19-09-2014, 10:18 AM
TERABIT SWITCHES AND ROUTERS
TERABIT S.pdf (Size: 255.17 KB / Downloads: 39)
ABSTRACT
Just a few years back, no one would have thought that internet traffic will increase at such a rapid rate that even
gigabit capacity routers in the backbone will be insufficient to handle it. Today, routers with terabit switching
capacities have become an essential requirement of the core backbone networks and gigabit routers find their
place at the mid-core or even the edge. This survey paper explains the issues in designing terabit routers and the
solutions for them. It also discusses about some of the key products in this area.
INTRODUCTION
In the present network infrastructure, world's communication service providers are laying fiber at very rapid rates.
And most of the fiber connections are now being terminated using DWDM. The combination of fiber and DWDM
have made raw bandwidth available in abundance. 64-channel OC-192 capacity fibers are not uncommon these
days and OC-768 speeds will be available soon. Terabit routing technologies are required to convert massive
amounts of raw bandwidth into usable bandwidth. Present day network infrastructure is shown in Fig 1. Currently,
Add/Drop multiplexers are used for spreading a high-speed optical interface across multiple lower-capacity
interfaces of traditional routers. But carriers require high-speed router interfaces that can directly connect to the
high-speed DWDM equipment to ensure optical inter operability. This will also remove the overhead associated
with the extra technologies to enable more economical and efficient wide area communications. As the number of
channels transmitted on a single fiber increases with DWDM, routers must also scale port densities to handle all
those channels. With increase in the speed of interfaces as well as the port density, next thing which routers need
to improve on is the internal switching capacity. 64-channel OC-192 will require over a terabit of switching
capacity. Considering an example, a current state-of-the-art gigabit router with 40 Gbps switch capacity can
support only a 4-channel OC-48 DWDM connection. Four of these will be required to support a 16-channel
OC-48 DWDM connection. And 16 of these are required to support 16-channel OC-192 DWDM connection with
a layer of 16 4::1 SONET Add/Drop Multiplexers in between. In comparison to that a single router with terabit
switching capacity can support 16-channel OC-192 DWDM connection. With this introduction, we now proceed
to understand what is required to build full routers with terabit capacities.
Evolution of Present Day Routers
The architecture of earliest routers was based on that of a computer as shown in Fig 2. It has a shared central bus,
central CPU, memory and the Line cards for input and output ports. Line cards provide MAC-layer functionality
and connects to the external links. Each incoming packet is transferred to the CPU across the shared bus.
Forwarding decision is made there and the packet then traverses the shared bus again to the output port.
Performance of these routers is limited mainly by two factors : first, processing power of the central CPU since
route table search is a highly time-consuming task and second, the fact that every packet has to traverse twice
through the shared bus.
SWITCHING Vs ROUTING
The basic difference between switching and routing is that switching uses 'indexing' for determining the next hop
for a packet in the address table whereas routing uses 'searching'. Since indexing is O(1) operation, it is much
faster than any search technique. Because of this, many people started thinking about replacing routers with
switches wherever possible and vendors flooded the market with several products under the name of switches. To
differentiate their products, vendors gave different names to them like Layer 3 Switch, IP Switch, Layer 4 Switch,
Tag Switch etc. and regardless of what a product does, it is called a switch [Decis97] [Decis96][Torrent].
Therefore it is important to understand the difference between all these different forms of switches.
Full Routing
Some of the latest products in the market perform full routing at very high speeds. Instead of using a route cache,
these products actually perform a complete routing table search for every packet. These products are often called
Real Gigabit Routers, Gigabit Switching Routers etc. By eliminating the route cache, these products have a
predictable performance for all traffic at all times even in most complex internetworks. Unlike other forms of
layer 3 switches, these products improve all aspects of routing to gigabit speeds and not just a subset. These
products are suited for deployment in large scale carrier backbones. Some of the techniques used in these products
to improve route lookup are discussed later in the paper.
EFFICIENT ROUTING TABLE SEARCH
One of the major bottlenecks in backbone routers is the need to compute the longest prefix match for each
incoming packet. Data links now operate at gigabits/sec or more and generate nearly 150,000 packets per second
at each interface. New protocols, such as RSVP, require route selection based on Protocol Number, Source
Address, Destination Port and source Port and therefore make it even more time consuming. The speed of a route
lookup algorithm is determined by the number of memory accesses and the speed of the memory. This should be
kept in mind while evaluating various route lookup techniques described below.
Route Search at Gigabit Speeds
The solutions described above solve the route lookup problem in most cases. But with media speeds going up, it
requires very careful implementation of one or more of the above techniques combined together to have possible
advantages from all of them. With 1.5 million packets coming in, a router has only 672 nanoseconds to validate a
packet, find an outgoing route and send the packet. Many vendors and research groups in universities have come
up with innovative solutions for this. Details of most of the proprietary solutions from vendors have not been
disclosed because of patent pending or similar reasons. Some of the well known solutions are mentioned below.
SUMMARY
It is very clear now that with deployment of more and more fiber and improvements in DWDM technology,
terabit capacity routers are required to convert the abundant raw bandwidth into useful bandwidth. These routers
require fast switched backplanes and multiple forwarding engines to eliminate the bottlenecks provided in
traditional routers. Ability to efficiently support differentiated services is another feature which will be used along
with total switching capacity to evaluate these routers. Switching is faster than routing, and many products in the
market combine some sort of switching with routing functionality to improve the performance and it is important
to understand what the product actually does. But the products which scale up all aspects of routing rather than
the subset of them, are bound to perform better with arbitrary traffic patterns. Route lookup is the major
bottleneck in the performance of routers and many efficient solutions are being proposed to improve it. Supporting
differentiated services at such high interface speeds poses some new challenges for the design of router
architecture and some solutions are discussed here. Finally a survey of leading market products is presented.