DATA COMPRESSION

 
DATA COMPRESSIION
Data compression is a standard feature of most bridges and
routers, as well as modems, especially those used for transferring
bulky files over wireless links. Compression improves
throughput by capitalizing on the redundancies found in the
data to reduce frame size and thereby allow more data to be
transmitted over a link. An algorithm detects repeating characters
or strings of characters and represents them as a symbol
or token. At the receiving end, the process works in reverse
to restore the original data.
There are many different algorithms available to compress
data, which are designed for specific types of data
sources and the redundancies found in them but do a poor
job when applied to other sources of data. For example, the
Moving Pictures Experts Group (MPEG) compression standards
were designed to take advantage of the relatively
small difference from one frame to another in a video
stream and so do an excellent job of compressing motion pictures.
On the other hand, MPEG would not be effective if
applied to still images. For this data source, the Joint
Photographic Experts Group (JPEG) compression standards
would be applied. JPEG is “lossy,” meaning that the decompressed image is
not quite the same as the original compressed image—there
is some degradation. JPEG is designed to exploit known limitations
of the human eye, notably that small color details
are not perceived as well as small details of light and dark.
JPEG eliminates the unnecessary details to greatly reduce
the size of image files, allowing them to be transmitted
faster and take up less space in a storage server.
On wide area network (WAN) links, the compression ratio
tends to differ by application. The compression ratio can be
as high as 6 to 1 when the traffic consists of heavy-duty file
transfers. The compression ratio is less than 4 to 1 when the
traffic is mostly database queries. When there are only “keep
alive” signals or sporadic query traffic on a T1 line, the compression
ratio can dip below 2 to 1.
Encrypted data exhibit little or no compression because
the encryption process expands the data and uses more
bandwidth. However, if data expansion is detected and compression
is withheld until the encrypted data are completely
transmitted, the need for more bandwidth can be avoided.
Types of Data Compression
There are several different data-compression methods in use
today over WANs—among them are Transmission Control
Protocol/Internet Protocol (TCP/IP) header compression,
link compression, and multichannel payload compression.
Depending on the method used, there can be a significant
tradeoff between lower bandwidth consumption and
increased packet delay.
TCP/IP Header Compression With TCP/IP header compression,
the packet headers are compressed, but the data payload
remains unchanged. Since the TCP/IP header must be
replaced at each node for IP routing to be possible, this compression
method requires hop-by-hop compression and
decompression processing. This adds delay to each com-
76 DATA COMPRESSION
DATA COMPRESSION 77
pressed/decompressed packet and puts an added burden on
the router’s CPU at each network node.
TCP/IP header compression was designed for use on slow
serial links of 32 kbps or less and to produce a significant performance
impact. It needs highly interactive traffic with small
packet sizes. In such traffic, the ratio of Layer 3 and 4 headers
to payload is relatively high, so just shrinking the headers
can result in a substantial performance improvement.
Payload Compression Payload compression entails the
compression of the payload of a Layer 2 WAN protocol,
such as the Point-to-Point Protocol (PPP), Frame Relay,
High-Level Data Link Control (HDLC), X.25, and Link
Access Procedure–Balanced (LAPB). The Layer 2 packet
header is not compressed, but the entire contents of the
payload, including higher-layer protocol headers (i.e.,
TCP/IP), are compressed. They are compressed using the
industry standard Lemple-Ziv algorithm or some variation
of that algorithm.
Layer 2 payload compression applies the compression
algorithm to the entire frame payload, including the TCP/IP
headers. This method of compression is used on links operating
at speeds from 56 to 1.544 Mbps and is useful on all
traffic types as long as the traffic has not been compressed
previously by a higher-layer application. TCP/IP header
compression and Layer 2 payload compression, however,
should not be applied at the same time because it is redundant
and wasteful and could result in the link not coming up
to not passing IP traffic.
Link Compression With link compression, the entire frame—
both protocol header and payload—is compressed. This form
of compression is typically used in local area network
(LAN)–only or legacy-only environments. However, this
method requires error-correction and packet-sequencing
software, which adds to the processing overhead already
introduced by link compression and results in increased packet delays. Also, like TCP/IP header compression, link
compression requires hop-by-hop compression and decompression,
so processor loading and packet delays occur at
each router node the data traverses.
With link compression, a single data compression vocabulary
dictionary or history buffer is maintained for all virtual circuits
compressed over the WAN link. This buffer holds a
running history about what data have been transmitted to help
make future transmissions more efficient. To obtain optimal
compression ratios, the history buffer must be large, requiring
a significant amount of memory. The vocabulary dictionary
resets at the end of each frame. This technique offers lower
compression ratios than multichannel, multihistory buffer
(vocabulary) data-compression methods. This is particularly
true when transmitting mixed LAN and serial protocol traffic
over the WAN link and frame sizes are 2 kilobytes or less. This
translates into higher costs, but if more memory is added to get
better ratios, this increases the upfront cost of the solution.
Mixed-Channel Payload Data Compression By using separate
history buffers or vocabularies for each virtual circuit,
multichannel payload data compression can yield higher
compression ratios that require much less memory than
other data-compression methods. This is particularly true in
cases where mixed LAN and serial protocol traffic traverses
the network. Higher compression ratios translate into lower
WAN bandwidth requirements and greater cost savings.
But performance varies because vendors define payload
data compression differently. Some consider it to be compression
of everything that follows the IP header. However,
the IP header can be a significant number of bytes. For overall
compression to be effective, header compression must be
applied. This adds to the processing burden of the CPU and
increases packet delays.
External Data Compression Solutions Bridges and routers
can perform data compression with optional software or addDATA
COMPRESSION 79
on hardware modules. While compression can be implemented
via software, hardware-based compression off-loads
the bridge/router’s main processor to deliver even higher levels
of throughput. With a data-compression module, the compression
process can occur without as much processing delay
as a software solution.
The use of a separate digital signal processor (DSP) for
data compression, instead of the software-only approach,
enables the bridge/router to perform all its core functions
without any performance penalty. This parallel-processing
approach minimizes the packet delay that can occur when
the router’s CPU is forced to handle all these tasks by itself.
If there is no vacant slot in the bridge/router for the addition
of a data-compression module, there are two alternatives:
the software-only approach or an external compression
device. The software-only approach could bog down the overall
performance of the router, since its processor would be
used to implement compression in addition to core functions.
Although an external data compression device would not bog
down the router’s core functions, it means that one more
device must be provisioned and managed at each remote site.
Summary
Data compression will become increasingly important to
most organizations as the volume of data traffic at branch
locations begins to exceed the capacity of the wide area links
and as wireless services become available in the 2.4- and 5-
GHz range. Multichannel payload solutions provide the
highest compression ratios and reduce the number of packets
transmitted across the network. Reducing packet latency
can be effectively achieved via a dedicated processor like a
DSP and by employing end-to-end compression techniques
rather than node-to-node compression/decompression. All
these factors contribute to reducing bandwidth and equipment
costs as well as improving the network response time
for user applications.