International Business Machines Corporation

William E Hammer

Walter H Schwane

Frederick J Ziecina

Daniel Frank Moertl

Bruce Leroy Beukema

David Otto Lewis

Wayne Melvin Barrett

Harold F Kossman

Thomas S Robinson

David W Marquart

Keith K Slack

John W Reed

Surinder P Batra

Gene A Lushinsky

A plurality of devices attached to a communications bus observe a burst transfer protocol which allows pausing only at pre-determined, fixed intervals of n data words, where a word is the width of the bus. In accordance with this protocol, once burst transfer is initialized the sending device transmits an uninterrupted stream of n data words over the communications bus, after which either the sender or receiver may cause transmission to pause. The sender may need to wait for more data, or the receiver may need to finish processing the data just received. The pause lasts as long as needed until both devices are ready to proceed. This cycle is repeated until the data transmission is complete. The sending and receiving devices do not relinquish control of the bus during a pause, and therefore are not required to re-initialize communications. In the preferred embodiment, after n data words have been transmitted, the sender and receiver toggle interlocking signals that accomplish a handshaking between the two devices. The sender de-activates its signal when it is ready to send more, and the receiver de-activates its signal when it is ready to receive more. Both devices are equipped with buffers large enough to hold n data words, but the buffers need not be as large as the longest possible burst communication.

Apparatus and method for burst data transfer employing a pause at fixed

The flow of work requests in a server driven process to process communication environment is described. Logical connections between processes and bus managers interfacing bus units to an I/O bus are assigned to connection groups for management by the bus managers. Each bus unit has its own connection groups for the logical connections. Bus unit resources are assigned to each connection group based on performance factors, and a series of bus unit messages are used to control the flow of work so that a group which has no more resources will not accept further work requests. The originator of the work requests will resequence rejected work requests and resend them when the connection group has freed up resources. A further mechanism is provided to facilitate work consistent with the server driven architecture when bus units do not have adequate DMA capabilities. Two ways of reversing control of transfer of work requests and data so that the server need not have master DMA capability are presented. Management of storage in a remote processor is used to transfer work and its associated data into storage accessible by a bus unit with slave DMA capability. The slave DMA bus unit then transfers the information into storage is manages. In another way of reversing the flow, a bus unit message is used to make the original server a requestor. The bus unit message contains information which varies the request sent by the requestor. In this manner, the server, which was the original requestor transfers information using its master DMA capability flow.

Bus flow control mechanism

A data management mechanism for a processor system provides for management of data with minimum data transfer between processes executing work requests. Each process has storage areas for storing data associated with work requests. The data is described with descriptor elements in the work requests which indicate the location and length of segments of the data. Data is transferred to a process only if it is required for execution of a work request. Further work requests can be generated by a process executing a work request which reference the data without the process actually receiving the data The segments of data may reside in storage areas of different processors with the descriptor elements of a work request defining a logical data stream.

Distributed data management mechanism for handling a data stream

Remote storage management mechanism and method

Workstations are connected via workstation controllers to two computer systems where one of the workstation controllers is a primary or controlling workstation controller and the other workstation controller is connected to appear to the primary workstation controller as a workstation and is designated as the secondary or standby workstation controller. The standby workstation controller has its line impedance matching resistor connected to function as a line terminator but it can also function as a line driver resistor when the primary or controlling workstation controller fails, the failure of the primary or controlling workstation controller being detected by the secondary or standby workstation controller upon the failure of being polled by the primary or controlling workstation controller within a predetermined period of time. The line impedance matching resistor of the failing primary or controlling workstation controller then functions as a line terminator resistor. The primary or controlling workstation controller and the secondary or standby workstation controller are connected backwards with respect to the connection of workstation to workstation whereby the line impedance matching resistor of the workstation adjacent to the filing primary or controlling workstation controller can function as a line terminating resistor to permit the failing workstation controller to be removed from the system for repair without creating a line termination problem for the workstations and the standby workstation controller which has become the controlling workstation controller.

Workstation takeover control

A data management mechanism for a processor system provides for management of data with minimum data transfer between processes executing work requests. Each process has storage areas for storing data associated with work requests. The data is described with descriptor elements in the work requests which indicate the location and length of segments of the data. Data is transferred to a process only if it is required for execution of a work request. Further work requests can be generated by a process executing a work request which reference the data without the process actually receiving the data. The segments of data may reside in storage areas of different processors with the descriptor elements of a work request defining a logical data stream.

Distributed data management mechanism

The flow of work requests in a server driven process to process communication environment is described. A mechanism is provided to facilitate work consistent with the server driven architecture when bus units do not have adequate DMA capabilities. Two ways of reversing control of transfer of work requests and data so that the server need not have master DMA capability are presented. Management of storage in a remote processor is used to transfer work and its associated data into storage accessible by a bus unit with slave DMA capability. The slave DMA bus unit then transfers the information into storage is manages. In another way of reversing the flow, a bus unit message is used to make the original server a requestor. The bus unit message contains information which varies the request sent by the requestor. In this manner, the server, which was the original requestor transfers information using its master DMA capability flow.

Mechanism and method for transferring data between bus units having

An interprocess data transfer facility provides transfer of data between two processes. Work requests are represented by notes that are placed on a queue of a server process for performing the work. The requestor process which created a work request does not transfer the work request from storage it controls until requested by the server. The actual transfer of the work request occurs without interaction of the requestor. The use of notes which represent the work requests permits complex queueing of notes and hence handling of the requests in the order desired by the server.

Flexible data transmission for message based protocols

In a paged, virtual memory computer system an apparatus is provided for enabling I/O device adapters to request the CPU to allocate or resolve virtual pages into main storage pages as required for I/O data transfers. The I/O adapter provides the channel virtual address information specifying the requested page and channel function information which indicates whether the requested page is to be transferred from secondary storage into main storage or merely allocated as a page in main storage. The channel forms the virtual address information and the channel function information into a page request function event which is stored in an I/O event stack in main storage for retrieval and processing by the CPU.

Growing community of inventors

Rochester, MN, United States of America

William E Hammer