Multiplex communications – Network configuration determination – Using a particular learning algorithm or technique
Reexamination Certificate
2000-04-29
2003-12-23
Nguyen, Chau (Department: 2662)
Multiplex communications
Network configuration determination
Using a particular learning algorithm or technique
C370S380000
Reexamination Certificate
active
06667960
ABSTRACT:
BACKGROUND
Server class computer products are constructed by the combination of modular sets of computer components. These components can consist of a number of processors, a global-shared memory environment, main memory, PCI (Peripheral Components Interface) controllers and other components as required. Server class computer products can also be configured using basic building blocks. Example building blocks include a cell, a crossbar system, a routing chip and a PCI-based input/output (I/O) subsystem. In this case, a cell consists of shared multiprocessor (SMP) system containing from one to four (or more) processors, a portion of system memory and a connection to an I/O subsystem. Normally the cell is designed such that the hardware will not limit the mixture of different types of cells within the system. Cells can also be added or removed while the system is running. In typical systems, the cell resides on a single PC board.
The components included in the system can communicate with each other through a common bus or through point-to-point communication. Point-to-point communications consist of a discrete path such as a dedicated or switched line from one system component to a second system component. In addition or as an alternative to individual point-to-point communications, a crossbar system, a second building block, can provide switched non-blocking point-to-point interconnection between a number of cells and their associated memory. In systems, the crossbars are expected to reside oh backplanes. The third basic building block, the routing chip, connects the crossbar system to a high speed link for connecting a number of nodes into a single large system. The routing chip forms a high availability firewall to prevent failures in one node from affecting other nodes. Links can also be added or removed while the system is running. The fourth basic building block, the I/O subsystem provides connections for a number of PCI buses. Each cell has a link to a single I/O subsystem which can be located in another cabinet. PCI cards or entire I/O subsystems can be added or removed while the system is running.
A node is comprised of a set of cells connected by crossbars. Node-to-node connections are made using an interfacing routing chip (RC) and associated cables. Nodes can also be connected to each other to form larger systems.
When the system architecture is fixed, individual processors within cells can be made aware of other elements in the system through an available hardware architecture map. This hardware architecture map can be provided to the processor through its inclusion in read only memory (ROM). In this configuration a processor accesses the hardware architecture map stored in ROM to determine which other system components are available and communicates accordingly.
If all the system components are connected to a common bus, a processor on the bus has access to addresses of other system components through bus converters. By traversing the bus, the processor is connected to bus converters which connect to other buses in the system. Using this information, a processor can construct a network architecture or topology which identifies other system components within the system. Within this system, when one processor addresses a message to a second processor, the bus converter and the bus become transparent and the messages are passed from the sending processor to the receiving processor. That is, there is no indication or information provided about message routing. Through the use of this network architecture or topology the processor is aware of the functional connections between system components. However, using this system the processor is unaware of the physical layout of other system components or of the overall connecting and messaging network topology.
A processor's knowledge of the topology is important to reduce overhead associated with interactions between system components. By reducing the pathways between cooperating system components, associated overhead expenses are reduced.
Identifying the topology in a point-to-point system is more difficult then when system components are connected with a common bus. One method of identifying the topology is an exhaustive search. In an exhaustive search a single processor determines other system components by sending messages to every possible address.
Alternatively, sideband signals can be used to identify connected system components. For example, if system components have six-bit addresses, six physical wires can be run from one processor to its neighboring hardware component. The processor can then put its six-bit address on these dedicated wires and the attached physical component can determine the processor's address through these wires. Additionally, six separate wires would have to be run from the hardware component to the processor so the processor could determine the hardware component's six-bit address over these six dedicated wires. In this configuration accommodating 64 component addresses, twelve (12) wires are required between each set of two components so that each component would be aware of its neighboring hardware component's address. These hardware addresses could be determined through the use of dip switches. Although this is the simplest way of passing address elements between components, it is also the most expensive in terms of wires run. In this case, software would not be required to pass component addresses since the physical wires themselves contain the addresses.
The number of wires could be reduced by the introduction of logic to serialize the exchange of address information. If a bidirectional wire is used between the two system components, a single wire can be used to exchange addresses between the two components. In this case coordination must be obtained through software or hardware components to ensure the bi-directional communication is satisfactorily obtained. However, with a bi-directional connection, sideband signals indicating neighboring components are not normally passed between the components.
Accordingly, a need exists for systems in which components can exchange address information while minimizing costs in terms of wire runs and software or hardware control components. A further need exists for a system that allows system components to generate and maintain a functional and physical topology of system components.
SUMMARY OF THE INVENTION
These and other objects, features and technical advantages are achieved by a system and method which, according to one aspect of the invention, include a method of mapping a network topology in a network that includes a plurality of nodes communicating with each other over dedicated links connecting pairs of the nodes. The method includes exchanging respective network identification information between adjacent pairs of nodes and, establishing communications with another of the nodes (i.e., the neighbor's neighbor node) using the network identification information. Network identification information of the other node from that node, using the network identification information to establish communications with other nodes, obtaining additional network identification information from those other nodes, repeating these steps until network identification information is obtained from all of the nodes of the network and using this information determine the network topology.
According to a feature of the invention, the network identification of immediately adjacent nodes (i.e., neighboring nodes) are stored in respective network identification registers. The nodes include both terminal nodes (i.e., data users and sources) and switching nodes (i.e., communications resources). Thus, the terminal nodes include processing cells and the switching nodes may include crossbar switching devices.
According to features of the invention, the steps of identifying and recording the identification information of neighbors may be performed either iteratively (e.g., by depth first probing to reconnect topology starting outward and progressing t
Dodge Benjamin
Lindsay Dean T.
Snyder Robert D.
Mills Donald L
Nguyen Chau
LandOfFree
Protocol for identifying components in a point-to-point... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Protocol for identifying components in a point-to-point..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Protocol for identifying components in a point-to-point... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3121478