LAN Switching: A Strategic Decision

chinchillatidyΔίκτυα και Επικοινωνίες

26 Οκτ 2013 (πριν από 3 χρόνια και 5 μήνες)

63 εμφανίσεις

LAN Switching: A Strategic Decision

[
SEMI
-
FINAL DRAFT • 11/21/95
]


This white paper helps network managers evaluate the long
-
term functional and economic
aspects of deploying LAN switches. The material takes a total network perspective, rather than

exami
ne the LAN switch out of its context, for only with this broader perspective can a fully
-
informed decision be made. This perspective also permits LAN switching to become a long
-
term
strategic

network element, instead of an interim, throw
-
away step to incr
easing
bandwidth.


The document is organized into four sections. Section one describes why routing remains vital
in switched LANs. The second section compares the three dominant architectures being touted
for switched LANs. Section three highlights st
rategic plans from six major network vendors,
indicating the fundamental architecture preferred by each. The fourth and final section employs
a hypothetical network configuration, created by
Data Communications
magazine, to compare
price/performance and o
verall costs of the different architectures.



Routing’s Role in Switched LANs


Routing’s Role in Switched Networks
[
SIDEBAR
]



Eliminates the hard
-
to
-
manage “flat” network topology



Provides a more scalable and dependable hierarchical arrangement



Del
ivers efficient utilization of bandwidth



Facilitates transparent connectivity among diverse network types and protocols



Adds excellent security and firewall protection safeguards



Enables static and dynamic Virtual LANs (VLANs)



Offers the best cours
e for migration to ATM


Contrary to ambitious claims that switching heralds the end of routing, routing remains an
essential function as networks migrate from shared to switched media. For a strategic approach
to network planning, routing and switching mu
st be considered concurrently. Indeed, routing
provides the very framework for understanding the myriad switching permutations and
combinations.


Routing remains vitally important for a number of reasons. Routing facilitates transparent
connectivity am
ong diverse network types and protocols. Routing adds stability by making a
network more predictable, dependable and manageable. Routing adds security and firewall
safeguards. Routing maximizes efficiency of limited resources, especially for access to c
ommon
servers and in relatively low bandwidth WAN links.


The advent of ATM has further highlighted routing’s continued importance. The emerging
Multiprotocol Over ATM standard (MPOA), key to protecting a company’s investment in
LAN adapters, wiring and

other equipment, depends on routing. So the question of routing
remains not whether, but how.



LAN Switching Architectures


There are three dominant LAN switching architectures being widely touted today: centralized,
split and distributed. All three

are dependent on routing; they differ solely on the
location

of
routing functions. With the centralized architecture, all routing functions are placed in a single
location. The split architecture is a cross between the other two with
centralized

route
d
etermination and
distributed

packet forwarding. In the truly distributed architecture multilayer
switches, which perform both routing functions, are deployed throughout the network in a peer
mesh or hierarchical topology.


Centralized Routing
,

the histo
rically dominant architecture, employs a single or multiple
routers in a common location, frequently a facility’s computer center. LAN segments from
hubs, bridges, and cut
-
through or layer 2 switches throughout the facility feed the centralized
router(s).

A separate router port is required for each LAN segment and/or each Virtual LAN
(VLAN). As a result, the centralized architecture requires traditional routers that support
dozens, and even hundreds of ports. The resulting topology is a collapsed backbo
ne at the
centralized router(s).


This architecture is popular for one simple reason: until recently, central routers have been the
only commercial option available. The principal advantage of centralized routing is easy
management. Among its disadvan
tages are the single point of failure, poor scalability,
suboptimal performance and the high cost of central mega
-
routers.


Both cut
-
through and layer 2 switches, because they provide no internal routing functions, must
be deployed using a centralized ar
chitecture. To facilitate scaling such a network, companies
must invest in a router with sufficient long
-
term capacity, even though much of that capacity goes

unused initially. As the network grows beyond that capacity, the router must be replaced with a

larger one or supplemented with additional routers.


Performance is similarly handicapped. Under typical traffic patterns, packets regularly leave
one switch and traverse the central router on their way to other switches. The external router
and route
s become bottlenecks that dramatically decrease overall network throughput, thereby
diminishing any ostensible performance advantage offered by these switched network
architectures.


Poor scalability and suboptimal performance undermine the very reasons
for switching, which is
why many organizations are considering newer alternatives to the legacy centralized architecture.



[
DIAGRAM OF CENTRALIZED NETWORK
]


Split Routing

is one response to the scalability and performance issues of a centralized
architec
ture. Split routing separates the router’s path determination and table creation function
from the packet/frame forwarding function, and places these in separate devices. The concept
exists only in theory today, but is being touted as a beneficial archit
ecture by several traditional
router vendors. With split routing a centralized “route server” determines routes for the entire
network. These routes are then conveyed to distributed “data forwarders” that perform the
actual packet forwarding.


Split ro
uting has its roots in centralized routing. As LAN switching becomes pervasive,
centralized routers become bottlenecks. Rather than obsoleting their big routers, router vendors
are developing software that converts a central routing giant into a central
route server. Users
must also buy special
-
purpose switches to perform the packet forwarding function, deploying
these in a distributed fashion.


While no commercial route server is yet available, the concept is useful for portraying investment

protectio
n for centralized routers. Of course, current investments in layer 2 switches are
similarly jeopardized by the debut of proprietary data forwarders.


The disadvantages of split routing are virtually identical to those found with a centralized
architectu
re: poor scalability, single point of failure, suboptimal performance, and high costs
associated with equipment conversion and displacement. The poor scalability forces companies
to over
-
invest in a route server with adequate long
-
term capacity. As the
network grows
beyond that capacity, the route server must be replaced or supplemented adding substantially to
the cost and management complexity. An additional drawback is that, until a standard is
accepted in the industry, route servers will remain propr
ietary.


Network performance does improve going from centralized to split routing because no actual
traffic traverses the route server. The route server remains a bottleneck, however, because all
of the data forwarders regularly query the route server (
and wait for its response) every time a
packet arrives with a destination address that is not listed in the local cache of addresses and
routes. This is the case both for new, unknown addresses and “old” addresses that have timed
-
out or expired in the cac
he memory.


[
DIAGRAM OF SPLIT NETWORK
]


While split routing may be an acceptable compromise for companies dependent on a particular
product or vendor, its poor price/performance is pushing others toward a fully distributed
architecture.


Distributed Ro
uting
utilizes “multilayer” switches deployed throughout a network. Multilayer
switches are capable of independent switching at layer 2 and at layer 3. For this reason, the
multilayer switch operates as a full switch and router in one unit. Each multila
yer switch handles
both route determination and packet forwarding. Multilayer switches communicate with one
another, using standard routing protocols, to create and maintain the collective network routing
configuration. The resulting topology is either a

hierarchy or peer mesh of switches. In the
hierarchical arrangement, a large “master” multilayer switch functions as a collapsed backbone
serving smaller subordinate switches. The peer mesh arrangement has no such master; all
multilayer switches communi
cate freely with one another as traffic patterns require.


Distributed routing has a number of advantages. The network is more dependable with
alternate routes that eliminate single points of failure. The topology is substantially more flexible
and sca
lable, and offers both multivendor interoperability and incremental migration. It supports
an unlimited number of static or dynamic VLANs. And distributed routing facilitates migration
to ATM by locating packet processing close to the users. Two
perceiv
ed

disadvantages,
dependent on specific implementations rather than the architecture itself, are slightly more
complicated management and higher costs. The section on “Total Cost of Ownership” will
show that these are indeed misperceptions.


Flexibility

and scalability are compelling aspects of a distributed architecture. Multilayer
switches can be deployed gradually, as needed to grow the network or its bandwidth, alongside
or as replacements for shared media hubs or non
-
routing switches. Because each

multilayer
switch maintains its own routing table, the network is self
-
configuring. Each multilayer switch
can also be redeployed in another location or serving another role just as easily. Such flexibility
makes migrating from a purely centralized arch
itecture incremental, manageable and affordable.


The management complexity of a distributed architecture results from the need to keep all
multilayer switch software at a compatible revision level. Advances in network management
platforms and applicati
ons are centralizing and simplifying this once
-
difficult task. Similar
hardware and software technology advancements are also making distributed routing attractive
as a price/performance leader. Deployment flexibility, combined with elimination of all ne
twork
-
wide bottlenecks, preserves the investment in multilayer switches while delivering optimal real
-
world performance.


[
DIAGRAM OF DISTRIBUTED NETWORK
]


With both distributed and split routing, there remain two important functions delegated to
central
ized routing: conversion among protocols not supported by the multilayer switches and
an interface to the wide area network. Routers performing these two functions would be
located normally in the common equipment room, with each connected to the corpora
te
backbone. Multilayer switches route and bridge packets as needed to these routers just as they
do to other multilayer switches in the network

all using standard protocols for maximum
interoperability.



The table below offers a summary comparison of

the three routing architectures:


Routing Architecture

Advantages

Disadvantages

Centralized

Easiest to Manage

Poor Scalability

Single Point of Failure

Performance Bottleneck

Requires Port for Each LAN
Segment & Virtual LAN

Split

Unlimited Number of VL
ANs

Facilitates Migration to ATM

Most Expensive

Poor Scalability

Single Point of Failure

Performance Bottleneck

Route Servers Not Yet
Commercially Available

Proprietary

Distributed

Most Scalable

Best Flexibility

Standards
-
based

Dependable Mesh Topology

Un
limited Number of VLANs

Facilitates Migration to ATM

Available & Field
-
proven

More Complex Management



LAN Switch Designs
[
SIDEBAR
]


Although all LAN switches increase bandwidth by increasing network segmentation, there
are three different designs availa
ble: cut
-
through, layer 2 and multilayer.


Cut
-
through switches
are simple devices that forward packets based on destination
addresses without any additional processing. They do not check for bad packets, must
buffer incoming data streams when outbound

ports are congested, and cannot be used to
create hierarchical LANs. In effect, their simple design makes these devices fast and
inexpensive, but also inflexible. Cut
-
through switches also work exclusively with a single
MAC type and speed, so a 10 Mbps
Ethernet cut
-
through switch can only forward packets
to another 10 Mbps Ethernet LAN, and not to FDDI or 100 Mbps Ethernet backbone LANs.


Layer 2 and multilayer switches are more sophisticated devices that employ a store
-
and
-
forward design. Store
-
and
-
f
orward switches check for bad packets, perform sophisticated
filtering and forwarding, and can translate a packet to a different LAN type on a higher
speed backbone LAN, then switch the packet either at the MAC layer (layer 2 and
multilayer) or the network

layer (multilayer only).


Layer 2 Switches

switch packets at layer 2, the media access control or MAC layer. A
layer 2 switch has many similarities with a multiport bridge. Both cut
-
through and layer 2
switches can be used only with a centralized rout
ing architecture, which is why most
vendors and users alike are turning to multilayer switches.


Multilayer Switches
, also known as intelligent switches, can switch packets either at the
MAC layer (layer 2) or the Network layer (layer 3). Because it can

switch at layer 3, a
multilayer switch provides the full routing functionality needed in a distributed architecture.
Normally traffic within a virtual LAN segment is bridged, while traffic to other VLANs is
routed. The multilayer switch is the most flex
ible because it is the only design that can be
deployed in all three architectures: centralized, split and distributed. This is particularly
important for migration from a centralized to a distributed architecture in manageable and
affordable steps. Bec
ause the multilayer switch is a permanent building block for ATM, it is

also a strategic choice for switching.


A more detailed comparison of switch designs can be found in another white paper titled
LAN Switch Designs: A Tactical or Strategic Choice
av
ailable from Alantec.



Leading Vendor “Marketectures”


With so many disadvantages of centralized routing, it is not surprising that five of the six major
vendor “marketectures” (IBM’s SVN, DEC’s EnVISN, Bay’s BaySIS, Cisco’s Fusion and
Cabletron’s Synth
esis) feature either split or distributed architectures at the heart of their
strategic directions. The remaining major player, 3Com with HPSN, is sticking with the
traditional centralized architecture

at least for the time being.


IBM
, the penultimate
centralized vendor, has embraced split routing with its SVN (Switched
Virtual Networking) architecture for migrating to ATM. SVN features a centralized route
server with packet forwarding provided by special
-
purpose switches. While IBM’s SVN is
currently

short on product details, the eventual rollout is certain to have a profound impact on
existing centralized networks.


DEC

has embraced distributed routing with its EnVISN (Enterprise Virtual Intelligent Switched
Networks) strategy for migrating to ATM.

EnVISN is purely distributed, with no need for a
route server. DEC expects to ship its first intelligent switching products late in 1995.


Bay Networks
is touting a purely distributed architecture, as well, under its BaySIS strategy.
This is a signif
icant step for a company that has its roots in Wellfleet, a leading vendor of large,
centralized routers. Bay is currently integrating route processing into its layer 2 switches for
availability early in 1996.


Cisco
, the leading router vendor worldwide
, is moving to a split architecture under the Fusion
vision. Cisco plans to migrate the role for its large, centralized routers to that of a route server,
and is integrating route processing into some of its Catalyst switches.


Cabletron

also plans to s
upport a split architecture under its Synthesis and Securefast Virtual
Networking strategies. Cabletron is working on a route server, and already has at least some
multilayer switching capabilities in its MMAC
-
Plus product.


3Com

is the only major netwo
rking vendor attempting to hold onto the status quo of the
centralized architecture with its HPSN (High Performance Scalable Networking) strategy.
3Com may be hedging its bets, however, because the company’s LANplex switches are
multilayer devices that cu
rrently support routing in distributed topologies. Or perhaps 3Com is
struggling with a heavy dependence on ASIC technology, which is complicating a migration to
distributed or split routing. Whatever the situation, if 3Com holds to its centralized archi
tecture,
it will stand alone among the major network players.



Total Cost of Ownership


Data Communications
magazine created a hypothetical application for the purpose of
comparing the three fundamental architectures (“Next Generation Routing: Making S
ense of the
Marketectures” in the September 1995 edition). The mock request for proposal (RFP)
included three different configurations of 50, 250 and 500 switched Ethernet ports, all
connected to an ATM backbone. The RFP requested a complete configuratio
n of switches
(LAN and ATM) and routers. Per
-
port prices were calculated by dividing the total cost of all
equipment (LAN switches, routers, ATM switches and route servers, if applicable) by the
number of switched Ethernet ports.


The range of per
-
port
pricing is given below for the three different configurations and the three
different architectures. Pricing for the centralized architecture is the average of bids submitted
by three leading networking vendors. Only a single vendor submitted a proposal
for the split
architecture. Pricing for the distributed architecture is Alantec’s configuration. All pricing is US
list.


Routing Architecture

50 Node Network

250 Node Network

500 Node Network

Centralized

$1,270

$1,120

$910

Split

$1,920

$1,520

$1,435

Distributed

$801

$730

$657


Alantec’s distributed solution consists of PowerHub 6000 Intelligent Switches (multilayer), each
with an ATM backbone interface and dual power supplies, and Fore’s ASX
-
200BX ATM
switch. For non
-
ATM applications, an FDDI modu
le would be substituted for the ATM
backbone interface. All configurations also include Alantec’s PowerSight network management
software. The actual number of ports in each configuration is 60, 252 and 504 for true per
-
port
pricing of $667, $724 and $652

respectively. Because a single PowerHub 6000 handles up to
60 Ethernet ports, the ATM backbone switch was not required for the 50 node network,
making its per
-
port pricing appear disproportionately low.


[
DIAGRAM OF ALANTEC’S CONFIGURATION
]


The fact t
hat Alantec’s distributed routing solution represents an average savings of 33% over
centralized and 55% over split architectures belies the common misperception that the
distributed architecture is the most expensive of the three. Considering the greater

performance
of a distributed architecture, the price/performance advantage is even greater. The cause of this
common misperception is perspective. In a simple box
-
for
-
box comparison, it is not unusual for

a multilayer switch to cost more than a layer 2
or cut
-
through switch with the same number of
ports. But this is an apples and oranges comparison. More important is total cost of
ownership, which can only be evaluated using a full network perspective. Because the
multilayer switch supports full routi
ng capabilities, it eliminates the need for expensive centralized
routers or route servers. As a result, the
real

cost of the
total

network is normally much lower.


While less tangible than total cost, network management considerations often fall victim

to the
same kind of apples and oranges comparison. A multilayer switch, being a more sophisticated
device, should be expected to require somewhat more complex management than a simple layer
2 or cut
-
through switch. But taking a full network perspective
reveals that switches are not the
only pieces in the puzzle. A distributed architecture with multilayer switches involves both fewer
types of equipment and fewer nodes. The variety of equipment required by centralized and split
architectures, each with i
ts own specialized management application, makes maintaining support
staff expertise a real challenge. In most situations, managing less is easier than less management.



Summary


Distributed networks with multilayer switches not only offer substantial
architectural advantages
over the alternatives, they offer the best price/performance ratio and the lowest overall cost of
ownership. The multilayer switch is also a strategic choice with both short
-
term flexibility for
migrating from a centralized archit
ecture, and long
-
term durability as a permanent building block
for future migration to ATM. Multilayer switches are standards
-
based for maximum
multivendor interoperability, and offer the optimal design for implementing Virtual LANs. When
deployed in a f
ully distributed architecture, the full network scales easily and incrementally, and
operates reliably with its mesh topology.


Multilayer switches are, and will remain, the state
-
of
-
the
-
art in LAN switching. And Alantec, as

the leader in multilayer swi
tching, has more experience than other vendors entering the market
segment, along with the most comprehensive product line available.



# # #