Beschreibung der möglichen Fehler, die im Ethernet auftreten können.

• Transmit Deferrals: is the number of times the NIC deferred to another transmitting node while waiting to transmit. This number increases as other stations contend with this NIC for access to the Eternet network.
• Recieve Overrun is the number of packets that cannot be stored in the recieve buffer because the PC's buffer is already full.
• Late Collisions occur, when another node on the Ethernet network does not properly defer once the NIC has started transmitting.
• Carrier Sense Lost is the number of packets transmitted with carrier sense loss. This normally occurs as a result of collisions. For more information, refer to an Ethernet reference guide.
• SQE Errors the number of times that the SQE signal was detected during transmission from the PCs NIC. This count should be either zero or the total number of packets transmitted. Any other number indicates that the NIC transceiver is malfunctioning.
• Multiple Collisions are a normal occurence on an Ethernet network; however, excessive collisions may indicate problems.
• Single Collisions are the number of packets that had one collision during transmission from the NIC.
• CRC Errors: Each packet is sent on an Ethernet network has a calculated cyclic redundancy check (CRC) appended to it. When the packet is recieved, this CRC is compared with the calculated CRC. If the calculated CRC is different from the CRC, the packet was corrupted, most likely by line noise.
• Excessive Collisions may indicate network or cabling problems.
• Transmit Underruns: Number of times the NIC detected an underrun condition, that is, the number of packets received were less than the number required in the buffer. Transmit underrun is the number of times a packet was transmitted without adequate data by NIC.
  A count of the number of packets that were not successfully transmitted because data coming into the adapter could not be accumulated fast enough to maintain transmission to the Ethernet media. This error results in a CRC error at the receiving station. The transmit is always retried automatically by the driver.
• Frame Alignment Errors are the number of alignment errors for the packets recieved by the NIC. An alignment error is caused when an incoming packet does not end on a byte boundary and the CRC does not match at the last byte boundary.

What is a collision, and how many collisions are bad?

Ethernet networking uses collisions as one of the contention access methods. When the network carrier is not active, any station can send information. If two stations attempt to send information at the same time, the signals overlap with each other, creating a collision. Collisions are not errors! Many people misinterpret a flashing collision light or a collision counter as a network problem! Although the term 'collision' may bring to mind a terrible crash, be assured that a collision is a normal part of Ethernet networking. The total number of collisions that occur on a network may be related to traffic patterns or utilization. Because of this variability of collisions, it is not applicable to define a 'good' or 'bad' level of collisions. In most cases, detailed analysis of collisions alone yields very little qualitative network health information.

What is the Signal Quality Error (SQE) Test?

The SQE Test is used to test for the collision present circuit between a transceiver and a network interface card (NIC). After data is successfully transmitted, the Ethernet transceiver asserts the SQE signal on the collision presence circuit of the NIC. The NIC sees this test signal as a verification that the transceiver will inform the NIC when a collision occurs. In most modern Ethernet networks, the SQE test is not used or applicable. Most NICs now have an integrated transceiver and therefore have a hard-wired AUI, so a test for the collision presence circuit is unnecessary.

What is jam?

When a collision is recognized by a transmitting station, a bit sequence called jam is transmitted. This jam is 32 bits long, which is long enough to traverse the entire collision domain so that all transmitting stations can detect the collision. Interestingly enough, the actual format of jam is unspecified in the 802.3 specifications. Most manufacturers have used alternating 1s and 0s as jam, which is displayed as 0x5 (0101) or 0xA (1010) depending on when the jam is captured in the data stream. In many Fast Ethernet implementations, the jam has been seen as other arbitrary values, such as 1101000 (0xD0) or 10000110 (0x43). The reasoning for this particular jam pattern isn't very obvious.

What is a late collision, and why is it bad?

A collision is considered late if the jam occurs after 512 bit-times, or 64 bytes. Collisions that occur after the first 64 bytes of a frame may be indicative of a network design problem (the network is so large the jam cannot traverse the entire length in 32 bit-times), or a hardware or Ethernet firmware issue. When collisions do not propagate the network quickly enough, a collision could occur between two stations without the stations aware that the packets collided. In this situation, the frames are simply lost, and the upper-layer protocols must begin a retransmission process to retransmit the information. These retransmissions can cause large delays, especially at the application layer.

Out-of-window (Late)Collisions occur when a station receives a collision signal while still transmitting, but more than 51.2 micro seconds (the maximum ethernet propagation delay) after transmission began. The two conditions which cause this type of error to occur: either the network's physical length exceeds IEEE 802.3 specifications, or a node on the net is transmitting without first listening for carrier sense (and beginning its illegal transmission more than 51.2 micro seconds after the first station began transmitting

What is a runt?

In Ethernet networks, any frame shorter than the minimum 64 bytes but with a valid CRC is considered a runt. Other frame-length errors in Ethernet called Giants are long frames, which are longer than 1518 bytes yet have a valid CRC.

A runt packet is one that is smaller that the minimum Ethernet frame size of 64 bytes (excluding preamble). This minimum size is tied to the maximum propagation time of an ethernet network segment; the maximum propagation time is 51.2 micro seconds, and it takes approximately 51.2 micro seconds to transmit 64 bytes of data. Therefore, every node on the segment should be aware that another node is transmitting before the transmission is complete, providing for more accurate collision detection.
runts can sometimes result from collisions, and as such, may be the natural by-product of a busy ethernet; however, they can also indicate a hardware (packet formation), transmission (corrupted data), or network design (more than four cascaded repeaters) problem.

As mentioned in the error statistics descriptions above, each Cabletron device employs an error priority scheme which determines how packets with multiple errors will be counted, and ensure that no error packet is counted more than once.

What is a CRC/Alignment error?

When a station sends a frame, it appends a Cyclical Redundancy Check to the end of the frame. This CRC has been generated from an algorithm and is based on the data in the frame. If the frame is altered between the source and destination, the receiving station will recognize that the CRC does not match the actual contents of the packet. All frames should end on an 8-bit boundary, but problems on the network could cause the number of bits to deviate from the multiple of 8. Both CRC errors and alignment errors are grouped together as the single CRC/Alignment error counter.
CRC, or cyclic redundancy check, errors occur when the packets are somehow damaged in transit. When each packet is transmitted, the MAC layer of the transmitting device computes a frame check sequence (FCS) value based on the content of the packet. The receiving station performs the same calculation; if the FCS values differ, the packet is assumed to have been corrupted and is counted as a CRC error. CRC errors can result from MAC layer hardware problem causing an inaccurate computation of the FCS value, or from other transmission problem that has garbled the original data. On most Cabletron devices, the error priority scheme is set up so that packets which are counted as CRC errors do not have alignment errors as well.
'misaligned packets are those which contain any unit of bits which is less than a byte. Misaligned packets can result from a MAC layer packet formation problem, or from some transmission medium (cabling) problem that is corrupting or losing data; they can also result from packets passing through more that two cascaded multi-port transceivers (a network design that does not meet the ethernet spec). Alignment packets typically also have CRC errors, although the priority scheme employed in most Cabletron devices does not distinguish between those that do and those that don't.

Giants / JabberGiants

are packets which are longer that the maximum ethernet size of 1518 bytes (excluding preamble). Giant packets usually occur when you have a jabbering node on your network. -one that is continuously transmitting, or transmitting improperly for short bursts-probably due to a bad transmitter on the NIC. Giants can also be caused by packets being corrupted as they are transmitted, either by the addition of garbage signals, or by the corruption of the bits that indicate frame size.
Giants are packets which are longer that the maximum ethernet size of 1518 bytes (excluding preamble). Giant packets usually occur when you have a jabbering node on your network. -one that is continuously transmitting, or transmitting improperly for short bursts-probably due to a bad transmitter on the NIC. Giants can also be caused by packets being corrupted as they are transmitted, either by the addition of garbage signals, or by the corruption of the bits that indicate frame size.