Embed Doc
Copy Link
Readcast
Collections

CommentGo Back

Download

Edwin W. Meyer, Jr.MIT Project MAC27 June 1970The method of flow control described in RFC 54, prior allocation ofbuffer space by the use of ALL network commands, has one particularadvantage. If no more than 100% of an NCP's buffer space is allocated,the situation in which more messages are presented to a HOST then it canhandle will never arise.However, this scheme has very serious disadvantages:(i) chronic underutilization of resources,(ii) highly restricted bandwidth,(iii)considerable overhead under normal operation,(iv) insufficient flexibility under conditions of increasing load,(v) it optimizes for the wrong set of conditions, and(vi) the scheme breaks down because of message length indeterminacy.Several people from Project MAC and Lincoln Laboratories have discussedthis topic, and we feel that the "cease on link" flow control schemeproposed in RFC 35 by UCLA is greatly preferable to this new plan forflow control.The method of flow control proposed in RFC 46, using BLK and RSM controlmessages, has been abandoned because it can not guarantee to quench flowwithin a limited number of messages.The advantages of "cease on link" to the fixed allocation proposal arethat:(i) it permits greater utilization of resources,(ii) does not arbitrarily limit transmission bandwidth,(iii)is highly flexible under conditions of changing load,(iv) imposes no overhead on normal operation, and(v) optimizes for the situations that most often occur.Its single disadvantage is that under rare circumstances an NCP's inputbuffers can become temporarily overloaded. This should not be a seriousdrawback for network operation.

The "cease on link" method of flow control operates in the following[Page 1]

NWG/RFC 59 Flow Control - Fixed Versus Demand Allocationmanner. IMP messages for a particular "receive" link may be coming into the destination HOST faster than the attached process is reading themout of the NCP's buffers. At some point the NCP will decide that theinput queue for that link is too large in relation to the total amountof free NCP buffer space remaining. At this time the NCP initiatesquenching by sending a "cease on link" IMP message to its IMP. This doesnothing until the next message for that link comes in to the destinationIMP. The message still gets transmitted to the receiving HOST. However,the RFNM returned to the transmitting HOST has a special bit set. Thisindicates to the originating NCP that it should stop sending over thatlink. As a way of confirming the suspension, the NCP sends an SPD <link>"suspended" NCP control message to the receiving HOST, telling it thatit indeed has stopped transmitting. At a future time the receiving pro-cess will have cut the input queue for the link down to reasonable size,and the NCP tells the sending NCP to begin sending messages by issuing aRSM <link> "resume" NCP control message.The flow control argument is based on the following premises:(1) Most network transmission falls into two categories:Type 1 - short messages (<500 bits) at intervals of several secondsor more. (console communication)Type 2 - a limited number (10 - 100) of full messages (8000 bits)in rapid succession. (file transmission)(2) Most processes are ready to accept transmitted data when it arrivesat the destination and will pick it up within a few seconds (longerfor large files). Thus, at any particular instant the great major-ity of read links have no data buffered at the destination HOST.This assumes a sensible software system at both ends.(3) Flow control need be imposed only rarely on links transmitting Type1 messages, somewhat more frequently for Type 2 messages.(4) Both the total network load and that over a single connection fluc-tuate and can not be adequately predicted by either the NCP or theprocess attached to an individual connection.(5) Assuming adequate control of wide bandwidth transmission (Type 2),the probability that an NCP will be unable to accept messages fromthe IMP due to full buffers is quite small, even if the simultane-ous receipt of moderately small messages over all active linkswould more than fill the NCP's input buffers.(6) In the event that an NCP's buffers do fill completely, it mayrefuse to accept any transmission from the IMP for up to a minutewithout utter catastrophe.