Is the end of the OS in sight?

THROUGHOUT THE ’90S, an IT buyer looking to add storage and other services to the company’s network simply bought more servers. Operating system-specific network clients and applications made integration so difficult that homogeneity, not interoperability, was the order of the day. Vendor consolidation, open source, and a decisive swing toward open, heterogeneous solutions have changed the shape of the OS market.

Looking ahead, it’s reasonable to wonder what role operating systems will play in the future of networked storage. Novell and Microsoft are watching the file-and-print market go up in smoke as customers rebel against rising per-seat license costs. When file servers are put out to pasture, they’re often replaced with cheap, low-maintenance, OS-agnostic NAS (network attached storage) appliances. Java and J2EE (Java 2 Enterprise Edition) have shifted application services out of the OS, in stark contrast to the OS-dependent model promoted by Microsoft. Does this spell the end of the OS?

As much as I hate to admit it, operating systems are becoming irrelevant. After all, until a few years back, the hardware dictated the OS you used, and you wrote applications to fit that OS. Now, operating systems are less about controlling hardware and more about providing a platform for application services. We saw this begin when Microsoft proved with Windows NT that a vendor could take a buggy, second-rate OS, paste a popular API on top of it, and watch the money roll in.

The horse race for the next decade isn’t about Windows vs. the ‘Nixes; it’s about Java vs. .Net, and this is coming from a self-proclaimed “not-developer.” For basics — file, e-mail, print, and simple network stuff — an appliance does the job and is easier to set up than a white box with an off-the-shelf OS.

True, NAS appliances present a comparatively appealing model in terms of lowering acquisition and operating costs. The plug-and-play approach has also taken hold in other areas, including fax and RAS (remote access services). But broader efforts to “appliantize” other network services have met with resistance. A general-purpose OS maximizes rack space by allowing each box to provide a range of services.

There is also the issue of freedom: Appliance vendors make the OS of each box invisible not only to reduce administrative overhead, but also to prevent users from expanding the unit’s software. Vendors would love to sell you one locked appliance for storage, another for e-mail, and yet another for your intranet.

P.J.: Tom, I love a good conspiracy theory as much as anyone, but I’m not buying this one of yours. Appliances are increasingly popular because they’re an inexpensive way to provide basic services.

You complain about expandability, but how big do appliances need to be? A Sun Cobalt Qube comes with provision for 150 users and takes less than half an hour to unpack, configure, and bring online. Sure, you wouldn’t run General Motors off a single Qube, but as a cheap, quick, purpose-built solution, it appeals to enterprise CTOs with shrinking IT staffs just as much as it does to small-business leaders. Even OS vendors such as Microsoft and Novell see this, and they’re trying to get their wares into appliances as we speak.

Tom: As do other popular low-profile servers, the Cobalt machines don’t try very hard to conceal their OS or try to block expandability. Cobalt’s primary added value is its simplified administrative interface, an appliancelike feature that other vendors, with help from Intel, are working to imitate. Those servers prove that manageability can be enhanced without locking customers into a hidden or proprietary OS.

Hooking a stack of drives to a general-purpose server (DAS, aka direct attached storage) may seem old-fashioned, but it’s still the most flexible and cost-effective way to add storage to a network. The popularity of NAS is rising fast, but 52 percent of readers polled in the 2002 InfoWorld Networked Storage Survey still use DAS. Running an OS that’s open to reconfiguration and enhancement makes it easier to scale up or repurpose existing hardware.

I think that adaptability is key, especially when IT has to squeeze every ounce of value from its assets. Closed systems such as storage appliances are easier to manage, but they’re more expensive to upgrade and subject to capability limits that don’t affect servers.

P.J.: Times are tough, and that means you have to spend money wisely. The main reason why respondents won’t be installing SANs isn’t cost — initial and ongoing — it’s the lack of need; 73percent of survey respondents say they don’t require the effort. Licensing costs are one issue driving the move to appliances — exactly who has time to count seats these days?

Then there’s the overhead of the whole server model, with the biannual upgrade pressure from vendors; the hassle of training and retraining administrators; and the peripheral gear, which is always on back order when you need it. I’ve grumbled about this for years. But until recently, there weren’t a lot of alternatives; so you either ponied up for the top-of-the-line solution or ran on substandard boxes.

Tom: Per-user charges are evil; that’s one point on which we agree. I’d rather pay a big flat fee for a server OS with unlimited client licenses than get nickel-and-dimed by a vendor that wants to charge me for every user (whatever a “user” is). The issue of rising, recurring per-seat license fees has driven a lot of companies away from Windows and NetWare. When asked to rank operating systems according to the quantity of storage trusted to each, survey respondents hand Unix a commanding lead over second-place Windows. That’s encouraging, and I expect Windows to continue to lose ground as a storage platform. I’d rather see companies save money by deploying Unix or Linux than by investing in closed, single-purpose appliances.