Browse Prior Art Database

High performance topology for clustering of storage devices

IP.com Disclosure Number: IPCOM000010462D
Original Publication Date: 2002-Dec-05
Included in the Prior Art Database: 2002-Dec-05
Document File: 1 page(s) / 8K

Publishing Venue

IBM

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 58% of the total text.

Page 1 of 1

High performance topology for clustering of storage devices

    To buildup high capacity, high speed storage system from aggregation of storage devices, 3 types of network topology are in use currently. First one is BUS type in which devices share a single media. E.g. SCSI. Total throughput of this system is equal to or less than media speed. Second one is LOOP type in which devices are on a loop and every nearby two nodes are connected by a media. E.g. FCAL. Again total throughput of this system is equal to or less than media speed. Third type is STAR type with switching device. In this case every pair of nodes are connected by a media and total throughput of this system is equal to or less than media speed times N where N is number of connections concurrently exist on the network. Third one has advantage in its performance speed but switching device is a rather expensive equipment. Typically, first two approaches are used in small and low cost system and third approach is adapted in large and high cost system.

The idea described here proposes new type of approach that can be adapted in large but low cost system with high performance speed.

Every storage device equips 3 or more physical interfaces for point-to-point link between devices. Typically, high speed serial link will be used. Devices and hosts are linked each other and built up in 2D or 3D lattice structure. Data write command is issued from a host on a lattice point. User data in a command is divided in fi...