History (1994): First IBM RAIDs for Mainframes

Before the end of June, IBM will announce 2 fault-tolerant disk array subsystems for its mainframes.

Big Blue is doing things small. It’s the first time that such disks, in the 3.5-inch form factor (2GB each), will be used on large IBM configurations. It could be the first RAID-5 for ES/9000 environment.

In the US, Encore Computer (Fort Lauderdale, FL) had made a RAID-5 announcement that will be used by its partner Amdahl (Sunnyvale, CA).

On its side, Storage Technology (Louisville, CO) didn’t officially announce the availability of Iceberg, unveiled in January 1992! Even if 60 units are actually tested by some customers, the first one known is Kaiser Permanente (Garden, CA).

At EMC (Hopkinton, MA), only RAID-0s and -1s are in the catalog. And, in all the cases, they are based on a double volume 5.25-inch disk.

However, except a coming announcement, IBM could be the first to ship RAID-5s for its mainframes since production in Mainz (Germany) and San Jose, CA, and shipments should start before the end of the year.

Big Blue already offered a RAID-3 and 5, based on its 5.25-inch drives, but only aiming at the small scientific computer market.

From an official internal source at IBM, 2 models in the 939X series are going to be unveiled, but 3 months late, temporarily under the code names of Chelan and Sequoia. Several Chelans will be connectable to the controller 3990 models 3 and 6, when the Sequoia, directly linked to a channel, will include its own integrated controller. They both will have about the same functions and performances.

Close to fault-tolerant IBM 9337 subsystems for AS/400s, they will be RAID-0, 1 and 5. The maximum useful capacity of the subsystems will be 90GB, 25% less than the total disk capacity on account of the necessary space to guarantee fault tolerance.

The subsystem’s footprint will be about one square meter only, and the power consumption and heat dissipation are 50% less then those of current 10.8-inch disks.

IBM is changing its strategy, its new RAIDs will be able to work or not under MVS or SMS.

“In most cases, the specs will be higher or equivalent to those of a 3390-3,” said one of IBM’s sources.

The trouble with the RAID-5, is that it is slow in writing mode, since several writings are necessary for one command, only to code parity bits after their calculation. IBM seems to have not wanted to completely eliminate some writing deterioration to no have to involve, like StorageTek, a huge amount of microcodes which explain Iceberg’s delay.

“The price per gigabyte will be slightly higher than those of Model 3,” adds one of IBM sources.

This article is an abstract of news published on the former paper version of Computer Data Storage Newsletter on issue 77, published on June 1994.