Net Results

NC: Network Computer Or No Contest

So what's the solution? The Network Computer. The NC.

Stop your snickering. I don't buy all the hype about $500 devices that will liberate us from the tyranny of Microsoft. Everyone's searching for a rainbow, but you're as likely to find a simple solution that will meet your needs as you are to find a nice day for a swim in Syracuse during December. That doesn't mean there won't be technologies available that can move yo u toward the ideal of the NC. In fact, many of them have been on the market for years, but a variety of factors have conspired to keep them off your network.

Hooked on PCs I remember the first day I did it. After laboring away as a statistician keypunching cards in search of the perfect correlation, I fell in love with a system that let me edit records online. I was hooked on time-sharing, courtesy of an HP 2000. After a while, I got bored with the Hewlett-Packard and moved on to stronger stuff. VMS. MVS. All were very powerful and very reliable.

Not feeling content to stick with what worked day after day, I purchased my first microcomputer, a Radio Shack TRS-80. Not long after, I purchased one of the early IBM PC clones (I wrestled with whether to buy a Compaq or a Columbia and went with the latter since it seemed to be the stronger company). It's been job security ever since. Unlike the time-sharing systems of the past, these things never worked reliably, so if you knew something about how they worked, or didn't work as the case may be, people were anxious to hire you.

There was no stopping the PC revolution at universities like the one where I work. Lots of good ed ucational and research-oriented software was written for them and freely available from FTP sites over our speedy 9.6-Kbps IP over DECNet Internet connection. Anxious faculty members screamed for the construction of microcomputer labs, and we obliged. What a mess. A class would be scheduled for the lab, only to find that half the machines that worked yesterday weren't working today because some wacky computer science student had reformatted the disk drives. Or there was the time when another student modified all the AUTOEXEC.BAT files late one night so the next day at boot time an erotic image appeared on every screen. Animal House in cyberspace.

My point is these early experiences suggested that the model of microcomputer user as microcomputer systems administrator was bound to lead to chaos. As PCs made their way onto secretaries' and administrators' desks, it should have been obvious that the support costs would be extraordinary. Today, despite all the hype about plug-and-play and desktop management, i t's still a mess. But people live with it because they are imprisoned, dependent on these damn machines to get their work done.

It's Super Dave to the Rescue No, not me. It was this other guy Dave and his buddy Don. Dave and Don showed me how you could run a microcomputer lab without all the headaches to which we were accustomed. You just took this thing called an Arcnet card, this software from a small company in Utah called Novell and this little chip called a boot ROM, and you were all set. You see, Dave and Don were old mainframe guys who had been enlisted to set up microcomputer labs at their university. They understood this truth: If you want the system to work every day, don't let users mess with it. Pretty simple, isn't it? It's still true today.

By running their PCs without disk drives and remotely loading software, including the operating system, over the network, Dave and Don delivered microcomputer services that achieved reliability levels approaching the mainframe. As Dave used to say, it just works--every day.

Of course, progress got in the way a little. Windows came along. Applications became bloated. Faster PCs raised user performance expectations. But we've continued along this track with a fair amount of success. In fact, we run Windows95 on 200 Pentium PCs, remote booting without hard disks from Novell NetWare 4.1 servers. We configure them with generous amounts of RAM to minimize paging over the network and try to design our network infrastructure in a manner that provides the fastest network server I/O (no router hops allowed). It works. The only gotcha we've experienced relates to support for 32-bit WinSock applications--a problem that we have been trying to resolve with Microsoft for several months. Microsoft seems intent on proving to us that when it comes to difficult network problems, it still doesn't have a clue. But we call Microsoft every day and leave voicemail for whomever is assigned to our problem this week.

Do we achieve performance levels comparable to wha t you'd see on a standalone system with a fast hard-disk? Of course not, especially for people who want to run many applications concurrently. But our users, who generally are quite demanding, do not complain much about performance. With our planned implementation of Ethernet switching to the desktop and full-duplex Fast Ethernet pipes to the servers, performance will improve significantly. We're just waiting for the cost of switched Ethernet to get down to that magic number of $100 per port. That will happen very soon, perhaps by the time you read this.

Is diskless Windows the right answer for your environment? Not necessarily, but it does take us a long way in the direction of the NC model. And Microsoft's recent announcement of a NetPC reference specification includes software enhancements that should overcome some of the current performance limitations.

Revenge of Time-Sharing The NC that you read so much about today calls for dynamic loading of applications across the network--much the sa me way it's done in the diskless environment mentioned previously. Like that model, your level of performance is directly affected by the available bandwidth. So what do you do if you can't afford to deploy Ethernet switches or, worse yet, if you have to support this kind of configuration over a WAN? Another alternative exists that may be a better approach.

Remember X Windows? It's a great remote-computing model that delivers a graphical applications environment with minimal desktop intelligence. Many organizations have embraced the model by deploying X terminals running applications on remote Unix servers, but despite the significantly reduced cost of desktop maintenance and enhanced reliability, this market has never really taken off. I won't call it a niche market, but in comparison with the Wintel Goliath, it looks a lot like David (again n ot me, but the kid with the slingshot).

The problem, of course, is not a fundamental architectural flaw, but rather a lack of applications. While Sun Microsyste ms, Silicon Graphics, HP and others are still promoting the power of Unix workstations, most of us have long since relegated Unix to its appropriate role as an engineering desktop and a back-end application server. Given this state of affairs, it seems unlikely that a mainstream market for Motif-based applications will ever appear.

But don't give up. What if you could combine the benefits of X with the applications base of the Wintel market? That was precisely the idea a small company out of Coral Springs, Fla., had a few years ago. Not so small anymore, Citrix Systems provides an NC solution through its WinFrame product and the concept is quite simple. Take a commodity-priced multiprocessor Intel server running Windows NT and load it with RAM. Add some clever extensions on the server, an efficient remote presentation protocol similar to that found in X Windows and a very thin client capable of running on your uncle's old DOS-based 286, and you've got mainstream remote computing. You get all the popular ap plications you've come to depend on and none of the desktop systems administration hassles--and it all works over very low bandwidth connections. Soon, you'll even be able to access this from within your Web browser. That sounds like a better NC to me than the one we've heard so much about in recent weeks and months.

On the Other Hand I know what some of you are thinking. This is too weird for us. Things are working just fine using standard Microsoft tools, including policy restrictions and System Management Server. Or maybe some of you are turning to third parties for similar desktop management solutions. Hey, if it works, that's great. But the interest in NCs wouldn't be there if there weren't problems with current fat-client models. Maybe you need to attend some of the meetings I do. This is not all a conspiracy against Microsoft. The is sue is cost of ownership for client/server computing. The costs are way too high today for most executives to stomach. So take another look at the alternatives with an open mind. And remember the bottom line.

Dave Molta is director of network systems at Syracuse University. He can be reached at dmolta@nwc.com.