*Should IS pay attention to WAN compression? (Tech View) (PC Week Netweek)
PC Week  August 29, 1994 v11 n34 pN17(2)
PC Week  August 29, 1994 v11 n34 pN17(2)

Should IS pay attention to WAN compression? (Tech View) (PC
Week Netweek)

by	Blakeley, Michael

Abstract
WAN router and bridge vendors are lauding the merits of their various
compression schemes, claiming they can provide users with more
bandwidth for their buck. Many users would like to believe them, since
they would rather continue leasing low-cost 56-Kbps lines instead of
upgrading to expensive T-1 connections. Tests show that saturated WAN
connections can handle almost two times as much traffic with
compression. Before purchasing the requisite hardware, however,
potential buyers should first measure how much of their WAN bandwidth
is being used with a WAN/LAN protocol analyzer. They should also
realize that older 386- and 68030-based bridges and routers cannot
handle the increased processing demands of compression. Vendors
currently employ proprietary schemes for compression, so buyers should
choose one carefully.

Full Text
Vendors of WAN routers and bridges promote their compression schemes
as providing more bandwidth without spending more money. Does WAN
compression make sense for corporate networks, or is it just hype?

WAN compression attempts to compress data before it enters the WAN
and decompress it on the other side. Vendors of WAN equipment argue
that compression increases throughput between sites at a lower cost
than buying new or faster WAN lines. Since the price differential for
both installation and connections on a 56K-bps line is dirt-cheap and a
T-1 line is still quite expensive, many firms would like to stay with
56K-bps lines for as long as possible.

At the same time, many WAN links are saturated, and corporate
productivity suffers from internetwork gridlock. If corporations mix
data and voice on a single line, WAN compression would ideally allow
more voice connections and data throughput on the same T-1 or
fractional T-1 line.

If your WAN link isn't saturated, of course, you won't see any
improvement from compression. Do you know how much of your WAN
bandwidth is used? Do you know how many packets are dropped (or
refused) by your WAN link? You need these numbers before you can
decide between WAN compression and a faster, more expensive data line.
Many routers and bridges can measure and log their packet-loss rates
and throughput rates. However, for completely reliable measurements,
you need a WAN/LAN protocol analyzer.

It may be difficult to measure packet loss on bridges, as bridges are
MAC-layer transparent. One possibility is to compile a list of "foreign"
MAC addresses and measure traffic on those addresses.

Let the IS manager beware

Purchasers of WAN infrastructure equipment may not realize, however,
that WAN compression isn't a free lunch. Compression requires CPU
bandwidth, vendor loyalty, and higher prices. Compression vendors also
exaggerate the bandwidth improvement that users can expect. Although
many vendors claim a fourfold or sixfold increase in bandwidth,
real-world improvements will depend on the compressibility of your
network traffic.

PC Week Labs believes that most saturated WAN links will see an
improvement of slightly less than twofold. Many WAN routers and
bridges are based on slower CPUs, such as Intel Corp.'s 386 and Motorola
Inc.'s 68030. If your WAN equipment doesn't have enough bandwidth to
handle compression and decompression of the WAN link, turning on
compression could hurt your performance. More packets will be dropped
or refused by the router, and data may not make it from one site to
another.

Vendors such as Cisco Systems Inc. have addressed this problem in two
ways. First, many routers and bridges already use either subprocessors
to off-load many tasks from the main CPU or multiple processors to
distribute the load over several CPUs. Second, the compression and
decompression chores may be handled on each WAN card with a dedicated
codec chip. Newport Systems Solutions Inc. uses this approach in its
LAN2LAN/MPR product (see Lab Note, at right).

A third approach off-loads the compression chores onto an external
device. The DSU Consortium is presently working on a standard for
DSU-to-DSU compression. In the meantime, proprietary solutions such as
Motorola Codex's 3512 SDC are available. Even with plenty of processing
power, however, compression and decompression of high-speed links
takes time. If slow RAM buffers and a relatively slow codec add latency
to the WAN link, user-perceived performance will be lower, despite the
increased bandwidth.

Since there is no open standard for WAN compression, each vendor must
invest development effort and licensing fees for a proprietary solution.
By increasing R&D costs, the lack of a standard increases end-user
prices. It also ensures brand-name loyalty, since no two WAN
compression standards will interoperate. WAN compression isn't right
for everyone, but it will improve some networks. Measure your traffic,
and if you still think compression will help, try it out with your favorite
vendor's equipment.


Topic:	MIS
	WAN
	Data compression
	Guidelines
	Hardware Selection

Record#	16 226 318

COPYRIGHT Ziff-Davis Publishing Company 1994