Browse Prior Art Database

Graceful Degradation in a Multiple Data Path Environment

IP.com Disclosure Number: IPCOM000079692D
Original Publication Date: 1973-Aug-01
Included in the Prior Art Database: 2005-Feb-26
Document File: 3 page(s) / 46K

Publishing Venue

IBM

Related People

Frush, DI: AUTHOR

Abstract

Described herein is a new configuration of existing elements which will allow an on-line transaction-oriented system to gracefully degrade, if one of its data paths fails in operation. In many cases, the isolation of the failed path is automatic, requiring no operator intervention.

This text was extracted from a PDF file.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 53% of the total text.

Page 1 of 3

Graceful Degradation in a Multiple Data Path Environment

Described herein is a new configuration of existing elements which will allow an on-line transaction-oriented system to gracefully degrade, if one of its data paths fails in operation. In many cases, the isolation of the failed path is automatic, requiring no operator intervention.

Consider first a typical multipath configuration such as shown in Fig. 1. The host-to-host link is necessary in the event the data path fails. The failure signals the alternate host to initiate alternate path access to the desired data. This is a workable system, but complicated because each host must store address conversion tables, to equate the requested logical data address with the physical data path addresses it possesses.

An improved system configuration is shown in Fig. 2. Note the addition of the channel switches in lieu of the host-to-host link, and the explicit method of interconnecting the switches. It is this explicit cross coupling of the data paths that provides the graceful degradation characteristic, by allowing the still- functioning data path to assume the identity of the failed path. First, the logic of the explicit interconnection will be explored, and then the operation of the interconnection during path failure shown.

The logical addresses of the control units are physically programmed into the units at the time of installation and cannot readily be changed by an operator. Assume for purposes of explanation that the upper control unit in Fig. 2 is to be the "Prime A/Backup B" controller, while the lower unit is to be the inverse - "Prime B/Backup A" controller. The cards for the A Channel of both controllers are plugged to A at installation time, and the upper control unit is physically cabled closer to the channel than the lower. They need not be adjacent, only that the channel cables attach to the upper control unit before they attach to the lower unit. Conversely, the B channel cables physically attach the lower control unit to the B channel before the upper control unit is attached.

Because of the architectural propagation of SELECT OUT, any time the B channel raises SELECT OUT and ADDRESS OUT with CU address B on Bus Out, the lower control unit will find an equal compare with the Bus Out address and its own address. It then turns SELECT OUT around back to the B channel as SELECT IN, without passing it physically further along the string of devices or control units physically attached to the B channel. Thus, selection of Prime B is always accomplished between the B host and the lower control unit, as long as the lower unit is "alive and rational". Similarly, the upper control unit is the Prime A for channel A selection sequences as the upper unit is "alive and rational".

Now, while there are an...