Tuesday, June 14, 2016

Tandem Computers, Inc.

The Tandem computer was the granddaddy of fault-tolerant systems. Tandem’s first system was delivered in 1976. Forty years later, its original architecture remains the dominant fault-tolerant technology. Then and now, a Tandem system was a loosely coupled multiprocessor system that contained anywhere from two to sixteen independent processors in a node. Up to 255 nodes could be included in a single network, linked via Tandem’s Expand communication network.


The processors in a node were linked via a duplexed, interprocessor messaging bus called the Dynabus, capable of a 26 megabyte/second data rate.

All device controllers were dual-ported so that there was always a path to a device even if a processor failed. All critical processes ran as process pairs in two different processors. One process was the primary process, and one was the backup process. The primary process kept its backup process synchronized via checkpointing messages. Should the primary process fail (presumably due to a processor failure), the backup process took over and continued processing with no apparent interruption to the user. (Tandem’s later inclusion of the Pathway process monitor eliminated the need for application programmers to write checkpointed process pairs.)

With Tandem’s second release of its product, the Tandem NS2, each processor could be configured with two megabytes of memory. Each mirrored disk pair could provide 128 megabytes of memory (yes, that’s megabytes, not gigabytes).

Tandem was acquired by Compaq in 1997, which then was acquired by HP in 2002. Tandem computers are now known as HPE NonStop computers. (HPE is HP Enterprise, one of the two companies that resulted from the split in 2015 of Hewlett Packard into HP, Inc., which sells HP personal computers and printers, and HPE, which markets HP server, storage, and networking systems.) 

No comments: