Creating frames , Pipelining and Reliable Transmission

What is a requirement for framing to even work in the data link layer?

• Recipient must somehow recognize start & end of frame

What are old or rather unsuitable methods that can be used for framing?

• Time of non-sending (=pause time)

• TLV (Type - Length - Value); today used on higher levels, not L2

• Frame tags: specific start/end markers

What are other technical means that can be employed for framing?

• Fixed redundancy for error detection/correction

  • 4B/5B, Manchester Encoding, ...

• Cell synchronization of data via checksum

  • Move starting points until checksum matches

Why are periods of "non-sending" usually not a good approach for marking the start and end of frames?

• Gaps can disappear due to time delays

What is frame tagging all about?

• Use of start and end characters to mark the boundaries of a frame

  • There are dedicated ascii characters to distinguish them from data (DLE, STX, ETX):
    

    

  • If data contains these dedicated characters -> they are escaped/stuffed

Which two variations of bit errors exist?

• single errors

• error bursts

Wie funktioniert parity?

• Codewort mit m Datenbits und r parity bits für error detection

• even = popcount gerade

• odd = popcount ungerade

What kinds of error detection / correction methods are there?

• Parity (extra bit(s) toggled such that group of bits is even/odd), with enough parity bits, can be used for correction
  

  

• CRC (cyclic redundancy check) complex checksums for error detection

Why is error correction via row and column parity generally not a good idea?

• 16 bits + 9 parity bits -> HUGE overhead

• No guarantee that the error is actually corrected

  • problematic when there are >= 3 bit errors

What is the hamming distance d(w1, w2) of two words w1 and w2?

• Number of bits in which two code words differ

  • Obviously d >= 1 for w1 != w2

What is the hamming distance of a code?

• Minimal d(w1, w2) over all different code word combination pairs of the code

Which hamming distance d is necessary to detect f single bit errors?

• d = f + 1

Which hamming distance d is necessary to correct f single bit errors?

• d = 2*f + 1

What is pipelining in the context of reliable transmissions?

Sending multiple data frames without waiting for ACK

What problem does pipelining solve?

Waiting for ACK on routes with long response times

What are the consequences of pipelining?

1. More memory for buffers -> keep, as long as frame is not acknowledged

2. Does not solve reaction time problem, only data rate increases

3. Only useful for a single stream

4. What if a frame is lost?

What is always necessary when something is transmitted incorrectly or lost?

Retransmission of the data frames in question

When is error detection and retransmission preferrable on L2?

Environments with low BER and minimal RTT

When is error correction + detection and retransmissino prefferable on L2?

Links with higher BER and relatively high RTT