February 1999
Are You Buying Too Many Software Routers?
BY TONY RYBCZYNSKI
While few would dispute that routing is essential, it may be time to rethink exactly
where the routing function should reside. At present, routers are well entrenched in
internetworks. Indeed, the culmination of recent router developments, the software-based
multiprotocol router, is sometimes portrayed as a panacea. It is, for example, prescribed
for a variety of networking ailments, from campus congestion to security. There are,
however, alternatives to software-based multiprotocol routers. In fact, a range of
opportunities is emerging, thanks to the rise of the Internet, convergence on the Internet
protocol (IP), and improved economics for hardware switching.
WHERE ARE WE AND HOW DID WE GET HERE?
In the beginning, there were terminals attached to mainframes. Then, there were
minicomputers, microcomputers, and PCs. Then, Ethernet was invented to provide printer and
file sharing using vendor-specific Layer 3 protocols. These LANs had the characteristic
that they were shared by all users on the LAN and that all traffic was seen by each user -
a party line for data. Then, Layer 2 LAN bridges were invented, and they extended shared
media domains across the campus, and even across the WAN.
As traffic grew, flat LANs spanning hundreds or even thousands of nodes became totally
unmanageable and unwieldy, and some smart folks developed intelligent bridges that evolved
into software-based multiprotocol routers. The evolution from bridges to routers was
required for WAN scalability (unicasts vs. broadcasts) and resiliency (rapid dynamic
rerouting vs. slow spanning tree convergence).
As traffic grew, software-based multiprotocol routers proliferated in the WAN, and (for
some of the same reasons) in campus nets as well. It would be fair to say that
software-based multiprotocol routers have ingrained themselves within internetworks, and
that they have been all but indispensable there for over a decade.
But even as the software-based multiprotocol router was reaching its apotheosis, other
developments were pointing in new directions. Specifically, three things happened:
- The Internet changed everything.
- IP became the protocol of choice for new applications.
- Switching technology changed the economics of networking.
For the purposes of this discussion, the Internet is significant for having moved IP
from being one more protocol to be supported (for the techie UNIX shops) to being the
source of virtually all growth in data networks. In fact, IP has become the standard for
electronic business and online consumer infotainment. A million developers (give or take)
are working on myriad applications ranging from e-commerce and enterprise resource
planning to distributed games and home appliances.
So much for the points about the Internet and IP convergence. Now, what about
switching? Well, thanks to dramatic price/performance improvements in hardware switching,
this technology has won favor in the LAN and WAN alike. Moreover, while the
price/performance of hardware switching continues to improve, the cost of software-based
multiprotocol routers is actually going up. The cost, incidentally, includes both initial
deployment and continued operations. (For example, the Yankee Group estimates that the
operations cost per WAN router is $5K and rising. This figure reflects, in part, the
complexity of managing and configuring these routers.)
Given the opportunities spurred by the Internet, IP convergence, and hardware
switching, why would anyone continue deploying software-based multiprotocol routers? If
the expectation is that continuing to do the same thing (deploying software-based
multiprotocol routers) will lead to a different result (less management complexity and
better price performance) - well, that's a common definition of madness.
Rather than institutionalize ourselves, let's consider achieving different results by
doing something different. Let's consider some new alternatives. We may decide that
software-based routers are now being oversold, and that routing, while as essential as
ever from a functional point of view, may need to be deployed differently.
CHOICES, CHOICES, CHOICES FOR IMPROVED VALUE
Routers in the Internet: Not an enterprise issue, but illustrative of where the industry
is going overall. This world is rapidly moving toward very high-capacity IP-only Terabit
switch routers with lots of high-performance hardware for packet forwarding, traffic
management, and quality of service (QoS). That's in the core. At the CO edge of the
Internet, we're moving toward highly specialized, high-density remote access switches for
leased line and switched access, as well as for cable modem and xDSL termination. The
software router is becoming a thing of the past.
Routers in the campus: Where many of the price/performance
improvements have taken place. Software-based routers have been deployed to provide LAN
segmentation in large campuses. These are now making way for Layer 3 switches, which are
best defined in terms of performance and price: Many millions of packets/sec for Layer 3
switches compared to hundreds of thousands of packets/sec for software routers; $5/Mbps
for Layer 3 switches (and going down) compared to $50/Mbps for traditional routers (and
going up). A Layer 2 switch with a router blade is not on the Layer 3 switch
price/performance curve, so beware.
Inter-site routers at the WAN edge: What routing was designed for, so
this is where the value of multiprotocol routers still makes sense. But expect significant
developments in dramatically changed economics with IP-optimized switching platforms. But
you don't have to wait for these to simplify your environment. You can take advantage of a
new class of products - enterprise network switches. These new switches integrate
multiprotocol routing on a common platform that also acts as a frame/ATM switch and a
multiplexor. This provides a single wide-area infrastructure for all of your inter-site
bit, cell, frame, and packet traffic that can't be matched by today's routers.
Routers in the branch: Another variant of WAN edge routers with yet
different alternatives. In this case, frame relay is the current transport technology of
choice supporting inter-LAN, SNA, and (increasingly) voice traffic. In multi-branch
environments, access bandwidth is a major concern, as is meeting QoS performance needs
across multiple traffic types. One option is a multiprotocol router.
A second option is a multiservice FRAD. The key advantage is that there are
efficiencies to be gained by mapping SNA and voice directly onto frame relay, rather than
onto IP onto frame relay (for example, 60 percent reduction in overhead in handling
voice). In addition, there is more direct control of performance by leveraging frame relay
QoS capabilities directly.
SOHO routers: Another variant of WAN edge routers. The specialized
low-cost, consumer-oriented products (for example, ISDN routers) that have emerged in this
category are notable inasmuch as they emphasize ease of configuration.
This category also includes another class of devices: instant Internet gateways. These
PC-based platforms include easy installation, security features, dynamic IP address
assignment, and network address translation.
Routers at the Internet edge: A relatively new area driven by
enterprises leveraging the connectivity to expand their business reach through extranets
and remote access VPNs. Software-based routers are adding functionality (for example,
IPSec) to meet enterprise security needs, but once again a new class of specialized
extranet switches have emerged, delivering routing, VPN capabilities, profile-based access
and user authentication, bandwidth management, and various security features such as
encryption.
WILL YOU HAVE TO THROW AWAY YOUR INVESTMENT?
No. If it's working for you, don't fix it. But don't ignore opportunities to dramatically
simplify your environment through convergence on IP, bottleneck elimination, and
engineering. Here are some strategies:
- Move to IP as fast as possible: It will simplify your environment and
allow you to leverage the immense industry investments in IP-optimized products - in
switches, applications, and management.
- Avoid feature list paranoia: The old workhorses obviously have feature
lists as long as your arm. For the campus environment, many of the features are functional
dinosaurs and have no value today (for example, TCP/IP header compression and protocols
such as OSI); others are subtle technical differentiators that only the technically
skilled can exploit (for example, BGP default routes); others meet mainstream enterprise
needs that can be met in a number of ways.
In short, limit your list of "must have" features to what's really important.
Otherwise, you may find yourself paying a 25 percent premium on something you don't need,
and missing the opportunity to leverage IP-optimized multi-layer switching.
- Question the wisdom of router upgrades: Be sure you consider
alternatives to forklift upgrades. There are real opportunities for tenfold (and better)
performance improvements for the investment dollar.
- Anticipate investments in new applications and more bandwidth: Save
your money for these items. You'll need it!
- Re-deploy existing routers: Use your existing routers as multiprotocol
gateways for your legacy protocols, or make a generous donation to the schools in your
community.
Tony Rybczynski is director of strategic technologies and marketing for Nortel Networks' Enterprise Solutions. This business unit offers a full range of enterprise workgroup,
campus, and wide-area unified networks, through direct and indirect channels. For more
information, visit the company's web site at www.nortelnetworks.com.
E-mail questions or comments to the author at [email protected]. |