US 5,630,173 AGrant
Methods and Apparatus for Bus Access Arbitration of Nodes Organized Into Acyclic Directed Graph by Cyclic Token Passing and Alternatively Propagating Request to Root Node and Grant Signal to the Child Node
Issue Date:1997-05-13
•13 Claims
•11 Drawing Sheets
Abstract
A bus arbitration scheme is implemented in a system where an arbitrary assembly of nodes on a system bus have been resolved into an acyclic directed graph. The hierarchical arrangement of nodes has one node designated a root while all other nodes have established parent/child relationships with the nodes to which they are linked. Each node may have a plurality of connected child ports with a predetermined acknowledgment priority scheme established. Fair bus access arbitration provides for bus granting in a sequence corresponding to the predetermined port priorities allowing all nodes a turn on the bus. The root node may always assert its priority access status to gain bus access which is useful for accommodating a root node which requires isochronous data transfer. Alternatively, a token passing arbitration scheme may be implemented where the token for bus access is passed around the nodes according to the above-described predetermined port priority scheme. Preemptive bus initialization may be triggered by any node upon detection of a necessitating error or addition or removal of a connection to an existing node.
Metadata
Assignee
- Apple Computer, Inc.
Inventor
- Florin Oprescu
Application Information
Application Number:US 9949836
Filing Date:1992-12-21
Priority Date:1992-12-21
Art Unit:237
Classifications
IPC:
G06F 1314G06F 13362G06F 13368
Field of Search:
395370200;325;293;285;287;299;300;301;302;731;863;860;295;729;296;73016;94.3
Patent Drawings (11 sheets)
Description
BACKGROUND OF THE INVENTION
1. Related Applications
This application is related to application Ser. No. 07/994,402, now U.S. Pat. No. 5,394,556 entitled "Method and Apparatus for Unique Address Assignment, Node Self-Identification and Topology Mapping for a Directed Acyclic Graph" and Ser. No. 07/994,128, now abandoned entitled "Method and Apparatus for Transforming an Arbitrary Acyclic Topology Collection of Nodes into an Acyclic Directed Graph" each of which is assigned to the assignee of the present application and filed concurrently herewith.
2. Field of the Invention
The present invention relates to computer systems. More particularly, the present invention relates to a method and apparatus for establishing and utilizing a communications scheme between a plurality of arbitrarily assembled elements of a computer system.
3. Background
Components within a given computer system need the ability to convey signals amongst themselves. In very simple systems, it is possible to have each element of the system directly wired to all of the other parts of the system. However, in reality, in order to make computers expandable and to accommodate an unknown number of system parts, computer architects long ago developed the concept of a communications bus.
A bus is a communications path, such as a wire or wires, running throughout the computer system. Each component of the system need only plug into the bus to be theoretically connected to each of the other components in the system. Of course, each component cannot simultaneously communicate with other components because there may be only a single communications channel between the components. It is necessary when utilizing a communications bus to establish some form a sharing arrangement so that each component may use the bus to communicate with other components in an efficient manner that does not leave critical pieces of information from one component hanging, waiting for bus access. The method by which components on the bus share the bus is generally referred to as a bus arbitration scheme.
In addition to the critical need to optimize the bus arbitration scheme so as to maximize the flow of important information, the physical (and logical/electrical) configuration of the bus itself can and should be optimized to minimize system delays while remaining as flexible as possible.
In order to communicate with other components attached to a bus, each component must be equipped with hardware such as transmitting and receiving circuitry compatible with the communications protocol implemented for the bus. One such communications standard is described in IEEE Standards Document P1394 entitled "High Performance Serial Bus", said document attached as Appendix A to this document. The standard described in P1394 is intended to provide a low cost interconnect between cards on the same backplane, cards on other backplanes, and external peripherals.
Prior art buses or networks required knowing what was being plugged in where. For example, the back of many computers have specified ports for specific peripherals. Some computers implement several buses, such as the Macintosh which uses a bus referred to as ADB for components like a mouse and keyboard and SCSI bus for other peripherals. These types of buses provide for daisy chaining elements together but connections are of limited topology. Other known buses/networks require that the nodes of the network be arranged in a ring, a loop which must be closed in order to operate. Finally, star, or hub-and-spoke arrangements required that each node be directly linked to a central master. Each of the prior art systems lacks a desirable measure of flexibility.
It would be desirable, and is therefore and object of the present invention, to be able to arbitrarily assemble elements of a computer system onto a bus where the arbitrary topology can be resolved by the system into a functioning system without requiring a predetermined arrangement of components.
SUMMARY OF THE INVENTION
It is an object of the present invention to provide a fair bus access arbitration scheme for a computer system bus or network in which the connections of nodes have been resolved into an acyclic directed graph.
It is also an object of the present invention to provide a priority bus access arbitration scheme for a computer system bus or network in which the connections of nodes have been resolved into an acyclic directed graph.
It is another object of the present invention to provide a method of token passing bus arbitration for a computer system bus or network in which the connections of nodes have been resolved into an acyclic directed graph.
It is yet another object of the present invention to provide a mechanism whereby a preemptive bus initialization may be triggered by any node in a network of nodes which are resolved into an acyclic directed graph when errors are detected or nodes are added or subtracted during operation.
These and other objects of the present invention are implemented in a system where an arbitrary assembly of nodes on a system bus have been resolved into an acyclic directed graph. The hierarchical arrangement of nodes has one node designated a root while all other nodes have established parent/child relationships with the nodes to which they are linked. Each node may have a plurality of connected child ports with a predetermined acknowledgment priority scheme established. Fair bus access arbitration provides for bus granting in a sequence corresponding to the predetermined port priorities allowing all nodes a turn on the bus. The root node may always assert its priority access status to gain bus access which is useful for accommodating a root node which requires isochronous data transfer. Alternatively, a token passing arbitration scheme may be implemented where the token for bus access is passed around the nodes according to the above-described predetermined port priority scheme. Preemptive bus initialization may be triggered by any node upon detection of a necessitating error or addition or removal of a connection to an existing node.
BRIEF DESCRIPTION OF THE DRAWINGS
The objects, features and advantages of the present invention will be apparent from the following detailed description in which:
FIG. 1 illustrates a block diagram of the hardware layer implementation utilized in accordance with the present invention.
FIGS. 2(a)-2(b) illustrate arbitrarily assembled collection of nodes, one being acyclic and the other including multiple cycles.
FIG. 3(a) is the arbitrarily assembled collection of nodes of FIG. 2(a) undergoing the graph transformation process in accordance with the present invention.
FIGS. 3(b)-3(d) illustrate alternative communications exchanges between nodes in implementing the present invention.
FIG. 3(e) graphically illustrates the directed graph resulting from the arbitrarily assembled network of nodes of FIG. 2(a).
FIG. 4 illustrates a symmetrical graph arrangement which requires resolving a root contention.
FIG. 5 illustrates a directed acyclic graph with a possible unique address assignment order indicated.
FIGS. 6(a)-6(f) illustrate the process flow for carrying out the graph transformation procedure in accordance with the preferred embodiment of the present invention.
DETAILED DESCRIPTION OF THE INVENTION
A method and apparatus for utilizing a bus having an arbitrary topology are described. In the following description, many specific details are set forth such as various computer components in order to provide a thorough understanding of the present invention. It will be obvious, however, to one skilled in the art that the present invention may be practiced without such specific details. In other instances, well-known control structures and coding techniques have not been described in detail in order not to obscure unnecessarily the present invention.
Throughout this detailed description, numerous descriptive terms are introduced to provide metaphorical clarity to the description. For example, frequent references will be made to parent-child relationships between nodes in a given topology. The purpose of this is to provide the concept of "direction" to the finally resolved graph. As will be described, once an arbitrary topology has been reduced to an acyclic directed graph, there will be one node identified as the "root" node. The root node will not have a parent node, all nodes logically immediately adjacent to the root node are the child nodes of the root. The "tree" metaphor is completed by the inclusion of nodes referred to as "branches" and "leaves".
The bus architecture described herein, though described with reference to components for a single computer, in general has a broader scope. The present invention for defining the bus topology may be applied to any arbitrarily assembled collection of nodes linked together as in a network of devices. One point that must be noted is that it is necessary to distinguish a node from a physical computer component. Each component to reside on the bus will have associated with it at least one node physical layer controller. In certain circumstance, a given component may advantageously be associated with multiple nodes but in the usual case there will be a one-to-one correspondence between devices or components on the bus and nodes.
Referring now to FIG. 1 a block diagram of a node 10 is illustrated. The physical implementation of a node is somewhat arbitrary. In the preferred embodiment implementation of the present invention, the nodes are designed to comply with the IEEE P1394 High Performance Serial Bus communications protocol which is attached as Appendix A. The node 10 includes arbitration state machine logic 11. This arbitration state machine logic incorporates all the logic circuitry for carrying out the methodologies and algorithms to be described herein. The circuitry may comprise a programmable logic array (PLA) or be uniquely designed to carry out the functions described herein. Those skilled in the art, once described the functions to be carried out by the node logic will be able to implement the present invention without undue experimentation. The node, by means of its logic, shall implement the minimum arbitration protocol including the bus initialization, tree identification, self identification, and the bus arbitration functions, all to be described in detail further herein.
The node 10 shown in FIG. 1 also includes transmission multiplexers 12 and 13 and data transmitter, receiver and resynchronizer 14. The node illustrated in FIG. 1 is coupled to local host 15. Local host 15 may be any device one wishes to attach to the bus such as a disk drive, CPU, keyboard or any other component which needs to communicate with other components in the system. The node 10 communicates with other nodes through communications links. A link is a connection between two ports and in immediate practical terms is a cable segment but in general it may be implemented as any physical communication channel. A link shall be able, at minimum, to provide a half duplex communication channel between the two ports which it connects. A port is the interface between a node and a link. In accordance with the present invention, a port must have the ability to transmit and receive data and arbitration signaling. A port needs to be able to determine whether or not it is connected to another port through a link. One method of facilitating this is by having connected ports apply a biasing voltage through the link which is detectable by the port at the other end of the link. Thus, if a port has a link attached which is not connected to a port at the other end, a naked link, the port will determine that it is not a connected port. In FIG. 1, the illustrated node 10 has three external ports 21, 22 and 23 with connecting links 17, 18 and 19, respectively.
Some of the rules of implementation for nodes in order to implement the present invention are that a node may have one or more ports. A node shall be able to transmit and receive data on any one of its ports. A node shall be able to receive data on one and only one of its enabled ports at a time and be able to retransmit this data on all remaining enabled ports. A node shall be able to receive and transmit signaling messages through all of its ports simultaneously and independently. Separate signaling transceivers, encoders and decoders are required for each port of a node. A minimum implementation node does not require a local host device. For example, such a node may function as a cable extension. From hereon devices and local hosts will be ignored and all references to bus topology will refer to nodes and node connections through various ports.
Graph Transformation
FIGS. 2(a) and 2(b) illustrate arbitrarily assembled collections of nodes. From hereon, nodes will be illustrated merely as circles, but are deemed to each incorporate elements equivalent to those described with respect to FIG. 1. Note, however, that each node may have more or less than the three external ports shown in that figure. The lines illustrated connecting each of the nodes are the method by which links are shown. Ports are not illustrated, but are impliedly the interface where a link and a node connect.
The bus arbitration methodology to be described herein requires that the arbitrary topology be resolved into an acyclic directed graph. In an arbitrary topology graph a collection of nodes and links may form a cycle. A cycle exists if starting from a specific node in the graph it is possible to return to the same node by traversing links and nodes without any link being traversed twice. FIG. 2(a) illustrates an acyclic graph because none of the nodes illustrated are connected within a loop. FIG. 2(b), however, is not an acyclic graph because the region in bounding box 25 contains a collection of nodes, 40-47 which form multiple cycles. The bus arbitration methodology to be described requires that there be no cycles, so a method of user intervention to resolve cycles will also be described further herein.
In addition to the requirement that a graph be acyclic, it must also be directed. A directed graph is one in which a hierarchical arrangement has been established between adjacent nodes. Initially, there are no established parent-child relationships between nodes. That is, for example, node 31 may be the "parent node" for node 34, or be the "child node" for node 34. Thus, it is necessary to take a given arbitrary topology graph and transform it into an acyclic and directed graph. The methods described herein will work to perform this transformation for any give arbitrary topology, regardless of the number of nodes or how they are physically linked and regardless of the signaling propagation time along the links.
Node Communication
Initially, the process of transforming an acyclic arbitrary topology graph into a directed graph will be described. The case where cycle resolution is required will follow. FIG. 3(a) shows the arbitrary graph of FIG. 2(a) wherein the nodes and links have status labels and communicated signals are indicated for the graph transformation process for directing a graph. It is instructive at this point to describe signal communications between nodes. FIG. 3(b) illustrates two nodes 50 and 51 (hereinafter node A and node B, respectively) coupled by link 52. As described, the link is the communications channel coupling transceiver ports of the respective nodes as described above with reference to FIG. 1. During the graph transformation process, it becomes necessary for nodes to establish parent-child relationships with adjacent nodes. Two nodes are said to be adjacent nodes if there is at least one link connected between a port of the first node and a port of the second node. In FIGS. 3(b)-3(d) it will be assumed that the relationship to be resolved is that node B is the parent of node A and that it is appropriate for the nodes to establish that relationship.
Prior to a direction being established, when it become appropriate for node A to establish node B as its parent, node A will transmit from its port to which link 52 is coupled the signal "You Are My Parent" (YAMP). This message content may take any form, so long as node A knows that it is signaling YAMP and node B is capable of understanding that the received message is YAMP. When YAMP signal 53 is received by node B, node B will respond to node A by sending "You Are My Child" (YAMC) through link 52 to node A. The arbitration state machine logic 11 of node A will keep track of the time delay between sending YAMP signal 53 and receiving YAMC signal 54. The time measured signifies twice the propagation delay between nodes A and B. Upon receiving the YAMC signal, node A will respond with a "You Are My Child Acknowledged" (YAMCA) signal 55. This provides node B with the ability to also determine the propagation time delay between the nodes equal to the time delay between sending YAMC and receiving YAMCA. For half duplex communication links, the YAMCA message also has the effect of properly orienting the communication channel.
For full duplex communications channels the three logical messages, YAMP, YAMC and YAMCA can alternatively be relayed by only two signal transmissions. In FIG. 3(c) this situation is illustrated where node A asserts the YAMP signal 56 continuously until it receives the return YAMC signal 57. The YAMCA signal is logically transmitted to node B when the YAMP signal is detected as no longer arriving.
The use of this described triple asynchronous message exchange provides a mechanism by which both nodes involved in the message exchange can determine the propagation time delay through the link. This delay value is used in resolving contention events to be described further herein as well as during normal bus arbitration to optimize bus performance. The dynamic extraction of this parameter is not mandatory. As an alternative a maximum propagation time delay can be apriory defined at the expense of optimum bus performance.
Once nodes A and B have exchanged messages signifying that node B is the parent of node A, the link can be said to be directed. Node A within its logic labels its port to which link 52 is coupled as a parent port (it talks to a parent node) and node B labels its port to which link 52 is coupled a child port (it talks to a child node). It is important to maintain the labels that ports achieve because the methods to be described below will be in terms of the labels assigned to nodes and ports at a given time. A short hand graphical notation is illustrated in FIG. 3(d) where the direction arrow 58 indicates that node B is established as the parent of node A and the link is directed.
Direction Determination
Referring back now to FIG. 3(a) and to process FIGS. 6(a)-6(e), the process of directing the overall arbitrary topology will now be described. It is necessary to introduce a few more colorful definitions to aid in explaining the topology transformation process. First, a "leaf" node is defined as a node with only one connected port. A node recognizes its status as a leaf node as soon as it is initialized after power-up or other bus initialization. A "branch" node is a node which has a least two connected ports. Through all but one of the connected ports a branch node will have received the YAMP signal and have acknowledged it. Through its remaining port, a branch node has sent the YAMP signal thus establishing that it has a parent node. A node does not achieve branch status until it has established that it has one parent (a node can have only one parent node) and all its other ports are connected to child nodes. Prior to achieving branch status, a node is considered a "cycle" node because until it is determined to be a branch the possibility exists that the node is part of a cycle which makes establishing direction impossible.
The graph transformation procedure begins at step 60 upon bus initialization (power-up or instigated) at which time the leaf nodes in the arbitrary topology recognize at step 61 and label themselves at step 68 as leaf nodes by determining that they have only one connected port at decision box 66. In the graph depicted in FIG. 3(a), nodes 33, 35, 36 and 37 are leaf nodes which, once initialized, at step 69 each transmits the YAMP signal through its only connected port to its adjacent node. The nodes receiving these signals will then propagate the YAMC signals back to the leaf nodes at step 70, thus establishing a direction for the given link between respective parent-child pairs when the YAMCA communication is completed. At step 71 each leaf node labels its one connected port as a parent port and each receiving port on the parent node is labeled a child port.
The nodes on the graph which are not initially leaf nodes are initially considered "cycle nodes" for the reason described above and proceed according to the Cycle Node Procedure 63. Any cycle node which has labeled all but one of its connected ports as child ports then propagates the YAMP signal from its remaining unlabeled port at step 85. When that direction is established for the link, the cycle node then becomes labeled a branch node. Thus, after leaf node 37 establishes that node 34 is its parent, node 34 has only one unlabeled port (having labeled the link connection to node 37 as being through a child port) so node 34 broadcasts the YAMP signal to node 31, resulting in node 34 becoming a branch node. Likewise, once node 31 has identified that nodes 33 and 34 are its children, node 31 broadcasts the YAMP signal to node 30. When one node has received through all of its ports the YAMP signal at decision box 75, that node becomes the root node. In FIG. 3(a) after node 30 has received the YAMP signals from nodes 31 and 32, its label changes from being a cycle node to being the root node. In the graph of FIG. 3(a), it is not necessarily the case that node 30 would become the root. If some of the links in the tree provided long propagation delays, node 30 might have received a YAMP signal on one port and then transmitted a YAMP signal through its other port. Any of the nodes may become the root, even a leaf, the root property taking precedence. FIG. 3(e) shows the resulting directed graph in response to the communicated signals shown in FIG. 3(a) with each node labeled and the directions indicated by dark arrows.
Root Contention
In certain circumstances a root contention situation may arise. This may happen for example in the case where the arbitrary topology has a symmetrical arrangement to it such as that shown in FIG. 4. In the arbitrary graph illustrated in FIG. 4, nodes 160 and 161 have each established that it is a parent to the two leaf nodes to which it is coupled. Then, each has propagated the YAMP signal to the other at nearly the same time. The root contention situation is recognized by both nodes involved at decision box 86. Each node is receiving a signal which designates it as a parent while it has sent the same signal out through the same port. Each of the contending nodes responds to the other with the YAMC signal at step 91 which allows each to determine the "decision time period" which is equal to twice the propagation time between the nodes.
The root contention situation is resolved by utilizing a random decision mechanism incorporated on each arbitration state machine logic unit 11 of each node. For every "decision time period" that elapses, each node will randomly decide at step 92 (with a 50% probability) whether to again transmit the YAMP signal to the other. Almost certainly within a finite number of the cycles, one node will decide to designate the other its parent without that one reciprocating. The one that is designated the parent becomes the root at step 95. Alternatively, predetermined selection criteria values may be assigned to nodes, the larger or smaller determining which dominates in a contention event. The dynamic determination of the "decision time period", while it offers optimum performance is not essential in implementing the present invention. As an alternative an apriory defined "decision time period" may be used as long as it is greater than the worst case link propagation that can be encountered in any bus using this algorithm. The same method used to resolve root contentions will also be used to resolve other contention events to be described further herein.
Root Assignment
As described above, the result of the graph transformation process is the assignment of the root attribute to one and only one node in the graph. The root node will have the ultimate decision in the bus arbitration scheme to be described and therefore can access the bus with maximum priority without the use of special priority time gaps. It is often desirable to be able to assign the root property to a predetermined node either when it is manufactured or dynamically (during run time) to optimize a given system. A given bus may include a node which requires isochronous data transfer. Isochronous data is data that must be transmitted at a given time to be of any value. For example, music from a compact disk needs to be transferred and output in the order in which it is to be heard and with no significant delays, unlike data files which may be transferred piecemeal and not necessarily in order.
Nodes can be classified into three categories with respect to root designations. These designations may be applied during manufacturing by hard-wiring the designation into the device, programming the arbitration state machine logic or by higher level software making the decision then initiating a reboot while preserving that decision. The three designations that a node may be assigned with respect to being designated a root are: nodes that do not want to be root, nodes that may (should) be root and nodes that shall be root. These designations are tested for at steps 81 and 83. A node designated in the first category will begin the graph transformation procedure as soon as it is directed to do so. This will usually be immediately following the completion of the bus initialization procedure. A node from the second category will delay the beginning of the graph transformation procedure for a predetermined amount of time after it is directed to begin the procedure at step 84. By this delay, the node increases its chance of becoming the root. (The YAMP signals are more likely to propagate to it due to the delay.) Despite the added delay, it is still possible that a "may be root" node will not wind up being designated the root. This will depend on the given topology and message propagation delays. The amount of delay can be defined during design to be greater than a reasonable worst case propagation delay through a fairly complex graph.
A node from the third category of root designation possibilities may only recognize the fact that it must be the root after the graph has already been transformed and all nodes have identified themselves. The arbitration state machine logic may make this determination or software running on the host system. When this occurs, the node that has to be root agrees with all other nodes on the bus that it is going to be the one and only root and restarts the graph transformation process by signaling a preemptive bus initialization signal which is described further herein. The node then waits at step 82 to become the root and does not participate in the graph transformation until it has received the YAMP signal on all of its ports, thus forcing it to be designated the root.
Once the root has been determined, the graph can be said to be directed. There is a defined relationship existing between all adjacent nodes on the graph.
Cycle Resolution
The procedures described above for directing a graph will only work for an acyclic graph. If there are cycles in the arbitrary topology, they must be broken by the procedure beginning at step 80. The existence of a cycle is detected at step 79 when, after a predetermined time-out period has elapsed, a node is still labeled a cycle node rather than a leaf, branch or root. The "cycle detect" timing starts immediately after the end of the bus initialization function. The time-out period need be no longer than the worst case duration of the graph transformation process (adding in delay time for a "may be root" node and a possible root contention event).
The "cycle detect" time-out event does not have to occur synchronous for all nodes of a graph as all message exchanges are asynchronous events. As such, it is possible for a node which has not yet reached its "cycle detect" time-out event to receive a message indicating that cycle resolution is ongoing. Such a node will terminate its cycle detect time-out interval and begin the appropriate cycle resolution process.
The method of cycle resolution in accordance with the present invention requires the user of the assembled collection of nodes to intervene. When a node encounters the "cycle detect" time out the system user may be notified at step 100 of FIG. 6(a) through an output device that a cycle exists and which nodes are then involved. The user will then be instructed to disconnect links to eliminate whatever cycles are present. The user will then return control to the graph transformation procedure.
Once each of the loops is broken, and no cycles remain, the procedure for transforming the graph as described in the earlier sections may proceed until the entire graph is both acyclic and directed.
Unique Physical Address Assignment
Once a directed acyclic graph has been established from the original arbitrary topology, it is then possible to assign unique physical addresses to each node on the graph. This process begins with all leaf nodes requesting the bus by transmitting through their single connected ports the Bus Request (BR) signal. The parent node receiving the signal will wait until it has received the BR signal from all of its child ports and then will propagate the BR signal to its parent. The BR signals will propagate through the graph until the root has received the BR signal from all of its children. Once the root has received a bus request through all of its child ports, it will make a decision for granting the bus through one port and propagating a Bus Denial (BD) signal through its remaining child ports. The method for selecting which bus request to grant may be an apriory decision such as that described above where, for example, ports are selected from left to right or based on port numbering, etc. The Bus Grant (BG) signal will be transmitted from the root to its requesting child. If that requesting child is itself a parent node which has propagated the bus request from one of its children, it will send the bus denial signal through all but one of its child ports in the same predetermined fashion as described above. Eventually one leaf node will receive the bus grant signal to which it will reply with a Bus Grant Acknowledged (BGA) signal which will be propagated back to the root node. The propagation of the BD and BGA signals serve to orient the communication links which may be necessary for the case of half duplex communications channels. All of the denied nodes will then wait for activity by the node which finally receives the BG signal.
The node which is finally granted access to the bus will transmit an address assignment packet. The node will transmit this packet on the bus and it will be received by all other nodes, each of which will count the number of address packets they receive. The transmitted address packet may have any arbitrary information. A node's unique physical address will be based on the number of address packets a node has counted before it transmits an address packet. Thus, no two nodes will acquire the same physical address despite not having address information assigned in advance. The actual composition of the address packet is arbitrary and may be any bit stream efficiently utilizable by the system. After transmitting the physical address assignment packet, a node will transmit a "Child ID Completed" signal (CIC) signal. The parent node receiving this on its child port will then transmit the "Child Identification Completed Acknowledgment" (CICA) signal and label the port as an identified child port. In response to the next BR signal propagation, the parent of the node which has just identified itself will then select its next child to transmit the physical address packet. Once all the child nodes of a parent node have identified themselves, the parent node will request the bus and, when granted the bus, will propagate its physical address assignment packet. This procedure will continue following the predetermined selection criteria until all nodes determine a unique physical address assignment by counting. FIG. 5 illustrates the graph of FIG. 3(e) in which a left-to-right predefined selection criteria is implemented. The nodes are uniquely assigned addresses where node 33 receives the first and as described, the root node 30 receives the eighth and last address.
When this procedure is completed, each node in the graph will have a unique physical address, which need not have been determined in advance and which may be utilized for system management or other purposes.
Node Self-Identification
The process of node self identification essentially follows the same routine as the physical address assignment procedure described above. As each node transmits its physical address assignment packet, that packet may include further information such as identification of the physical device comprising the identification of the local host related to the node, how much power it requires, and, for example, whether it supports a "soft power-on" attribute, etc. In fact, the node self-identification information may serve as the physical address assignment packet because the practice of sending any information at all provides the basis for counting to yield unique physical addresses.
With respect to the node self-identification packet, the particular information concerning the node need only be "listened" to by those nodes affected by the nature of the announcing node. This procedure, as with the above, proceeds until all nodes have transmitted their node self identification information.
Topology Mapping
The method of topology mapping follows along the same lines as physical address assignment and node self-identification. This procedure thus has each node, when it is going through the address assignment or node self-identification process, further transmit information concerning all of its ports such as the number of child ports it has and whether or not it has any disabled ports. With respect to disabled ports, it may be desirable to implement a communication protocol between ports that are disabling so that they can identify from whom they are disabled. Thus, when a port identifies a disabled port it will give an identifier indicating its own ID as well as the port ID from which it has become disabled.
By assembling all the topology information about all the ports received during the topology mapping procedure, the bus server, host or any software level application may logically reconstruct the resolved bus topology. This is useful for many purposes including implementing redundancy where if a link unexpectedly goes down, previously disabled links may serve to prevent the loss of communication channels to any nodes.
Fair Bus Access Arbitration
Once the topology mapping, node self identification or physical address assignment routines have completed, the bus can be considered up and running. One arbitration scheme implemented in accordance with the present invention is that of fair bus access. When a node desires access to the bus, it sends through its parent port (unless it is the root) a bus request (BR) signal. The parent, upon receiving the BR signal from one child sends a bus denied signal (BD) through all its other child ports. The parent then propagates the BR signal upward through its parent until the signal reaches the root. The root issues a bus grant signal (BG) responsive to the first BR signal it receives and sends the BD signal through all of its other child ports which propagate downward thereby orienting the links. The BG signal propagates downward through the graph until it reaches the requesting node which then sends Bus Acknowledge (BA) signal followed by the packet of information that the node needed to send on the bus. When the packet is completed, all nodes return or enter into an idle state.
In the case where the root receives nearly simultaneous requests for the bus, the predetermined selection criteria for the root node will be used for granting to one of the nodes bus access. This may be the same predetermined priority selecting criteria as described above.
A further aspect of fair bus access arbitration is that a parent node has priority over its children. Thus, when a parent node wants the bus, it sends the BD signal through all of its child ports, then propagates the BR signal up toward the root. One potential problem with this mechanism is that if the parent has a large quantity of information to transmit on the bus a child node may have trouble getting adequate bus access. There is therefore introduced a gap system which is widely used and well-known in the art. After a node has utilized the bus, the node must wait for one gap period before it can again request the bus. This gives equal chance of being granted the bus to every node on the bus independent of its topological placement on the bus. In order to guarantee a fair arbitration protocol the length of the gap has to be greater than the worst case signal propagation delay through the bus. The gap value can be predetermined and hard-wired into the node logic but such an approach will result in all but the most extreme case in a sub optimal utilization of the bus. The topology mapping capability together with the measurement of the propagation delay between adjacent nodes performed during the graph transformation phase enables the calculation of an optimal fair gap that will maximize the bus performance for any specific implementation.
Priority Bus Arbitration
In the bus arbitration scheme implemented in accordance with the above fair bus access arbitration, it may be desirable that the root always have bus priority. When this is implemented, the root node may grant the bus to itself at any time. This is done by first sending the BD signal down through all of the nodes in the graph. Priority bus access for the root is very useful for the case where the root node is required to perform isochronous data transfer.
Token Passing Bus Arbitration
As an alternative to the fair and priority bus access arbitrations schemes described above, the present invention may be utilized in implementing a token passing bus arbitration scheme. Metaphorically speaking, token passing bus access refers to the notion that a node may communicate on a bus when it is in possession of a token that is passed between nodes. The token is passed from node to node in a cyclic fashion so that each node receives the bus in a predetermined point in the cycle. Token passing is implemented in the present invention in following the same manner as the physical address assignment routine described above. The predetermined selection mechanisms implemented are used to select the order in which the token will be passed from node to node. This order resembles the order as shown in FIG. 5 which dictates the order of unique address assignment. Each node, when it is assigned the token will propagate its information packet on the bus while the remaining nodes listen. The node will then pass the token to the next logical node based on the predetermined sequencing method as described above.
Preemptive Bus Initialization
An important feature that may be implemented in accordance with the present invention is the notion of preemptive bus initialization. The state machine logic incorporated on each node is capable of triggering a bus initialization (BI) signal to be propagated from the node through all of its ports upon certain conditions. When a node has determined it is necessary to signal a bus initialization condition, it will propagate the BI signal out through all of its ports for a length of time sufficient to guarantee that all adjacent nodes have received it and then released. A node will then go into the initiating procedures which then lead to the graph transformation process in the above described procedures.
There are a number of situations which may make it necessary or desirable to trigger a preemptive bus initialization. First, this may be a node response to an unforeseen error. Additionally, at the host level, it may be determined that a different node should acquire the root attribute, for example, an isochronous data transfer node. This assignment will be preserved throughout the bus initialization routine thereby causing the desired node to wait during the transform procedure until it receives the root designation. Another condition leading to a preemptive bus initialization may be the breakage of a link, in which case it may be necessary to calculate a new acyclic directed graph for the attached nodes. Finally, an important situation in which a preemptive bus initialization should occur is when a device is added to the network, referred to as "hot addition" of peripherals. The port to which a new device is connected will detect the presence of a new node and trigger a bus initialization which will be transparent to the user of the system but which allows the addition and subtraction of peripherals, for example, without having to shut down and repower. A new acyclic directed graph is calculated which includes the presence of the added node. It is possible that upon removing certain nodes, it will not be necessary to trigger a bus initialization, for example, when a leaf node is removed, there is no harm to the network. However, if a branch node is disengaged from an operating bus, it is likely to be necessary to reconfigure the graph.
Although the present invention has been described in terms of preferred embodiments, it will be appreciated that various modifications and alterations might be made by those skilled in the art without departing from the spirit and scope of the invention. The invention should, therefore, be measured in terms of the claims which follow.
Claims
I claim:
1. In a computer system comprising a plurality of components interconnected by a plurality of communications links, said plurality of components each having at least a first communications node wherein said communications nodes interface their associated component with a communications link through a node port, said nodes being capable of having a plurality of ports, said configuration of nodes and communications links comprising a directed acyclic graph wherein one node is designated a root node, all nodes coupled to only one adjacent node are designated leaf nodes, all other nodes in the graph being designated branch nodes, said acyclic directed graph having established hierarchical parent-child relationships between all adjacent nodes proceeding from the root node down to any leaf nodes wherein a leaf node has only one parent node and all nodes adjacent to the root node are child nodes with respect to the root node but parent nodes with respect to other adjacent nodes, the root node being defined as having no parent node, a method of fair bus access arbitration comprising the steps of: transmitting a "Bus Request" (BR) signal from a requesting node to its parent node, said requesting node having a packet of information to propagate on said bus; all nodes receiving the BR signal forwarding the BR signal to their respective parent node; the root node, upon receiving a BR signal from one adjacent node, responding to said adjacent node with a "Bus Grant" (BG) signal; all nodes receiving the BG signal from a parent node then propagating the BG signal to the child node which previously forwarded the BR signal, unless the node is the requesting node; and the requesting node, upon receiving the BG signal propagating said packet of information on the bus, wherein after the requesting node has propagated said packet of information, the requesting node waits for a gap period before again requesting access to said bus, said gap period being greater than a worst case signal propagation delay through the bus.
2. The method of claim 1 further comprising the step of before said transmitting a BR signal step, transmitting a "Bus Denied" (BD) signal from said requesting node to any child nodes it has.
3. The method of claim 2 further comprising the step of propagating a BD signal from all forwarding nodes to all child nodes that were not the source of the BR signal.
4. The method of claim 3 further comprising after said step of said root node responding to an adjacent node with a BG signal, the step of said root node propagating a BD signal to all other adjacent nodes.
5. The method of claim 4 further comprising the step of all nodes receiving a BD signal propagating said BD signal to all adjacent child nodes.
6. The method of claim 5 further comprising before said requesting node propagating said packet of information step, the step of the requesting node acknowledging receipt of the BG signal.
7. The method of claim 6 further comprising the step of all nodes receiving a BG signal acknowledgment, forward said acknowledgment to a parent node.
8. The method of claim 1 further comprising the step of the root node, upon determining that it requires access to the bus, propagating a BD signal to all adjacent nodes and granting itself the bus.
9. The method of claim 7 further comprising the step of the root node, upon determining that it requires access to the bus, propagating a BD signal to all adjacent nodes and granting itself the bus.
10. The method of claim 1 wherein the root node solves conflicting simultaneous requests for the bus by granting the bus based on a predetermined adjacent node selection criterion.
11. In a computer system comprising a plurality of components interconnected by a plurality of communications links, said plurality of components each having at least a first communications node wherein said communications nodes interface their associated component with a communications link through a node port, said nodes being capable of having a plurality of ports, said configuration of nodes and communications links comprising a directed acyclic graph wherein one node is designated a root node, all nodes coupled to only one adjacent node are designated leaf nodes, all other nodes in the graph being designated branch nodes, said acyclic directed graph having established hierarchical parent-child relations ships between all adjacent nodes proceeding from the root node down to any leaf nodes wherein a leaf node has only one parent node and all nodes adjacent to the root node are child nodes with respect to the root node but parent nodes with respect to other adjacent nodes, the root node being defined as having no parent node, a method of fair bus access arbitration comprising the steps of: transmitting a "Bus Denied" (BD) signal from a requesting node to all of its child nodes, said requesting node having a packet of information to propagate on said bus; transmitting a "Bus Request" (BR) signal from a requesting node to its parent node; all nodes receiving the BR signal forwarding the BR signal to their respective parent nodes and propagating a BD signal to all child nodes that were not the source of the BR signal; the root node, upon receiving a BR signal from one adjacent node, responding to said adjacent node with a "Bus Grant" (BG) signal, said root node propagating a BD signal to all other adjacent nodes; all nodes receiving the BG signal from a parent node then propagating BG signal to the child node which had previously forwarded the BR signal, unless the node is the requesting node; the requesting node, upon receiving the BG signal acknowledging said signal and propagating said packet of information; and the root node, upon determining that it requires access to the bus, propagating a BD signal to all adjacent nodes and granting itself access to the bus.
12. The method of claim 11 wherein the root node solves conflicting simultaneous requests for the bus by granting the bus based on a predetermined adjacent node selection criterion.
13. In a computer system comprising a plurality of components interconnected by a plurality of communications links, said plurality of components each having at least a first communications node wherein said communications nodes interface their associated component with a communications link through a node port, said nodes being capable of having a plurality of ports to which communications links to adjacent nodes couple, each node having a predetermined selection criterion established for selecting adjacent nodes coupled through its ports, said configuration of nodes and communications links comprising a directed acyclic graph wherein one node is designated a root node, all nodes coupled to only one adjacent node are designated leaf nodes, all other nodes in the graph being designated branch nodes, said acyclic directed graph having established hierarchical parent-child relations ships between all adjacent nodes proceeding from the root node down to any leaf nodes wherein a leaf node has only one parent node and all nodes adjacent to the root node are child nodes with respect to the root node but parent nodes with respect to other adjacent nodes, the root node being defined as having no parent node, a method of token passing bus access arbitration wherein a metaphorical token is passed from node to node in a cycle through the graph, the node having the token being the node with bus access, said method comprising the step of passing the token through the acyclic directed graph in an order determined by the predetermined selection criterion each node has established for selecting adjacent nodes .
Patent Citations (13)
| Patent | Date | Inventor | Cited By |
|---|---|---|---|
| US4344134 | 1982-08-01 | Barnes | |
| US4485436 | 1984-11-01 | Austruy et al. | |
| US4698752 | 1987-10-01 | Goldstein et al. | |
| US4897833 | 1990-01-01 | Kent et al. | |
| US5018133 | 1991-05-01 | Tsukakoshi et al. | |
| US5027342 | 1991-06-01 | Boulton et al. | |
| US5301333 | 1994-04-01 | Lee | |
| US5353412 | 1994-10-01 | Douglas et al. | |
| US5434984 | 1995-07-01 | Deloddere et al. | |
| EPX841010838 | 1984-02-01 | ||
| EPX914028097 | 1991-10-01 | ||
| EPX924013410 | 1992-05-01 | ||
| WOX9206436 | 1992-04-01 |
Non-Patent Literature (13)
- Melen et al. "Distributed Protocols for Access Arbitration in Tree-Structured Communication Channels", IEEE 1988 pp. 1539-1546.
- Even et al., "How to Keep a Dynamic Distributive Directed Graph Acyclic and Yet Grant All Requests of Edge Additions", Oct. 22-25, 1990, pp. 414-425.
- Jovanovic, "Software Pipelining of Loops by Pipelining Srongly Connected Components", IEEE, 1991, pp. 351-365.
- Belik, "An Efficient Deadlock Avoidance Technique", IEEE Transactions on Computers, Jul. 1990, pp. 882-888.
- Chaudhary, "A Near Optimal Algorithm for Technology Mapping Minimizing Area Under Delay Constraints", ACM/IEEE Design Automation Conference, Jun. 1992, pp. 492-498.
- Ravindran et al., "Multicase Models and Routing Algorithms for High Speed Multi-Service Networks", IEEE, 1992, pp. 194-201.
- Neilson et al., "A Dag-Based Algorithm for Distributed Mutual Exclusion", IEEE, May 1991, pp. 354-360.
- Bowie, "Distributed Operating Systems", Abstract, Computer Science Conference, Feb. 20, 1975, Washington, D.C., p. 24.
- Cidon and Gopal, "Dynamic Tree Detection in Computer Network", Proceedings of IEEE, Infocom, Apr. 2, 1987, pp. 181-187, San Francisco, U.S.
- Wu, et al., "Prototype of Star Architecture--A Status Report", AFIPS Conference Proceedings, Jul. 18, 1995, Chicago, U.S., pp. 191-201.
- Bokhardi and Raza, "Augmenting Computer Networks", Proceedings of the 1984 International Conference on Parallel Processing, Aug. 24, 1984, Ohio, U.S., pp. 338-345.
- Toshinori Sueyoshi, "Hierarchical Routing Bus", Systems and Computers in Japan vol. 16 (1985) No. 6, pp. 10-19.
- Database Inspec. Institute of Electrical Engineers, Stevenage, GB, Inspec. No. 898064, Bowie `Distributed Operating Systems`, see abstract and Computer Science Conference, 20 Feb. 1975, Washington D.C., US. p. 24.