rcallon@PARK-STREET.BBN.COM (Ross Callon) (09/09/87)
In the ISO IP, the DEC bit is located in an options field. In particular, the "Quality of Service Maintenance" option contains three possible classes of encodings, one of which is called "globally unique". One bit of this globally unique QOS option is the congestion experience bit. The ISO spec explicitly specifies which functions must be implemented, and which are optionally implemented. Support for the QOS option is not required. Thus not only is the presence of the DEC bit optional, but even if it is there gateways are not required to look at it. In addition, "Congestion Notification" is also optional. Even if the source uses the globally unique QOS option, and the gateways update the DEC bit appropriately, it is not required (at least by IS 8473) that the destination will notify the Transport Implementation. Clearly, optional implementation of the congestion bit is a problem, especially since if some end systems pay attention to it and others ignore it the "good guys" will end up suffering. This may be overcome by requiring use of the DEC bit in implementers agreements (such as produced by the COS). The placement of the bit in an option field means that if you want to use it for all packets you are requiring the gateways to actually parse the options fields for every packet. This increases the processing demand on gateways. This may be okay for low bandwidth internets, but gets more undesirable for higher bandwidths (what constitutes higher bandwidths depends upon how much processing power you are willing to put in your gateways). I don't think the DEC bit has actually been used in an operational environment (does anyone out there know differently?). What we see in the Arpanet seems to be a relatively large number of short and/or discontinuous flows such as electronic mail and telnet traffic. Transport based flow control methods may be more appropriate for a smaller number of longer term flows. The combination of slow start (which is part of the way that the DEC bit is to be used) plus regulating the growth rate of flows on each connection, will prevent short term connections from "jumping in" with a large increment of traffic all at once. However, for very short lived connections, this may be nearly equivalent to simply using a fixed very small window size. This approach also assumes that all users (or at least the users associated with the overwhelming majority of network traffic) will use a window based transport protocol. All of this suggests that the DEC bit is more an interesting research approach for use in some common environments, rather than a proven technique. Ross