GigNet Conference White Paper:
Didn't get to go, but got this -
Four Trends, Four Technologies and the Future of Gigabit Networking
A White Paper by Lawrence Gasman
PresidentCommunications Industry Researchers
E-mail: LDG@CIR-INC.COM
Table of Contents
l.0 Introduction: The Changing Focus of Gigabit Networking
2.0 Four Applications Trends Shaping Gigabit Networking
2.1 Distributed Computing
2.2 Strategic Networking
2.3 Multimedia Networking
2.4 "Content Inflation"
2.5 Gigabit Applications: A Summing Up
3.0 Pumping Gigabits: Gigabit Ethernet, Fibre Channel and ATM
3.1 Gigabit Ethernet: Gigabits for the Masses or
Something Less?
3.2 Fibre Channel: HIPPI's Worthy Successor?
3.3 ATM: Networking Savior
4.0 SONET, ATM and Gigabits in the WAN
5.0 Conclusion: The Future of Gigabit Networking
-----------------------------------------------------------------------
l.0 Introduction: The Changing Focus of Gigabit Networking
Defined in terms of throughput, Gigabit nodes have been with us for
years. Indeed, if aggregate capacity is the defining characteristic of Gigabit networking, then Gigabit networking seems hardly worth singling out for special attention -- ATM switches with capacities in the tens of Gigabits per second are already commonplace. Even a large voice switch might qualify as a Gigabit switch if the total circuit switched capacity was considered.
Important work on Gigabit nodes continues, especially work on Gigabit
routers, which will be needed if the Internet infrastructure is to
scale up within its current technological paradigm. But the leading
edge of Gigabit networking is not in Gigabit nodes, but rather in
technologies that are capable of delivering 1 Gbps or more to an access point, either at the desktop or at some concentration point on the customer premises.
Even this type of Gigabit networking is not entirely new, the High-
Performance Parallel Interface (HIPPI) was developed at Los Alamos
National Laboratory, beginning in the 1980s. HIPPI in its initial form was capable of delivering a full-duplex point-to-point channel of up to 1.6 Gbps. But the story of HIPPI illustrates well why Gigabit
networking has been of marginal concern until recently. Specifically,
HIPPI is a technology/standard with a very narrow focus -- it was
primarily aimed at linking large computers (mainframes, and especially
supercomputers) with peripherals at high speeds. It was originally
designed to deliver Gigabit speeds between two points separated by no
more than 25 meters. As such, HIPPI was barely a networking technology at all and was limited in use to the computer rooms of the largest corporations, government departments and research universities.
But that focus is changing. Gigabit networking is moving outwards from the computer room into campus backbones, metropolitan area networks, and wide-area networks. HIPPI is being reengineered to meet these new needs, although new Gigabit networking technologies -- notably Fibre Channel, Gigabit Ethernet and ATM -- are likely to prove more potent in the long-run. In addition, the physical layers of networks are being constructed (and reconstructed) with SONET/SDH and wave division multiplexing (WDM) making them capable of what would have once been unimaginable speeds.
The reason Gigabit networking is exciting is because it is becoming
commercialized. Understanding why and how that is happening is the
purpose of this white paper.
2.0 Four Applications Trends Shaping Gigabit Networking
Communications Industry Researchers has identified four applications
trends that are driving ATM into the mainstream. These are :
distributed computing, strategic networking, multimedia networking and
"content inflation." All of these applications trends are big
promoters of the needs for networking bandwidth, but each also has
other requirements, whether conducted at Gigabit speeds or not. Thus
multimedia networking puts significant demands on networks to provide
good quality of service (QoS) characteristics. And strategic
networking entails good security.
In addition, each of these trends has an impact at differing locations
in the network. Distributed computing -- for the time being at least -- is a trend whose bandwidth inflating effects are felt mostly at the
desktop and across the campus backbone. Strategic networking is having an impact across workgroups, campus networks and WANs.
<snip>
3.0 Pumping Gigabits: Gigabit Ethernet, Fibre Channel and ATM
HIPPI, the oldest Gigabit technology, has been rejuvenated in several
ways. HIPPI has been extended from a point-to-point technology to a
true networking technology with the introduction of HIPPI crossbar
switches. HIPPI-SONET and HIPPI-ATM adaptors connect the localized
HIPPI environment to MANs and WANs. Meanwhile, so-called Serial HIPPI
will run HIPPI protocols over single mode fiber for up to 10 km, while
the latest version of HIPPI (the HIPPI 6400 standard), bumps up HIPPI's top data rate to 6.4 Gbps.
With such improvements it can be assumed that HIPPI will be around for
quite some time. However, the focus of most user interest and product
development in the Gigabit networking world is on three other
approaches: Gigabit Ethernet, Fibre Channel and ATM. All of these
appear to offer advantages over HIPPI in terms of versatility and/or
cost. And Gigabit Ethernet, in particular, promises to bring Gigabit
networking into the mainstream business networking market. For the
wide-area market space, SONET/SDH networking has also become a critical part of the Gigabit networking equation.
3.1 Gigabit Ethernet: Gigabits for the Masses or Something Less?
Ethernet is one of the great success stories of the networking
industry. Ethernets are now to be found in the smallest of small
businesses and in the largest corporation. In addition, Ethernet has
shown itself capable of scaling from the original 1 Mbps and 10 Mbps
versions to the 100 Mbps, "Fast Ethernet." Now comes word that
Ethernet may be scaled up to 1 Gbps and perhaps beyond. Products are
appearing already and by the spring of 1998, the IEEE 802.z group will
announce a formal standard for Gigabit Ethernet too.
But while Gigabit Ethernet may be behind the other technologies in
terms of standardization and product development, its potential
benefits are enormous. It will run comfortably in an existing Ethernet environment and will be immediately compatible with existing Ethernet software and hardware. No other Gigabit networking technology can make such a claim. Gigabit Ethernet is expected by most observers to follow the same economics as Fast Ethernet, whose price is now dropped to a point where it is just a little more expensive than standard Ethernet.
So in a few years we may see Gigabit Ethernet interfaces available at
Radio Shack for $150 and they will have plug-and-play capability, so
that they can be installed by anyone capable of installing standard
Ethernet. Again, no other Gigabit networking technology can boast as
much.
CIR believes, however, that some caution is required of vendors, end
users and investors, who are currently betting on a Gigabit Ethernet
future. First, perhaps to state the obvious, Gigabit Ethernet is an
unstandardized and barely commercialized technology. Its prospects
look great -- but then again, so did those of FDDI! More importantly,
Gigabit Ethernet's weaknesses are exactly in those areas where the
needs for Gigabit networking appears to call for strengths.
Gigabit Ethernet is, after all is said and done, just Ethernet, and has that technology's cons as well as its pros. There can be inherent
delays in any kind of Ethernet, there are no built in QoS guarantees
and as little as 40 percent of the available bandwidth may actually be
available because of the fact that Ethernet (again inherently) uses a
media access approach based on contention of data that are trying to
use the same bandwidth. (As far as the latter is concerned, however, it is not entirely clear whether Gigabit Ethernet will use collision
detection, although a networking approach that used something else
would certainly cause Ethernet purists to demand that it be called
something other than Ethernet.)
Since, as we have seen, the applications for Gigabit networking are
frequently those that require low latency, good QoS control and (by
definition) lots of bandwidth, these features of Gigabit Ethernet are a little worrying. Yet another limiting issue is that Gigabit Ethernet is really a LAN technology, which is only extensible to 25 meters on copper, although, using single-mode fiber, it can be extended up to 2 km. Hence Gigabit Ethernet may also be used to build campus backbones.
The good news is that considerable attention is being given to propping up Gigabit Ethernet in those areas where it is weakest. Since everyone believes that multimedia is the wave of the future, QoS guarantees are of great importance. In Ethernet, one approach is to create virtual LANs with packet headers that identify traffic priority. Another approach is to use the RSVP protocol to allow for bandwidth reservation over an Ethernet. Meanwhile, while there are no plans to turn Gigabit Ethernet into a WAN technology, copper-based Gigabit Ethernet is being extended to 100 meters utilizing CAT-5 UTP as a medium.
3.2 Fibre Channel: HIPPI's Worthy Successor?
Fibre Channel has sometimes been billed as the successor to HIPPI in
that it attacks a very similar market space -- the interconnection of
powerful workstations, computers and peripherals. Some of Fibre
Channel's biggest backers once predicted that this technology would
become common for LAN backbones, but this has not really ever occurred. Instead Fibre Channel is being used mostly for linking large computers to storage arrays -- RAID storage systems now frequently come with a Fibre Channel option. It is also being used for connecting up multiple processors for parallel processing applications.
Beginning in 1988, the Fibre Channel standards were developed by the
ANSI X3T11 committee, whose objective was to provide a low overhead,
high-speed approach to transfer data among workstations, large
computers and peripherals. Fibre Channel is also designed to carry many different kinds of data including ATM, HIPPI, SCSI and TCP/IP formats.
Fibre Channel is also very versatile in terms of topology and offers
three basic types: point-to-point, arbitrated loop and fabric
switching. The first and last of these topologies are self-
explanatory. The arbitrated loop is essentially an unswitched LAN,
something like a token-ring. Finally, Fibre Channel is defined for a wide variety of media types including STP, various types of coax and (of course) multimode and single-mode fiber. Using single-mode fiber, Fibre Channel can be extended up to 10 km and the standard supports four data rates: 133 Mbps, 266 Mbps, 531 Mbps and 1.06 Gbps. ANSI has approved 2.134 Gbps and 4.25 Gbps speeds, but these are not yet available in commercial products.
On the face of it, Fibre Channel has many of the characteristics that
appear to be needed for Gigabit networking. It offers a high bandwidth which is efficiently used and there are few latency problems. Fibre Channel is also extensible to a point that it can be used to create quite large networks, this being the main advantage that Fibre Channel has over HIPPI. Finally, the Fibre Channel standard offers four classes of service, so that it can resolve at least some of the QoS concerns associated with Gigabit networking.
3.3 ATM: Networking Savior
Asynchronous Transfer Mode (ATM) has been billed as the ultimate
networking technology. It can be used for LANs, MANs and WANs. It has good latency characteristics and arguably the best QoS characteristics of any comparable networking technology and also is relatively efficient in terms of its bandwidth usage. However, ATM remains relatively expensive and difficult to implement and, despite early predictions that this technology would soon dominate the network from end to end, ATM faces considerable challenges from frame relay in the WAN and Fast Ethernet (and soon Gigabit Ethernet) at the desktop and local backbone level.
Most importantly in the context of this report, ATM is not really a
Gigabit technology. While the capacity of even the smallest ATM
switches are measured in the Gigabit range, the ATM Forum only recently defined a Gigabit-capable ATM interfaces. This operates at the OC-48 level (2.4 Gbps). But this interface has yet to become widely available for commercial applications. When it does it will probably be first used in service provider backbones. It is possible, however, to create ATM networks by combining several lower-speed ATM interfaces, although it is not clear how often this is actually done in practice.
4.0 SONET, ATM and Gigabits in the WAN
Some day ATM may be the underpinning of all Gigabit networking. It is
highly scalable and there would, for example, be no major technical
problems associated with creating 10 Gbps ATM interfaces. For the time being, however, there is no major R&D focus on this type of work. The near term future for Gigabit networking to the desktop and in campus and other local backbone environments will be found a mix of ATM, Fibre Channel, HIPPI and Gigabit Ethernet.
This may not be very intellectually satisfying, but it is the most
likely scenario and it is rendered more palatable by the fact that all
of these Gigabit networking technologies interwork with relative ease.
For example, ATM can encapsulate a wide variety of protocols for
transport in ATM cells or can emulate LANs including Gigabit Ethernet.
Meanwhile, Fibre Channel is designed to map HIPPI and ATM over its
physical layer, the structure of which it shares with Gigabit Ethernet.
Once we start moving into the realm of the WAN, only ATM really offers
a potential solution to the Gigabit networking problem. But ATM needs
a physical layer to sit upon. In the WAN, this layer is likely to be
SONET in North America or the almost identical SDH standards in Europe
and elsewhere. SONET and SDH offer a complete hierarchy for
transmission standards above the traditional T1/E1 standards.
Potentially, this hierarchy is scalable up to Terabits per second, but
for the time being the three Gigabit transmission speeds in the
SONET/SDH hierarchy that are actually in use commercially are OC-48
(2.5 Gbps), OC-96 (5 Gbps) and OC-192 (10 Gbps). The highest levels of
the SONET/SDH hierarchy are mainly used in major trunking facilities,
including submarine cables.
SONET/SDH is more than just a transmission hierarchy, the technology
associated with it also offers special features such as add/drop
multiplexing and a very effective self-healing mechanism. These are
not issues that are directly related to Gigabit networking. Gigabit
networking does, however, begin to put some significant strains on
conventional SONET/SDH technology and this strain is creating some
interesting new innovations.
The most important of these innovations is wave length division
multiplexing (WDM). WDM is essentially a version of frequency division multiplexing and enables multiple signals to be transmitted at different wavelengths along the same physical circuit. In the past
fiber optics has offered enough bandwidth to fill the usable bandwidth
with one signal, but the higher levels of Gigabit networking are
pushing the capabilities of conventional fiber optics to the limits.
With WDM it is possible to double or, quadruple the capacity of fiber
optic systems.
It is also possible that Gigabit networking will lead to new R&D on the glass used for fiber optics. In the early, 1980s when fiber was first becoming widely commercialized there was a considerable amount of work done on creating glass pure enough to transport data at very high rates. Eventually fiber was created that was considered pure enough to support any networking requirements then conceivable. Now this issue has reemerged and there are even discrete mutterings by some observers of the fiber optic industry that there will eventually need to be major recabling of long-distance networks, although this may be little more than alarmist talk.
5.0 Conclusion: The Future of Gigabit Networking
CIR believes that while some parts of the Gigabit networking future are still quite unclear, other parts are now next to inevitable. In the WAN, the future is clearly one defined by ATM-over-SONET. In the LAN and campus networks, the answer is less clear. There will certainly be a mixture of ATM, Fibre Channel, HIPPI and Gigabit Ethernet. But the question is: which one will predominate? CIR's answer to this question, at least for the next couple of years, is that Gigabit Ethernet will be the most popular choice. This networking technology not only has technical and cost advantages, but it is also backed by most of the large networking companies. (Marketing clout is a distinct advantage in such matters.)
In the more distant future -- which in networking parlance means in a
couple of years -- Gigabit Ethernet may be overtaken by ATM. Or there may be new and faster versions of Ethernet. Terabit Ethernet may sound like science fiction now, but Gigabit Ethernet would have had the same ring about it just a few years ago. And what of Fibre Channel? This Gigabit technology is the Rodney Dangerfield of networking; it seems to get very little respect. Fibre Channel does seem to have some very attractive characteristics, but has never been taken up by the mainstream networking industry and seems to be overshadowed by ATM.
But however Gigabit networking does shape up, it is clear that Gigabit
networking does indeed have a future -- the applications that we
discussed at the beginning of this White Paper make this certain.
* * * * * * * * *
This White Paper was prepared by Communications Industry Researchers, Inc. which is solely responsible for its content, and is made available by Technology Transfer Institute, producers of the GIGNET Conference and Exhibition on the GIGNET web site: tticom.com
Additional information about the event is available on the web site or by contacting custserv@tticom.com.
* * * * * * * * * |
