Electrical computers and digital processing systems: multicomput – Computer network managing – Computer network monitoring
Reexamination Certificate
1997-06-18
2001-01-23
Rinehart, Mark H. (Department: 2756)
Electrical computers and digital processing systems: multicomput
Computer network managing
Computer network monitoring
C709S232000, C709S233000, C370S468000
Reexamination Certificate
active
06178448
ABSTRACT:
FIELD OF THE INVENTION
The present invention relates to communications networks and more particularly to communications networks having multiple physical links, paths, connections or virtual circuits between two nodes.
BACKGROUND OF THE INVENTION
In recent years there has been a proliferation in the networking of computer systems. The recent expansion of the Internet is just one example of the trend toward distributed computing and information sharing. In most forms of computer or communication networking there are communication paths between the computers in the networks. These paths may include multiple links or hops between intermediate equipment in a path. Thus, a communication may be originated by a first computer and pass through several links before reaching the destination computer. The control over these communications is typically carried out under a networking architecture. Many networking architectures exist for defining communications between computers in a network. For example, System Network Architecture (SNA) and Transfer Control Protocol/Internet Protocol (TCP/IP) are two examples of existing network architectures.
One existing network architecture for controlling communications between computers is known as Advanced Peer to Peer Networking (APPN). APPN, like many networking architectures, is based upon the transmission of data packets where a communication is broken into one or more “packets” of data which are then transmitted from the source to the destination over the communication path. Packet based communications allows for error recovery of less than an entire communication which improves communication reliability and allows for packets to take multiple paths to an end destination thus improving communication availability.
One error condition which many networks attempt to correct for is packet loss. Packet loss in a network may be broadly characterized as resulting from congestion on the path from the source to the destination or from loss of data (bit error) by links in the path. Congestion may result from too high a data packet rate for a path. Bit error may, however, result from any number of failures in a communication link. For example, sun spots may adversely impact microwave transmissions and cause loss of data. However, bit error occurrences are generally highly correlated. As a result, a time averaged bit error rate (BER) alone may not accurately describe line quality. Line quality is, therefore, usually described using a combination of an average BER over some time period along with the number of seconds in the time period in which one or more bit errors occur.
While APPN has proven to be a reliable networking architecture, as computer networking demands have increased these demands have created a demand for network architectures which utilize the higher performance communication systems and computer systems currently available. In part because of these demands, High Performance Routing, which is an enhancement to APPN, was developed. Processing capability has increased and become less expensive. This has driven the need for larger peer-to-peer networks. Link technology has advanced by several orders of magnitude over the past decade. Advances in wide area links have dramatically increased transmission rates and decreased error rates. Thus, to take advantage of these advances HPR provides high speed data routing which includes end-to-end recovery (i.e. error recovery is performed by the sending and receiving systems) and end-to-end flow and congestion control where the flow of data is controlled by the sending and receiving systems.
HPR consists of two main components: the Rapid Transport Protocol (RTP) and automatic network routing (ANR). RTP is a connection-oriented, full-duplex transport protocol designed to support high speed networks. One feature of RTP is to provide end-to-end error recovery, with optional link level recovery. RTP also provides end-to-end flow/congestion control. Unlike TCP's reactive congestion control, RTP provides an adaptive rate based mechanism (ARB).
ARB provides end-to-end flow control to prevent buffer overrun at the RTP endpoints, a rate based transmission mechanism that smooths input traffic and a preventive congestion control mechanism that detects the onset of congestion and reduces the RTP send rate until the congestion has cleared. The ARB preventive congestion control mechanism attempts to operate the network at a point below the “cliff” (shown in
FIG. 1
) and to prevent congestion. A reactive mechanism, on the other hand, detects when the network has entered the region of congestion and reacts by reducing the offered load.
In RTP, the ARB mechanism is implemented at the endpoints of an RTP connection. Each endpoint has an ARB sender and an ARB receiver. The ARB sender periodically queries the receiver by sending a rate request to the ARB receiver who responds with a rate reply message. The sender adjusts its send rate based on information received in the rate reply message.
The mechanism used to control the send_rate is as follows. A burst_size parameter sets the maximum number of bytes a sender can send in a given burst at a given send_rate. During each burst_time, defined by burst_size/send_rate, a sender is allowed to send a maximum of burst_size bytes. The receiver continuously monitors network queuing delay looking for the initial stages of congestion. Based on this assessment and also based on the current state of the receiver's buffers, the receiver sends a message to the sender instructing it to either increment the send_rate by a rate increment, keep the send_rate the same, decrement the send_rate by 12.5%, decrement the send_rate by 25%, or decrement the send_rate by 50%.
The receiver initiates error recovery as soon as it detects an out of sequence packet by sending a gap detect message that identifies the packets that need to be resent. When the sender receives a gap detect message, it drops its send_rate by 50% and resends the packets at the next send opportunity. If the sender does not get a response to a rate request within a time-out period, the sender assumes the packet is lost and cuts the send_rate by half, increases the rate request time-out exponentially (exponential back off), and transmits a rate request at the next send opportunity.
Thus, like many forms of networking, in RTP packet losses are assumed to result from congestion rather than bit errors. Such an assumption may often be valid for modern digital wide area links which exhibit low loss rates. However, these loss rates may not apply to all communication links around the world or even to high quality links all the time.
Furthermore, as RTP provides end-to-end flow control, the send rate of packets on a path may be limited by the slowest link in the path (i.e., the bottle-neck link). Thus, despite a path having high-speed links in the path if a single low-speed link is present, the sender and receiver will pace the transmission of packets to accommodate the low speed link. Thus, a congestion problem or the presence of one low speed link in a path may degrade the throughput for the entire path.
One way to improve congestion problems or to compensate for differing transmission rates on a communications path is to provide for multiple links between connection points that may be the bottle-neck in the path. HPR provides for such concurrent links through a Multilink Transmission Group (MLTG). Similarly, TCP/IP provides for multiple links with multi-link Point to Point Protocol (PPP). A transmission group is a logical group of one or more links between adjacent nodes that appears as a single path to the routing layer. A MLTG is a transmission group that includes more than one link. Links in a MLTG are referred to herein as sublinks. An MLTG can include any combination of link types (e.g., token-ring, SDLC, frame relay). MLTGs provide increased bandwidth which may be added or deleted incrementally on demand. Furthermore, the combined full bandwidth is available to a session since session traffic can flow over all sublink
Gray James P.
Martin James J.
Cardone Jason D.
International Business Machines - Corporation
Myers Bigel & Sibley & Sajovec
Ray-Yarletts Jeanini S.
Rinehart Mark H.
LandOfFree
Optimal link scheduling for multiple links by obtaining and... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Optimal link scheduling for multiple links by obtaining and..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Optimal link scheduling for multiple links by obtaining and... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2557137