AccuracyThe overall goal for accura

Accuracy
The overall goal for accuracy is that the data received at the destination must be the same
as the data sent by the source. Typical causes of data errors include power surges or
spikes, impedance mismatch problems, poor physical connections, failing devices, and
noise caused by electrical machinery. Sometimes software bugs can cause data errors
also, although software problems are a less common cause of errors than physical layer
problems. Frames that have an error must be retransmitted, which has a negative effect on
throughput. In the case of IP networks, Transmission Control Protocol (TCP) provides
retransmission of data.
For WAN links, accuracy goals can be specified as a bit error rate (BER) threshold. If the
error rate goes above the specified BER, the accuracy is considered unacceptable. Analog
links have a typical BER threshold of about 1 in 10
5
. Digital circuits have a much lower
error rate than analog circuits, especially if fiber-optic cable is used. Fiber-optic links
have an error rate of about 1 in 10
11
. Copper links have an error rate of about 1 in 10
6
.
For LANs, a BER is not usually specified, mainly because measuring tools such as protocol analyzers focus on frames, not bits; however, you can approximate a BER by comparing the number of frames with errors in them to the total number of bytes seen by the
measuring tool. A good threshold to use is that there should not be more than one bad
frame per 10
6
bytes of data.
On shared Ethernet, errors are often the result of collisions. Two stations try to send a
frame at the same time and the resulting collision damages the frames, causing cyclic
redundancy check (CRC) errors. Depending on the size of the Ethernet network, many of
these collisions happen in the 8-byte preamble of the frames and are not registered by
troubleshooting tools. If the collision happens past the preamble and somewhere in the
first 64 bytes of the data frame, this is registered as a legal collision, and the frame is
called a runt frame. A general goal for Ethernet collisions is that less than 0.1 percent of
the frames should be affected by a legal collision (not counting the collisions that happen
in the preamble).
A collision that happens beyond the first 64 bytes of a frame is a late collision. Late collisions are illegal and should never happen. Ethernet networks that are too large experience late collisions because stations sending minimum-sized frames cannot hear other
stations within the allowed timeframe. The extra propagation delay caused by the excessive size of the network causes late collisions between the most widely separated nodes.
Faulty repeaters and network interface cards (NIC) can also cause late collisions.
Collisions should never occur on full-duplex Ethernet links. If they do, there’s probably a
duplex mismatch. Collisions on a properly configured full-duplex link have no meaning.
Both stations sending at the same time is normal. Receiving while sending is normal. So,
Chapter 2: Analyzing Technical Goals and Tradeoffs 39
there is no need for collision detection and collisions shouldn’t occur. Chapter 3 has more
to say about duplex mismatch problems and how to recognize if they cause errors on
your networks.
Collisions also never occur on WAN links. Unfortunately, the output of the show interface serialcommand on Cisco routers includes a collision count. It should be ignored.
Cisco programmers used a template for this part of the output. The template is based on
the output from the show interface ethernetcommand. There are no collisions on a serial
interface, regardless of the encapsulation or technology. Collisions occur only on carrier
sense multiple access (CSMA) networks including Ethernet, 802.3, LocalTalk, Aloha, and
802.11 networks. Collisions are a normal part of the “management-by-contention”
approach that defines CSMA. (And although LocalTalk and 802.11 use CSMA with collision avoidance, collisions can still occur.)
Accuracyusually refers to the number of error-free frames transmitted relative to the
total number of frames transmitted. Accuracy can also characterize how often the network reorders sequences of packets. Packet reordering occurs in many situations, including the use of parallel switching fabrics within a single network device and the use of parallel links between routers. Although upper-layer protocols, such as TCP and Real-Time
Transport Protocol (RTP), correct for the reordering of packets, the problem can cause
minor performance degradation. Some applications don’t use a protocol that corrects the
problem and thus might be more severely affected. Because the problem is often corrected, it can be hard to detect. IP routers are not designed to detect, let alone correct, packet reordering, and because they do not detect this condition, they cannot report the
problem to network management software. Measurements must be made at end hosts.
For example, you could use a protocol analyzer on an end-station host to detec