International Business Machines Corporation

Terrence K Zimmerman

Robert S Capowski

Daniel Francis Casper

Matthew A Krygowski

Darwin W Norton, Jr

Clifford T Williams

Frederick T Blount

David S Wehrly

Henry R Brandt

Joseph M Kusmiss

David L Silsbee

Harold F Cavagnaro

Eric T Shalkey

Robert F Geller

Arlin E Lee

Lewis W Wright

Richard C Bono

Lawrence R DelSonno

This embodiment provides apparatus and method for implementing a High Speed Link (HSL) such as the newly proposed ANSI High Performance Parallel Interface (HIPPI) standard on processors complexes like the IBM 3090 having a paging store with an independent bus. A high speed link adapter (HSLA) including input and output buffers and controls is coupled to the independent bus under program control. Program access to high speed link is obtained by an extension to the Page-in and Page-out instructions.

Computer system high speed link method and means

A priority circuit handles requests by three components of a data processing system for access to several resources of the system that can be accessed one at a time on each operating cycle of the system. A logic circuit receives requests by the requesters and grants access to one requester on a priority basis. The logic circuit has means for establishing a particular priority sequence, and the priority circuit includes means for stepping the logic circuit through a cycle of different priority sequences. In a repeating cycle of these steps, each requester is given the highest priority at least once. In a specific embodiment, the stepping means is a counter and a cycle is called a counting cycle. The stepping means is responsive to a control code to establish a particular stepping sequence. In a longer sequence, the lowest priority requester is given the highest priority once and higher priority requesters are given the highest priority several times so that in a complete cycle of steps the requesters have a different relative priority. In shorter cycles, the lowest priority requester is still given highest priority once and the higher priority requesters are given highest priority a few times so that in the shorter cycles the lowest priority requester has a higher relative priority. The priority circuit also includes means for matching a request for a resource with the availability of the resource so that a requester contends for priority only if the resource is in fact available.

Priority circuit for channel subsystem having components with diverse

A channel communicator CC has a storage array for holding an entry for each channel and for holding a busy vector and an interrupt vector that each have a bit for each channel. The CC is connected between the input bus and the output bus that connect the channels and an I/O processor IOP to processor main store. A message to the CC includes the ID of the channel that the message is to or from and the CC uses this ID as an address for accessing a channel entry or a bit in a vector. The message also carries a command that controls the CC to store a data portion of a message or to fetch a channel entry or a vector from the array and load it onto the output bus addressed to the IOP or to one of the channels. The command also controls addressing a particular one of the vectors. The CC also has means responsive to a command to perform a sequence of operations for testing the vectors and for testing fields of an entry in the array and for using the results of the test to control the execution of the command.

Channel communicator

Storage access requests are forwarded from plural input/output channels to shared main storage. An address word in each request designates the identity of the source channel (CHID) and 'destination' address (of a doubleword space in storage relative to which one, two or four 'data' words shall be transferred). EOT tag signal provides demarcation of requests and also uniquely identifies '1-wide' input (Store) requests. Quadword (QW) tag, presented with '4-wide' requests, enables the storage access system to use a single address in the request to locate two contiguous doubleword spaces in storage. Data tags (D1, D2), presented on a selective basis enable the access system to selectively steer (reorder the positions of) data words in an input request relative to word halves of the addressed space. Consequently the data repositioning function associated with Read Backward operations does not require reordering of data words in the channel buffers (saving handling time and expense of 'steering' circuits in the individual channels). In association with a '1-wide' input request a single data tag, D1 or D2, is used to steer the single data word of the request into either half of the addressed space on a selective basis.

Request forwarding system

The Channel Bus Controller (CBC) transfers information between groups of input/output channels and processor storage. Storage receives or dispenses two data words per access operation. Interfaces for transfers from the channel groups to the CBC are advantageously one word wide; since each output (fetch) request consists of a single request word. Information sent by each group is assembled into three-word units (a request word and zero, one or two data words) in a respective channel bus assembly register (CBAR). The assembled unit is passed from the CBAR to a respective area of an In Buffer array and from that array to storage. Zero filler words are inserted into unused data word positions. A channel request may be tagged to designate a transfer of four data words. If the transfer is an input the four data words are sent to the CBC with a single request word. The third and fourth data words are written in the CBAR over the first and second data words as (or after) the unit formed by the request and first and second data words is advanced to the In Buffer. The same request and the third and fourth data words are transferred as a second unit from the CBAR to the In Buffer. The low order bit in the address part of the request is inverted by the CBC to designate the 'next' storage address. This saves time by eliminating a request unit transfer from the source channel group. Request transfers from a group are permitted when a vacancy exists either in the respective CBAR or in a respective area of the In Buffer. Outputs from storage (acknowledgments of data inputs and fetched/output data) are returned to the respective channel group via a respective area of an Out Buffer array. Returns to a group are ordered in the input sequence of respective requests to the CBC although the requests may be applied to storage in another sequence. Tags generated by the CBC are used to maintain the correct output order without delaying evacuation of the In Buffer. The area partitioning of the in and out buffer arrays provides balanced group access to storage and simplifies handling of group traffic. Channel identity information in the request words is looped through the buffer arrays and returned to the channel groups with respective outputs. This permits the CBC (and storage) to ignore channel origins of group traffic and thereby further simplifies handling of traffic.

Channel bus controller

Growing community of inventors

Red Hook, NY, United States of America

Terrence K Zimmerman