How Stop and Wait Protocols Can Fail over the Internet

Abstract

Abstract. The correct operation of computer protocols is essential to the smooth operation of the distributed systems that facilitate our global economy. Formal techniques provide our best chance to ensure that protocol designs are free from errors. This invited paper revisits the class of Stop-and-Wait protocols that incorporate retransmission strategies to recover from transmission errors. This is motivated by the fact that their basic mechanisms are important for practical protocols such as the Internet's Transmission Control Protocol (TCP). Stop-and-Wait protocols have been shown to operate correctly over media that may lose packets, however, there has been little discussion regarding the operation of these protocols over media that can re-order packets. The paper presents an investigation of these protocols operating over a medium, such as that provided by the Internet Protocol, that does allow reordering of data. Coloured Petri Nets are used to build a model of a Stop-and-Wait Protocol parameterized by its maximum sequence number and the maximum value of the retransmission counter. The model is analysed using a combination of hand proofs and automatic techniques. We identify four problems. We firstly prove the counter intuitive property for a Stop-andWait protocol that the number of packets that are stored in the network can grow without bound. This is true for any positive values of the maximum sequence number and the maximum number of retransmissions. We further show that loss of packets is possible and that duplicates can be accepted as new packets by the receiver. These first three properties hold even though the sender and receiver perceive that the protocol is operating correctly. The final problem is that the protocol does not satisfy the Stop-and-Wait service where sends and receives alternate. Finally, we provide a discussion of the relevance of these results to the Transmission Control Protocol.

Transducer-Based Algorithmic Verification of Retransmission Protocols over Noisy Channels

“…Higher-order logic specifications of the protocols in the language of PVS are designed in [31]. Colored petri net models are used in [10] for modeling of stopand-wait protocols. In [15], timed state machines are used for modeling sliding window protocols.…”

“…Similar to most of the above approaches, we consider non-duplicating FIFO channels. The approaches [33,10] permit channels to re-order messages, whereas, [28,15] permit both re-ordering and duplication. Most of the above approaches assume lossy channels and do not model message contents.…”

An investigation of credit-based flow control protocols

“…The combination of abstraction techniques and model-checking allowed the sliding window protocol to be verified for a relatively large window size of 16, using the SPIN model checker [13]. Billington and Gallasch [1] showed how stop and wait protocols can fail over reordering channels and later verified a parameterised model of the class of Stop-and-wait protocols operating over lossy in-order channels by obtaining an explicit algebraic expression for the infinite set of state spaces as a function of the different parameters (retransmission limit and maximum se- 6 7 ] Free Buffers=2 [ 8 9 ] 5.…”

“…If we examine the first marking of place Pkt_Channel, when the list is at its maximum for the case when MaxRetrans=2, MaxBufSize=2 and MaxSeqNo=7, we obtain the following result (assuming the first packet to be transmitted has SN=0): 0 1 0 1 1 1 2 3 2 3 2 3] Examining the structure of this result we can see that the first full window (01) has been retransmitted twice, then (1) has been retransmitted twice, given the successful receipt and acknowledgement of (01). This then allows the next window (23) to be transmitted and then retransmitted twice (MaxRetrans=2).…”