Dismiss
InnovationQ will be updated on Sunday, Oct. 22, from 10am ET - noon. You may experience brief service interruptions during that time.
Browse Prior Art Database

Method for an unintrusive preemptive arbitration scheme

IP.com Disclosure Number: IPCOM000009612D
Publication Date: 2002-Sep-04
Document File: 2 page(s) / 38K

Publishing Venue

The IP.com Prior Art Database

Abstract

Disclosed is a method for an unintrusive preemptive arbitration scheme. Benefits include improved performance.

This text was extracted from a Microsoft Word document.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 55% of the total text.

Method for an unintrusive preemptive arbitration scheme

Disclosed is a method for an unintrusive preemptive arbitration scheme. Benefits include improved performance.

Background

� � � � � Buffer unit (B-unit or BU) streams, especially CPU reads, are essentially latency-sensitive streams. A requirement exists for a method to service B-unit streams opportunistically without affecting the high-bandwidth low-priority (LP) stream.

� � � � � Conventionally, a Dram controller unit (D-unit) services requests from multiple agents to perform read/write accesses to memory. Agents include Hublink interfaces and low and high-priority (HP) graphics agents. LP streams are relatively ­latency insensitive but have huge bandwidth requirements. HP streams (such as display streams) are extremely latency-sensitive but do not require relatively large bandwidths. An arbiter in D-unit (DUARB) fairly arbitrates between B-unit and LP streams. B-unit and LP have their corresponding grant counters that indicate the maximum number of grants far each stream in a given arbitration phase. When there is a HP request or any fixed-priority requests (such as page close and refresh), the fair arbiter goes to a SUSPEND state and services the high priority agents.

Description

� � � � � The disclosed method uses a new feature in DUARB that selectively preempts LP streams with B-unit requests. Preemption occurs only when the B-unit stream is accessing an unrelated bank as the current LP stream. This approach ass...