Browse Prior Art Database

Method for servicing multiple WQE requests simultaneously in an Infiniband HCA

IP.com Disclosure Number: IPCOM000010973D
Publication Date: 2003-Feb-05
Document File: 3 page(s) / 91K

Publishing Venue

The IP.com Prior Art Database

Abstract

Disclosed is a method for servicing multiple work queue element (WQE) requests simultaneously in an Infiniband hybrid channel allocation (HCA). 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 50% of the total text.

Method for servicing multiple WQE requests simultaneously in an Infiniband HCA

 

Disclosed is a method for servicing multiple work queue element (WQE) requests simultaneously in an Infiniband hybrid channel allocation (HCA). Benefits include improved performance.

Background

              Infiniband is a trademarked name owned by Infiniband Trade Assn. Specification 1.0 has a release date of October 24, 2000.

              During the processing of Infiniband architecture work requests, engines and arbiters perform many tasks that require a relatively long time to complete, including:

•             Work request entry (WRE) fetches from host memory

•             Address translation requests that require host memory access

•             Queue pair context (QPC) accesses to side random-access memory (RAM)

•             Message context accesses to side RAM

              These tasks are typically serviced serially for a given work request before beginning the next work request. As a result, large latencies occur between work request completions. A requirement exists to reduce the latency to improve performance.

Description

              The disclosed method is the servicing of multiple work queue element requests simultaneously in an Infiniband hybrid channel allocation (see Figure 1). The method is completely developed with digital hardware as apposed to microengines and/or microcontrollers. The method depends on the definition of queue pair context and message context so that multiple WQE fetches may be processed concurrently with minimal restrictions.

              The key elements of the method include:

•             Multiple digital hardware engines that independently service WQE fetches

-         WQE transaction port (WTP)

-         WQE fetch engine (WFE)        ®..