The dream OS

An OS tester fantasizes about building the best, most stable operating system.

My worst nightmare would be to wake up as either Linus Torvalds, heavyweight Linux wrestler of the world; Ray Ozzie, holder of the future of Windows; or Jonathan Schwartz, alter ego of Sun’s Scott McNealy. All are the lightning rods in a never-ceasing thunderstorm of new technology advances in hardware and mind-numbingly sophisticated software applications. To live in their respective roles requires ego, panache and vision – and Kevlar armor.

Each of these protagonists had one of his products tested by Network World last year. If we were to take the successful components of each and graft together our own perfect server operating system, the result would be recognizable – but barely.

First, we’d take all of the bloat from every network server operating system. Every superfluous driver would be stripped, and every extraneous piece of code placed onto a spare DVD to be used only if we called for it. Hardware compatibility at installation time mandates having all drivers ever conceived loaded onto the server’s storage media. This is like bringing in the 5th Army division when a single sniper is needed.

We’d remove all but the core command sets. All obscure executable items and 99% of such things as fonts, sounds and stock pictures should be loaded only when requested. Ask for the media or, better yet, download it from a secure/authenticated Web site at run-time, when needed. These two actions – deleting drivers and culling obscure executables – would allow most operating systems to be held on a single gigabyte USB flash drive fob. No flexibility would be denied. Bloat would go away.

Moving on from the hardware, we’d give our operating system the rapid porting stability of Linux 2.6. We’d add in Sun’s DTrace (a tool that rapidly determines where code is wasting time or is in an error condition), which helps eliminate the finger-pointing when code doesn’t execute to expectation.

We’d take the speed of Windows 64-bit editions, late as they were to the 64-bit race, or Solaris 10’s seemingly unfettered responsiveness. Both are comparatively ugly, however, so we’d add the luxurious Apple’s Panther/OS X.4 GUI.

All the best cross-platform virtualization features of EMC’s VMware would be mandatory. But we’d add in the ease of use of Microsoft’s Virtual Server, the Redmond giant’s own flavor of virtualization software that lets you run multiple instances of Windows on the same physical server. Because we’d want to host multiple instances of whatever we’d like, and we’d also want to isolate them as need be, the operating system would require the highly structured, kernel security model of Red Hat’s Advanced Server 4.0.

This operating system would have to understand multicore CPUs. They’re the future of computing as higher CPU clock speeds become more difficult for CPU vendors to support. Supporting numerous flavors of filing systems – moving beyond the chest thumping that one is better than another – also would be a must. Give us our choice of WinFS, or Reiser or ZFS. Cross-license them – free – to each other and everyone so we no longer must appease proprietary, standards-less, incompatible applications.

The kicker here is that we’d like this super file system to run on all of the processor families available. No special editions for 32-bit, 64-bit or multicore. No extra surcharges for the Intel Itanium, AMD 64, UltraSPARC or even million instructions per section. No tiny editions, medium editions, standard editions, enterprise editions, huge fat corporate editions or galactic cluster editions. Just one set of code for developers and administrators and network engineers and civilian users.

In turn, there must be no second-class clients. You either support them in a festival of harmony and egalitarianism, or say, “We don’t support them; now go away.” No half-attempts, no client ghettos allowed.

OK, we understand that we’re totally daydreaming about complete compatibility in a competitive environment, but we do feel it’s something to shoot for. Our antecedents drilled into us what compatibility and standards mean: We fail when they become an oxymoron.

Henderson, a Network World Lab Alliance member, is principal researcher and managing director of ExtremeLabs. He can be reached at thenderson@extremelabs.com.

Previous story: Aruba’s A2400 and A800 switches and A61 access point

Next story: What to know before you buy >