IBM has announced a significant delay to its latest 3990 disk controller. IBM says the delay is irrelevant because it didn’t plan to deliver the FastWrite software until next year anyway. Alan Knight, chairman of Computer Management Group user group and a practising DP manager, argues that users’ plans should now point them towards the plug-compatible manufacturers.
IBM has recently announced delays in the availability of its new top-end 3990 disk controller (CI No 904). The delays particularly apply to the Model 3 cached version but all models are affected to some degree. The controller was announced last year at the same time as the triple capacity 3380K disk drive. The original schedule for the shipment has been met as far as the drive is concerned but the October 1988 date for the 3990, itself over a year after the announcement, will slip into November or December for the basic hardware (or even later for those users who did not submit orders or letters of intent on announcement). However, those users who require the cached Model 3 version of the controller will have to wait until, as IBM puts it, the first half of 1989. Cynical To the cynical this sounds suspiciously like June 30 1989. The major problem that is causing the six month plus delay is associated with microcode. IBM always planned a gap of about three or four months between the availability of the basic cache functions and the advanced or extended functions, but this gap has disappeared and all the cache functions will appear together – late. The basic functions are those that supply the cacheing facilities already available on the current 3880 Model 23 controllers plus the Cache Fast Write function which provides fast input-output facilities for volatile data sets such as sort work areas. The extended functions consist of Dual Copy, where nominated data sets are automatically written in duplicate without software intervention, and DASD Fast Write which essentially gives cache capability for write operations as well as for read. These new facilities rely in part on the 4Mb of Non-Volatile Storage which is new to the 3990. The ambitious design of the 3990 was impressive on announcement but now seems to be causing problems. The original IBM cached controller, the 3880, was a very simple device and provided modest performance gains at a relatively high cost, in terms of both price and management effort. It was these inadequacies that the 3990 was intended to overcome. However, it now seems IBM misjudged the effort needed to develop and test the sophisticated microcode controlled features of the 3990-3.
It was only recently that responsibilty for disk controllers was moved from San Jose, home of the 3380 disk, to Tucson, home of the 3480 tape system. Perhaps the delays are due to inexperience in the new location. Certainly, the potential data integrity problems inherent in a cached write operation are awesome and very rigorous testing of the function is essential, especially in error recovery situations. Whatever the cause of the delays may be, there are serious implications for potential users. IBM will probably recover the short delays in the supply of the basic hardware by increasing production capacity but the problems associated with the cache microcode cannot be so easily overcome. Many installations will have made their DASD plans on the basis of 3380K and 3990-3 configurations being available as originally announced and will now have to have a serious rethink. IBM has offered various schemes to overcome the users’ difficulties but these circumventions each have their own disadvantages. For example, a user may take the 3990-3 hardware at his requested delivery date but be supplied only with the Model 2 microcode. A suitable reduction in the cost of the box would presumably be made, although how this would be achieved if the equipment were being leased is far from clear. Now should the user take this option it must be assumed that he didn’t buy the box for the performance benefits that the cache features are designed to give. In this case the logi
c behind the original order must be questioned. Alternatively, IBM is offering to supply 3880-D23 cached controllers on short term rental until the 3990-3 with full function microcode becomes available. This option seems more attractive but implies a subsequent upheaval in terms of taking out the old controllers and installing the untried, delayed new ones – something of a gamble. This approach is even less attractive than it seems at first sight since, at the point where the controller swap will take place, the attached disks will presumably contain important data and this must not be corrupted or lost during the swap. Such issues as these are a well known nightmare for operations staff and must be viewed with much disfavour in installations affected by these delays. Silver lining So far all the news has been bad. But every cloud has a silver lining and in this case the silver will probably end up lining the pockets of IBM’s competitors. Several compatible suppliers, such as StorageTek, Amdahl and NAS, see the 3990 as a response to their own efforts in the cached controller field – and with some justification. The current NAS cached controller, for example, has all the features of the 3990 except for the extended functions, which are, anyway, not available from IBM yet. The NAS controller can process operations between channel and controller in parallel with those between controller and device, full tracks are staged into cache on random reads and four tracks for sequential operations. All these features, which have been running in user installations for a considerable time, become available from IBM only when the delayed 3990-3 finally appears in over one year’s time. IBM seems, then, to have pulled a master stroke! It announced a high profile product well before the anticipated availability date, subsequently could not produce the goods, announced inadequate palliative measures, upset an important part of its customer base and finally opened the doors wide to its competitors. (C) Copyright 1988 Alan Knight