardware advances such as PCI I/O virtualization, and at the virtual machine management layer not the hypervisor which is destined to become the 21st century BIOS.
- Cloud computing - Operating systems for cloud computing are not just stranger than we imagine today, they are stranger than we can imagine. There's lots of work going on in this area, but's too early to say what the end result will be in anything more than the broadest terms. For more on Microsoft's thinking around cloud computing and its impact, see Ray Ozzie's recent interview.
- Microkernels make a comeback - The idea of microkernel operating systems have been around for a while, but haven't caught on in commercial implementations. I suspect that the new generation of hardware may change this and that we'll see a successful server-grade commercial microkernel in the next decade.
Posted by: Nik Simpson
It seems, to summarize, a Linux-based pDistro is probably best suited to tracking evolving hardware trends, while a Linux-based vDistro provides the hosting container for you enterprise services. And the Network OS is Federated, Open, Services Oriented, Secure and Agile. FOSSA. That sounds about right :-) http://www.novell.com/coolblogs/?p=689
Posted by: Robert Wipfel | April 10, 2008 at 12:39 PM