Word reaches us from the US and the UK user grapevines that IBM has been spilling the beans on its storage plans – accompanied by some very loud blowing of its own trumpet. The overall message – rather tarnished by the embarrassing problems with the 3380K drives – is that IBM has got disk technology […]
Word reaches us from the US and the UK user grapevines that IBM has been spilling the beans on its storage plans – accompanied by some very loud blowing of its own trumpet. The overall message – rather tarnished by the embarrassing problems with the 3380K drives – is that IBM has got disk technology taped, and that magnetic disks will see users through to the Year 2000, and however rapidly their storage needs increase, the things will get smaller and more capacious at a higher rate, so users needn’t worry too much about running out of floor space. IBM reckons that storage represents 33% of data processing expenditure, and that average demand for capacity is growing by 50% to 55% a year – and if you’re growing at a faster rate than that, well IBM will be able to keep up and stay ahead of you. Over the next four years, the space needed per Gigabyte of disk storage, IBM promises, will fall by a factor of 10, and the contribution to the greenhouse effect from disk farms will moderate – power required per Gigabyte will fall by a factor of five. How will this be achieved? We’ve talked about the clustered 5.25 platters that are in the plan – for next year now – and IBM is promising that by 1991 a single box containing the controller, an A drive and three Bs will store 200Gb. The upcoming 3390s will have only 10% as many components as the 3380Ks, the ones after those just 1%. The storage density on the 3380D platter was 12M-bits per square inch, the E took that up to 23M-bits, the K to 35M-bits, 100M-bits is the density on the ones due out any week, IBM is developing ones with 200M-bit and is researching 1.5Gbits-per square inch, and reckons that known technology – thin film ion deposition – can be squeezed to 16G-bits per square inch. Compaction Another area that is helping out is developments in compression algorithms – IBM thinks of data as snow and talks of compaction – and says that 3:1 compaction is easy – with the result that a 3380E cost $18 per Mb, a K is $14 per megabyte but would cost $5 per megabyte with compaction. But from some minatory noises IBM is making it seems likely that it will not make its compression algorithms available to all and sundry they will be embedded in DFSMS, so if you haven’t moved over to System Managed Storage, you’ll start losing out on the cost savings that your cannier rivals are seeing. Fault-tolerance A key reason for all the effort going into upping disk storage capacity at San Jose is that IBM reckons there is an enormous untapped market out there – it estimates that of all the information business uses, just 2% is currently on disk and 2% on microfiche, and its medium term aim is to get the other 96% onto disk too! And that means that embarrassments like the one that hit those 3380K drives just must not be allowed to happen in future. It has cast a dark shadow over IBM’s claim of 18 to 20 years’ mean time between failures for its current disk drives, and the company is working on forms of disk fault-tolerance, notably a schema in which each of the 8 bits in each byte would be stored under a different actuator – when data starts to be organised like that, the need for system-managed storage begins to sound pressing. And – hidden from the user there will be a leap in the number of actuators so that the data per actuator will not go up very much: that’s one of the secrets of those clustered 5.25 platters. The cache memory needed in the controller as disk speeds and data volumes leap ahead of the speed of the channels highlights the need for enormous amounts of solid state memory in the controller – cache. Don’t look for controller cache to be merged with Expanded Storage – the bulk store that extends the main memory of the 3090s – that is not in IBM’s plan. Instead, controller caches are going to get bigger and bigger in their own right, one reason why IBM is pushing ahead as fast as it can in 4M-bit and 16M-bit memory chips. The curve the company is working on will see maximum cache in the controller going to between 3Gb and 5Gb from the present 256Mb in the not too distant futur
e – and will be using 16M-bit memory chips by 1993. Controller cache cost $22,000 per megabyte in 1984, today it’s $3,000 per megabyte, by 1995 it will cost just $20 per megabyte, at which point it will not be an optional extra but bundled with the controller. How will these brave new disk farms talk to the processor? The first answer is over considerable distances: the company is working with one of its investments on fibre optic technology – the 25% stake it took in the Chatsworth, California PCO Inc subsidiary of Corning Glass Works (CI No 1,091) that will enable all devices to be installed up to three miles from the processor by 1992. Having been stuck at 3Mbytes per second for eight years or so, the channel data rate is set to start climbing rapidly – 4.5Mbytes-per-second at least when the new 15Gb and 22Gb 3390s are announced, then going from 9Mbps to 18Mbps to 24Mbps over two or three years. Yes, say impatient users, but all the microcode that is needed to enable the 3990-3 controller do its thing is still not available: the answer on that one is that the Cache Write feature is now shipping and the the last module, Dual Copy, will follow before the Alpine ski resorts start opening for business this autumn. No heart for tape IBM has had a long-time love affair with the magnetic disk drive – even if it does insist on insulting it by calling it a Direct Access Storage Device – but it heart just isn’t in tape any more: it reckons that very soon, anything tape can do, disk will do more cost-effectively as well as more quickly. Perhaps tape will nevertheless still need to be used for back-up, and the new compression algorithms and recording formats to be launched this month will enable a 3480 cartridge to store 600Mb, 800Mb and 1Gb against a present 206Mb. IBM has 3.5Gb per cartridge working, and within four years you’ll get 5Gb to 10Gb on a cartridge, and 20 cartridges in an IBM autoloader, against six now – so you’ll have 400Gb on half a string – so who needs an automated tape library? You see, IBM is very embarrassed at the success Storage Technology Corp – a company it thought it had successfully seen off – is having with its automated tape library, and takes every opportunity to rubbish the product. Problem is, IBM’s alternative solution is promises, promises, whereas the StorageTek product is here and now. But IBM really doesn’t like being bested by a competitor, so although it’s not going to make its own robot tape library because it believes the things are too reliable, it hopes to have its cake and eat it by buying in a tape library from a West German company and offering it as a Request Price Quotation, RPQ!