Distributed shared resource management

• US 20040215773A1
• Filed: 04/24/2003
• Published: 10/28/2004
• Est. Priority Date: 04/24/2003
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Strait, Gary E., Van Huben, Gary A., Walters, Craig R.
• Granted Patent: US 7,383,336 B2
• Time in Patent Office: Days
• US Class Current: 709/225
• CPC Class Codes: G06F 9/50 Allocation of resources, e.g. of the central processing unit [CPU]

Abstract

A method for processing data in a computer system using two main concepts for addressing this situation, from which numerous other implementations is achieved using a first and second main concept. The first is a method of managing a common data path among a plethora of facilities with a decentralized distributed management scheme. The second concept is a method for managing a shared data buffer or group of buffers between multitudes of facilities. By employing the concepts discussed in this invention, one can contemplate a complex dataflow consisting of a multiplicity of resources and data paths, whereby virtually any combination of sharing is possible. A single data path can be shared among multiple sources or sinks. A single resource can be shared, but may have many separate data paths, or all elements of the present invention can be combined to comprise a dataflow with shared resources and shared data paths, both of which are actively managed by completely independent controllers on a separate chip or on the same chip, but with a segregated dataflow.

Claims

1. A method for resource management, comprising the steps of providing a decentralized, distributed resource management in a computer system and then sharing data in the computer system with a segregated dataflow.

2. The method according to claim 1 whereby a shared data path is managed by a multitude of decentralized, distributed priority stations in a centrally pipelined system.

(a) the distributed priority stations have no method of direct communication between each other.

(b) all communications with the central pipeline use existing signals

(c) requesters requiring use of the shared resource are required to make a pipe pass to initiate use of said shared resource

(d) said pipe pass results in information being broadcast from the central pipeline to requesters and priority stations in the system

(d) said broadcast is used to trigger independent, free-running blocking logic in said priority stations to prevent use of said shared flow resource until it is available.

3. The method according to claim 1 providing the steps for remotely managing one or more data flow elements shared by a plurality of homogenous requesters wherein said requester

(a) comprises a local means of reserving said data flow elements on behalf of one requester while

(b) initiating a synchronous data transfer using a simple command initiated through a synchronous pipeline without requiring advanced reservation or arbitration within said dataflow thereby allowing simple data elements such as data busses, buffers, registers, DRAMs, etc. to be shared and

(c) said requester initiates pipe passes whereby coherency actions such as directory accesses, address translations, etc. may be performed simultaneously with said data transfer and

(d) wherein said data flow initiates data transfers among said shared data flow elements using simple data movement controls without knowledge of which requester the data belongs to and without the need for collision avoidance or detection logic and

(e) wherein said plurality of requesters can process storage access operations simultaneously thereby allowing each requester access to said data flow element utilizing a common priority mechanism such as rank or round robin comprised within said local reservation means.

4. The method according to claim 1 wherein a plurality of homogenous requesters employ a data flow comprising any combination of dedicated and shared data paths and storage elements such as, but not limited to, a single shared storage element with dedicated data paths for each requester, a single shared data path with dedicated storage elements for each requester or a single shared data path and shared storage element.

5. The method according to claim 1 providing the steps for remotely managing one or more data flow elements shared by a plurality of homogeneous requesters and providing a means for said data flow element to be also be shared by an external interface controller operating completely independently of said homogenous requesters wherein

(f) said homogenous requesters initiate synchronous data transfers utilizing a synchronous pipeline while

(g) said external interface controller initiates asynchronous data transfers and

(h) wherein a local reservation facility exists to remotely manage asynchronous and synchronous data transfers in a manner by which collisions are prevented and

(i) wherein no additional communication is required between said homogenous requesters and said external interface controller beyond a simple request and grant semaphore to gain access to said external interface.

6. The method according to claim 1 wherein a synchronous pipeline is used to initiate one data transfer and wherein subsequent data transfers are queued up to enter the pipeline as soon as the present data transfer completes.

7. The method according to claim 1 wherein said local reservation facility permits subsequent homogenous requesters to enter said synchronous pipeline in order to perform non-data coherency actions while a simultaneous synchronous data transfer is in progress by a second homogenous requester.

8. The method according to claim 1 wherein said local reservation facility permits a sequence of synchronous data transfers to occur using a multitude of shared data path elements in order to maximize efficiency by allowing a write operation for a second homogenous requester to begin using a shared data path element while a simultaneous read operation for a first homogenous requester is still in progress.