However, the killer app for 10 Gigabit Ethernet may be iSCSI, which takes advantage of the speeds and low latency that 10 Gigabit offers (see "iSCSI: A New Era in Storage"). New MANs (metropolitan area networks) will use 10 Gigabit Ethernet, thanks to the standard's ability to traverse long distances.
In addition, 10 Gigabit was designed for compatibility with SONET OC192 framing and, therefore, can be integrated with SONET networks, lowering equipment costs.
All this should translate into more high-speed services at lower costs, but there's less incentive than ever to innovate because so few CLECs (competitive local exchange carriers) have managed to stay in business. (For more on 10 Gigabit Ethernet, see "Networking for the Next Generation".)
Network Devices, Protocols Get Smarter
F5 Networks' Big-IP 5000 is a Layer 2 switch with intelligence through to Layer 7 (see "Big-IP 5000 Switch Marks New Territory for F5 Networks").
The Big-IP can load-balance requests among multiple Web servers. It can also distribute requests to Web servers based on the content requested. And, in addition to support for SSL acceleration, the Big-IP can redirect packets based on nearly any user-defined string in the packet.
Quality of service additions also have made networks smarter. QoS is usually done by setting up multiple transmit queues on routers and switches and dividing traffic among the queues based on assigned priority. Although most Layer 2 and Layer 3 devices now have this capability built in, it will become a necessity for those implementing VoIP (voice over IP).
One limitation to QoS, however, is that it can't affect traffic until it passes through the switch or router on which QoS is implemented. This means incoming traffic can fill your Internet pipe before your edge router has a chance to do its job.
Packeteer's PacketShaper 8500 and NetReality's WiseWan 5.0 Internet traffic-management products, however, are able to extend the reach of QoS by getting inside the TCP protocol and slowing undesirable traffic before it reaches your network's borders. These products seek out telltale bit patterns anywhere in the packet and use TCP to control the flow of data even before it leaves its source (see "PacketShaper 8500: Traffic Management Gets Smart,", and "New & Improved: NetReality WiseWan").
This capability is great for optimizing expensive Internet connections and has been the savior of many universities that have been brought to their knees by Napster and its kin, which like to hide inside Port 80.
Newcomer RouteScience Technologies' PathControl 1.1 solves a different problem by adding intelligence to the Border Gateway Protocol (see "Making a Science out of Routing on the Internet").
If you've purchased a second ISP connection and justified its expense by convincing yourself it would load-balance traffic and provide needed backup, you know that BGP and the complexities of the Internet make employing the second connection sucessfully, in a process known as multihoming, nearly impossible. Yes, adjustments can be made, but doing so takes lots of time and plenty of skilled labor. RouteScience's PathControl automates the process and provides dynamic balancing, which immediately reflects changing Internet conditions. PathControl eliminates the need to fiddle with BGP attributes and route maps.
Let the Chips Fall Where They May
This past year, the long-awaited 64-bit Merced chip--jointly developed by Intel Corp. and Hewlett-Packard Co. and now called Itanium--appeared in servers from all the major vendors. Not many computers were sold with the new chip, however, and because of a distinct lack of OS and application support, sales aren't likely to pick up anytime soon.
Microsoft's Windows 2000 Advanced Server Limited Edition runs on this 64-bit platform, but it's purely a developer's version. Microsoft is aiming to give developers operating system code to work with. More complete support is likely to follow in 64-bit versions of applications under Microsoft's .Net. The application will have to be recompiled to take advantage of the 64-bit technology and Intel's redesigned EPIC (Explicitly Parallel Instruction Computing) architecture.
As for the competition, Advanced Micro Devices has announced its answer to Itanium, code-named Hammer. Hammer is due to ship by year's end; Sledgehammer, the server version, is expected soon after that. If you think AMD lags behind Intel, consider that AMD's Hammer will run existing 32-bit apps faster than Itanium does. In fact, Intel's new architecture is so different that 32-bit applications won't run on Itanium without going through an emulator to translate between the two technologies. This means that 32-bit apps will run more slowly on Itanium than on Pentium 4 chips. AMD's Hammer doesn't have that problem because its basic architecture hasn't changed.
Ultimately, however, the trade-off will be that applications compiled for the new Intel architecture are going to perform a lot better--if they're ever written. Also, AMD's past success has been on the desktop, but Sledgehammer targets the server market. AMD's problem is simple: No major computer vendor is putting AMD chips in its servers, and none has announced plans to do so. The only OS support comes from SuSE Linux. But hold all bets. Intel has a new-generation Itanium chip, code-named McKinley, that the company says is twice as fast as the original.
And Intel is working on a version of the chip optimized for 32-bit apps, similar to AMD's approach (for more on AMD versus Intel, see "New McKinley Chip Details Surface,", and "AMD Processors in the Enterprise: Are They Ready for the Big Time?").
Linux Makes Strides in the Enterprise
It's been a good year for Linux, as the open-source OS continued to establish credibility in the enterprise (see "Special Report: Are We There Yet?"). Red Hat has had its woes but is wisely increasing its emphasis on the enterprise and on lowering total cost of ownership. These efforts seem to be paying off. Over the past year, Red Hat signed on some high-profile customers, including Amazon and Cisco, and IBM introduced hardware, including a mainframe, that supports Linux.
We tested multiple versions of Linux this past year, putting them through the same battery of tests we use to evaluate all OSs, and were so impressed (see "Special Report: SuSE Queues Up for a Clean Sweep") that we created a Well-Connected award for best Linux distribution.
Stalwarts claim Linux is staking claims on the desktop, but we've seen few such signs. Sun Microsystems released StarOffice 6.0 beta software; Lindows.com keeps threatening to release LindowsOS, which it says lets users run Windows apps on Linux without buying Windows; and CodeWeavers added CrossOver Office, designed to let specific Windows programs run on Linux. These are steps in the right direction, but don't expect Linux to take over the desktop anytime soon.
Peter Morrissey is a full-time faculty member of Syracuse University's School of Information Studies and a contributing editor and columnist for Network Computing. Send your comments on this column to him at ppmorris@syr.edu.