Electrical computers and digital processing systems: multicomput – Ring computer networking
Reexamination Certificate
2000-04-20
2002-05-07
Geckil, Mehmet B. (Department: 2152)
Electrical computers and digital processing systems: multicomput
Ring computer networking
C709S249000, C709S238000, C370S902000, C370S222000, C370S223000
Reexamination Certificate
active
06385657
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
This invention is related to the field of computer systems and, more particularly, to interconnecting multiple microprocessors within a computer system.
2. Description of the Related Art
Computer systems have been achieving increased performance through the increasing performance of the microprocessors included therein and through including multiple microprocessors. As performance has increased, the applications to which these computer systems may be applied has increased as well. Applications which were formerly allocated to mainframe-style computers may now be performed using less expensive workstations. Furthermore, applications which were formerly unachievable in computer systems are now achievable. Continued advances in computer system performance are desirable to make yet additional applications achievable and to improve the efficiency at which current applications are performed.
As the microprocessors included in computer systems have increased in performance (through microarchitectural improvements and increased operating frequencies made possible by advances in semiconductor fabrication technologies), the bandwidth requirements of the microprocessors have increased as well. The increased numbers of instructions executed per second and the increased amount of data processed per second lead directly to bandwidth increases. Multiprocessor configurations increase the bandwidth requirements as a function of the number of processors included in the configuration.
Computer systems typically employ a bus structure for interconnecting microprocessors, memory, input/output (I/O) devices, and other features. The bus structure may be hierarchical, in which a bridge is included for conveying transactions between the hierarchical levels. Unfortunately, the bandwidth available from a bus structure is becoming insufficient for serving the requirements of modern microprocessors, memory, I/O devices, etc. Generally, a bus structure comprises a shared set of communication lines (the “bus”) at each level of the bus hierarchy. The devices provided at each level attach to the bus. Access to the bus is typically controlled through an arbitration scheme. For example, a round-robin scheme may be used in which each requesting device is eventually allowed an opportunity to control the bus. Alternatively, a fixed priority scheme may be used in which the highest priority requester is allowed to control the bus. Unfortunately, the time elapsing during arbitration for the bus increases the latency of the bus for any given requester. The requester must arbitrate for control of the bus before transmitting a transaction upon the bus.
Additionally, because the bus at any level in the hierarchy is a shared resource, only one device may initiate a transaction at any given time. Therefore, the bandwidth available to any given bus master is a fraction of the total bandwidth available on the bus. The fraction depends upon how frequently the bus is granted to the given master as compared to other masters on the bus. Yet another problem associated with bus structures is the need for large queues for bus transactions, particularly in the bridge devices between bus levels. If a bus transaction is presented and the queue within the receiver of the transaction is full, then the transaction must be retried back to the master of the transaction. The master must then attempt the transaction again later. If the master is a bridge device, it must retain the queue position for the transaction until it is successfully presented upon the target bus. In order to reduce the number of times the queue is full, a relatively large number of queue positions may be implemented.
One way to increase bandwidth on a bus is to increase its width. More data may be transferred per bus cycle, thereby increasing the overall available bandwidth. Unfortunately, wider buses are more expensive. Costs may be increased by increasing the number of layers of PCB board needed to route the larger number of lines between the various devices attached to the bus. Furthermore, connectors for attaching removable devices to the bus must become larger. The line width of the conductors is limited by impedance considerations as well as board fabrication technologies. Line spacing is affected by the fabrication technologies as well as by electrical cross-coupling tendencies.
Another way to increase bandwidth on a. bus is to increase the operating frequency of the bus. Typically, a bus uses electrical signalling techniques. Electrical signalling techniques are beginning to reach physical limitations in modern computer systems. As bus frequencies increase, the length of the bus conductors becomes a problem. Essentially, the longer conductors become antennae which broadcast the signals being conveyed thereon. Cross-coupling between bus conductors increases, and the electro-magnetic interference (EMI) produced may exceed FCC specifications. Solving the cross-coupling and EMI problems can be an extremely expensive and time-consuming activity. Still further, as frequencies are increased the bus conductors are increasingly dominated by transmission line effects. The transmission line effects limit the frequency at which a particular bus may operate. An additional problem with high frequency buses is that differential signalling often must be employed. Each bus signal requires a pair of conductors when differential signalling is employed, increasing the overall number of conductors and thereby incurring the problems with wider busses discussed above.
SUMMARY OF THE INVENTION
The problems outlined above are in large part solved by a computer system in accordance with the present invention. The computer system employs a hierarchical ring structure for high frequency communication between devices included therein. Computer system elements (such as microprocessors, memory, etc.) are configured into modules with ring interface hardware, and the modules are coupled to one or more rings. Bridge modules may be included for transmitting between rings in the hierarchy. The rings are time division multiplexed, and each time slot on a ring carries a frame. According to an address carried within the frame, bridge modules determine whether or not to transmit a frame circulating on a source ring onto a target ring. The target ring may be higher or lower in the hierarchy than the source ring. If the address of the frame indicates a module upon the source ring, the bridge module retransmits the frame on the source ring. Otherwise, the bridge module transmits the frame on the target ring. The bridge module operates in this fashion at any level of the hierarchy. Advantageously, a simple and rapid protocol for transmitting frames is replicated at each level of the ring hierarchy. Because an address masking and match operation may be performed quickly, a short ring transmit time may be maintained at any ring level. Therefore, intra-ring transfers may be performed rapidly, providing high bandwidth communication.
Since time slots are assigned to modules upon a given ring, each module is provided bandwidth without requiring arbitration. Advantageously, time spent arbitrating for access is eliminated from data transfer time. Additionally, the owner of a time slot is permitted to release the time slot for use by other modules, allowing for a module experiencing low data transfer traffic to relinquish bandwidth to a module experiencing high data transfer traffic. The bandwidth available on the ring can thereby be allocated to high traffic modules, and may be fully utilized even when traffic is unevenly dispersed among the modules. The source of a frame in a time slot may be identified via a return address field within the frame. The return address is a local address to the ring, occupying just a few signal lines. Still further, the owner of a time slot can reclaim a released time slot using an arbitrationless signalling protocol. To reclaim a time slot, the owner marks the time slot owned. The module using the time slot, upon detecting t
Conley Rose & Tayon PC
Geckil Mehmet B.
Merkel Lawrence J.
Prieto Beatriz
Sun Microsystems Inc.
LandOfFree
Chain transaction transfers between ring computer systems... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Chain transaction transfers between ring computer systems..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Chain transaction transfers between ring computer systems... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2817966