« PrécédentContinuer »
(12) United States Patent (io) Patent No.: US 6,388,992 B2
Aubert et al. (45) Date of Patent: *May 14,2002
(54) FLOW CONTROL TECHNIQUE FOR TRAFFIC IN A HIGH SPEED PACKET SWITCHING NETWORK
(75) Inventors: Denis Aubert, Saint Paul de Vence;
Anne Chabannet, Le Bar sur Loup;
Jean-Jacques Legoll, Cagnes sur Mer;
Edmond Plattier, Antibes; Michel
Szczap, La Gaude; Rene Mariotti,
Nice, all of (FR)
(73) Assignee: Cisco Technology, Inc., San Jose, CA (US)
( * ) Notice: Subject to any disclaimer, the term of this patent is extended or adjusted under 35 U.S.C. 154(b) by 0 days.
This patent is subject to a terminal disclaimer.
(21) Appl. No.: 09/748,593
(22) Filed: Dec. 22, 2000
Related U.S. Application Data
(63) Continuation of application No. 08/935,758, filed on Sep. 23, 1997, now Pat. No. 6,167,027.
(30) Foreign Application Priority Data
Sep. 9, 1997 (EP) 97480060
(51) Int. CI.7 G01R 31/08
(52) U.S. CI 370/232; 370/236
(58) Field of Search 370/230, 232,
370/235, 236, 237, 389
5,359,593 A 10/1994 Derby et al 370/234
5,367,523 A * 11/1994 Chang et al 370/235
5,448,567 A 9/1995 Dighe et al 370/233
5,541,852 A 7/1996 Eyuboglu et al 709/232
5,596,576 A 1/1997 Milito 370/450
5,625,622 A 4/1997 Johri 370/232
5,790,521 A * 8/1998 Lee et al 370/230
5,815,492 A * 9/1998 Berthaud et al 370/234
6,118,791 A * 9/2000 Fichou et al 370/468
6,167,027 A * 12/2000 Aubert et al 370/230
Shaaban, Transport Flow Control, EECC694, May 11, 2000.
Rexford, Jennifer L., et al., A Fair Leaky-Bucket Traffic
Shaper for ATM Networks, 1994 or later (not dated).
* cited by examiner
Primary Examiner—Min Jung
(74) Attorney, Agent, or Firm—Cesari and McKenna, LLP; A. Sidney Johnston
A flow control mechanism for use in a high speed packet switching network where calls are multiplexed on network trunks with each connection using a reserved amount of the total bandwidth. Data terminal equipments access the network via access nodes. Each access node includes a Leaky Bucket component which maintains a refillable token pool. Each time an incoming packet is received by the leaky bucket component, the number of available tokens is compared to two predetermined threshold values. If the number of tokens is less than the low threshold, acknowledgments of received packets are stopped, inducing an interruption of packets transmitted by the emitting terminals. Interrupting packet transmission will lead to a regeneration of the number of tokens in the token pool. If the number of tokens reaches the high threshold, acknowledgments are again generated to restore packet transmissions. The first and second thresholds are greater than zero, reducing the chances that a packet will be marked as being discardable.
16 Claims, 2 Drawing Sheets
FLOW CONTROL TECHNIQUE FOR
TRAFFIC IN A HIGH SPEED PACKET
CROSS-REFERENCE TO RELATED 5
The present application is a continuation of U.S. patent application Ser. No. 08/935,758, which was filed on Sep. 23, 1997, by Aubert et al. for a "Flow Control Technique for X.25 Traffic in a High Speed Packet Switching Network", issued on Dec. 26, 2000, U.S. Pat. No. 6,167,027.
The present invention relates to high speed packet switch- 15 ing networks and more particularly to a mechanism for controlling X.25 traffic data flow at an access node of a high speed packet switching network.
BACKGROUND ART 20
Major users have many disparate networks such as SNA, DECnet, TCP/IP and X.25 data networks as well as logical networks of fax machines and one or more voice networks. For management reasons users want to have only one 2J network for all traffic. The emergence of high speed/ bandwidth networking technologies, such as ATM cellbased technology, now makes possible the integration of multiple types of traffic like speech, video and data over the same communication network. The communication circuits 3Q which may be shared in such network include transmission lines, program controlled processors, nodes or links, and data or packets buffers. In such network an access node should thus be designed for supporting the access through users' existing communications equipment such as X.25 3J devices as well as for accommodating new equipment, such as equipment supporting ATM.
The present invention relates to access nodes in high speed cell switching networks supporting the attachment of devices using different protocols and particularly X.25 40 devices. In such networks, access nodes perform a set of functions known collectively as bandwidth management and congestion control, the role of which is to manage the network's resources (bandwidth in particular) such that the overall resource utilization is maximized while continuing to 45 satisfy user requests for performance guarantees, usually referred to as quality of service or QOS. Congestion control is based on bandwidth reservation at the network connection. For each connection requiring a specific quality of service according to the protocol used, a certain amount of 50 bandwidth is reserved on the links traversed by the connection. The bandwidth reservation contract is enforced by specific techniques such as the well-known leaky bucket algorithm.
In classical packet switching networks, flow control algo- 55 rithms are based on indowing and buffer depletion techniques. A windowing scheme consists of defining packet credits (often referred to as a window). A credit represents the number of packets that an origin device, also referred to as Data Terminal Equipment or DTE, can send across the 60 network without receiving an acknowledgment either from the network or from the destination device. This credit mechanism prevents the origin DTE from sending data to the destination DTE when the network or the destination DTE cannot reliably expect to receive such data, e.g. when 65 the network is congested. Under network congestion conditions, no acknowledgment will be received by the
origin DTE, which will stop sending data as soon as packets already in its current window have been transmitted, i.e., when its credit has been decremented to zero.
Buffer depletion schemes rely on monitoring of access node buffers which are used for storing packets in anticipation of such routing or while awaiting availability of an output link. The principle is to define two thresholds, which correspond to percentages of the total amount of buffer space available, and two corresponding states: congested and not congested. Buffer depletion techniques are often associated with windowing schemes.
More specifically, with regard to the X.25 protocol, the X.25 ITU-T recommendation specifies the windowing mechanism mentioned above as a flow control technique. This technique is currently implemented at LAP-B (Link Access Protocol-Balanced) level and at PLP (Link Access Protocol) level. The general principle of a common implementation relies on the existence of three windows which are defined in a X.25 virtual circuit. A first window (origin window) is associated with the interface between the origin X.25 DTE and the input access node of the network. A second window (destination window) is dedicated to the interface between the output access node of the network and the destination X.25 DTE. A third window (internal window) is associated with the internal part of the network, that is, all network elements on the path between the input and output access nodes. In operation, the destination DTE indicates it is congesting by not sending acknowledgments of received packets. When the output network access node does not receive expected acknowledgments, it sends packets only until its existing credit (destination window) is depleted. Similarly, when the input access node fails to receive the expected acknowledgments, it sends packets only until its own credit (internal window) is depleted and then stops transmitting packets into the network. Finally, an origin DTE which does not receive the expected acknowledgments sends packets only until its existing credit (origin window) is depleted. This windowing mechanism can be associated with buffer depletion techniques that allow detection of the existence network congestion. A common buffer depletion technique consists of computing the amount of buffer space used by LAP-B frames which have entered the network access node and which have not yet been handled, and then comparing this number to predefined low/high buffer utilization thresholds. The LAP-B acknowledgments are stopped when the utilization of buffers exceeds the high threshold, and re-started when buffer utilization falls below the low threshold; this scheme is intended to avoid network congestion.
The above-cited X.25 common flow control technique is quite satisfactory for classical packet switching networks. But when X.25 traffic is routed through high speed digital networks using bandwidth reservation, such ATM or Frame Relay backbones networks, the X.25 common flow control scheme is not efficient enough to avoid network congestion. Indeed, in such networks an access node can support a large number of X.25 data terminal equipments (e.g. 4096 DTEs). Consequently, the total buffer size that would be classically be required for avoiding congestion is the sum of the packet window sizes required for all the logical channels entering the node, which correspond to the attached DTEs. For example, for 4096 windows, each including 7 packets of 1048 bytes, an access node would need a buffer size of 4096*7*1048 bytes=30 megabytes). Thus, a significant number of buffers would be required for this common technique to be efficient, which would increase significantly the cost of the system since buffer memories, which use high-speed, high-performance technologies, are expensive.