RFC 1110 (rfc1110) - Page 2 of 3
Problem with the TCP big window option
Alternative Format: Original Text Document
RFC 1110 Comments on TCP Big Window Option August 1989 is contained in a 32-bit field and therefore "wraps around" after the transmission of 2**32 bytes of data. Two additional mechanisms are used to insure the effective uniqueness of sequence numbers; these are the TCP transmission window and bounds on packet lifetime within the Internet, including the IP Time-to-Live (TTL). The transmission window specifies the maximum number of bytes which may be sent by the source in one source-destination roundtrip time. Since the TCP transmission window is specified by 16 bits, which is 1/65536 of the sequence number space, a sequence number will not be reused (used to number another byte) for 65,536 roundtrip times. So long as the combination of gateway action on the IP TTL and holding times within the individual networks which interconnect the gateways do not allow a packet's lifetime to exceed 65,536 roundtrip times, each sequence number is effectively unique. It was believed by the TCP designers that the networks and gateways forming the internet would meet this constraint, and such has been the case. The proposed TCP Big Window option, as described in RFC 1106, expands the size of the window specification to 30 bits, while leaving the sequence number space unchanged. Thus, a sequence number can be reused after 4 roundtrip times. Further, the Nak option allows a packet to be retransmitted (i.e., potentially duplicated) by the source after only one roundtrip time. Thus, if a packet becomes "lost" in the Internet for only about 5 roundtrip times it may be delivered when its sequence number again lies within the window, albeit a later cycle of the window. In this case, TCP will not necessarily recreate at the destination an exact copy of the data stream generated at the source; it may replace some data with earlier data. Of course, the problem described above results from the storage of the "lost" packet within the net, and its subsequent out-of-order delivery. RFC 1106 seems to describe use of the proposed options in an isolated satellite network. We may hypothesize that this network is memoryless, and thus cannot deliver packets out of order; it either delivers a packet in order or loses it. If this is the case, then there is no problem with the proposed options. The Internet, however, can deliver packets out of order, and this will likely continue to be true even if gigabit links become part of the Internet. Therefore, the approach described in RFC 1106 cannot be adopted for general Internet use. McKenzie
