From Computer Finance, a sister publication
Large systems storage costs dropped nearly 40% in 1995. Was that an aberration or a sign of things to come? Although disk pricing is continuing to spiral downward, new product introduction often distort the price and supply curves temporarily. For instance, in 1994, the market anticipated that IBM would introduce new 3.5-inch RAMAC RAID drives in February. In fact, IBM didn’t introduce the product until June, while volume shipments didn’t occur until 4Q94. The results: when IBM announced the delay, rivals (especially EMC) cut back on discounting. Demand was soft in anticipation of the new IBM offerings. When the RAMAC RAID devices finally reached general availability, pent-up demand was so strong as to overwhelm IBM’s conservative forecasts (which didn’t factor in the product’s delay). Prices temporarily spiked up to $4.25/MB before settling back to reality by 1Q95. IBM is not the only company impacted. In 1995, a similar delay by Hitachi in releasing the new 7700 series forced customers to accept obsolete 7690 models for the first few months, adding another $0.40/MB to the initial $3.40/MB asking price. As production ramped up by this year, HDS 7700s can now be had for $1.70/MB.
The price decline appeal
Such price declines are music to the ears of IT budget planners. As with any commodity product, manufacturing gets cheaper as the product matures. Research by analysts at Meta Group has shown that vendor production costs are declining by an industry average of 30-35% annually. Yet, with ample supplies and intense competition for market share, DASD vendors have been under pressure to cut prices at a slightly higher 35-40% clip. The result is that vendor margins are thinning. Meta predicts that mainframe storage prices will intersect with high-end Unix models by year end. In fact, most of the underlying hardware technology is already there. For instance, IBM RAMAC now uses the same drives as the company’s AS/400 and RS/6000 disk subsystems. Like most RAID systems, much of the components are commodity items that could appear in almost any level of product. According to IBM, magnetoresistive recording head technology has the potential to provide a 60% annual increase in recording density well into the next century. The company is forecasting it will be producing a 3.5-inch drive with 20GB capacity by as early as 1997, with 90GB capacity (equivalent to a full 3390-3 string array) by the year 2000. With all these improvements to its cost picture, IBM still has several critical challenges. The first is manufacturing cost. IBM has hopped on the industry bandwagon to outsourcing commodity parts and has shifted assembly to low labour cost nations such as Thailand, Singapore and Hungary. However, Meta believes that IBM’s manufacturing costs are higher than those of EMC, which has become primarily a systems integrator. EMC realizes that the true return is in the software, microcode and systems integration, which it performs itself. Meta estimates that EMC’s gross margins are 50%, compared to only 30% for IBM in the DASD arena. The next challenge is technology shift. Within the next year, IBM will replace its aging 3990 controller technology with Seastar. This is a critical step for IBM to reclaim its technology leadership in the DASD marketplace. However, as with the introduction of any new product, the release of Seastar will distort pricing and residual values of RAMAC drives equipped with the older controllers. Meta expects residual values of the 3990-6s alone will drop from $40,000 by late 1996 to $14,000 in 1997. For RAMAC pricing, Meta believes that IBM must drop them to $2/MB to keep the combination (with the aging 3990-6 unit) competitive. The recommendation is that any RAMAC drives with the old controllers must be equipped with a minimum of 512MB cache to offer comparable performance to the new systems. IBM’s various leasing arms play a central role in keeping IBM’s prices competitive. The challenge is daunting. Until IBM introduces Seastar, it will continue to be at a technology disadvantage compared to leaders such as EMC. By offering attractive, creative leasing options, IBM leasing is striving to enable IBM to further cut the real cost of its offerings, and cut the risk of customer investment in transitional technology. Meta believes that IBM Credit Corporation and the like will play a growing role in coming years as IBM (and its competitors) try to lock in user loyalty. That will be a challenging task. Most DASD systems today are almost plug and play – the advice must be to shop around and not be hesitant to have equipment from different sources. To handle the Seastar transition in the US, IBM has introduced the RAMAC Technology Refresh Option (TRO). Under TRO, users gain the option to upgrade to ‘current’ IBM technology (e.g. Seastar) by late 1997 at ‘no charge’. TRO isn’t free however; the provision itself carries a 5% premium above standard leases. Meta believes that TRO has little practical value. Historical procurement patterns indicate that users generally do not buy controller technology alone. Instead, they prefer to deploy the controller technology as part of a strategy to replace obsolete complete systems such as the 3380 or 3390. Meta advises users to consider IBM’s lease offerings if book value, budget, software, and environmental issues are concerns. It issues the following guidelines:
* Insist on low-cost purchase: Users who prefer purchasing arrangements should insist on terms equivalent to the present value of IBM lease-financed deals. For instance, users offered the 3990-6/RAMAC subsystem for $1.30/MB on a three-year ICC lease should demand the right to purchase it outright for less than $2/MB.
* Consider extendible leases: Although short-term (e.g. three-year) leases may be attractive, the useful life of DASD is often longer. Many users who sign three-year leases often extend the term or purchase the equipment, resulting in a useful life extending up to five years.
* Consider straight leasing: Users are usually better off signing basic leases without special terms such as the TRO option. These terms often yield few extra benefits for the added cost.
Measuring the impact of RAID
User confusion has arisen over the relative cost-benefits of RAID (Redundant Array of Inexpensive Disks). The most popular forms of RAID on the market include:
* RAID 1 (mirrored).
* RAID 3 (multiple drives with single parity disk).
* RAID 5 (multiple drives, multiple parity disks).
* RAID 6 (offers mirroring, striping).
The relatively high margins of S/390 storage is prompting IBM rivals such as EMC and Hitachi to provide RAID 1 mirroring at marginal cost of $1.50-2.25/MB, which is equal to or less than the cost of more complex RAID 5 implementations. Mirroring is deceptive when it comes to costing. Although conventional wisdom assumes that mirrored doubles the unit cost of a storage subsystem, in reality, the added cost to the vendor is small. Meta estimates the marginal cost of the extra MBs to the vendor is roughly $0.15/MB. Furthermore, more complex RAID configurations such as RAID currently sport heavy processing overhead. Meta expects that by 1997, S/390 and high-end Unix storage pricing curves will begin to intersect. By the 1998-99 timeframe, Meta expects that other RAID configurations, supporting log-structured file systems (like those from Storage Technology) that provide compaction and compression, will become the norm as marginal disk prices decline to $0.50/MB.
Residual value implications
The cold, hard facts are that residual value of large systems and peripherals is becoming almost a thing of the past. Research from the most recent Meta semi-annual report on storage system residual values indicates that by the end of year two, most DASD systems are almost worthless – from a remarketing standpoint. For instance, the residual values for RAMAC model 1 is 7%, while the two-year residuals for models 2 and 3 are both 3%. The values are slightly higher for some 3390 units. For instance, the 3390-B3C was worth 6% of its original value by 4Q95, four years after the model’s introduction. However, the end is near for this model; by 4Q96, it will be worth only 1% of its original list price. However, unless selling or trading is part of the business plan, residual value is irrelevant. The internal value of aging DASD has more to do with the derived business benefits versus the cost of upkeep. If the equipment was purchased (or bought after the end of the lease on favorable terms), it can be used for more esoteric functions. Here, it becomes a factor of how much it costs to operate the equipment versus the business benefits derived from extending the storage system into new areas.
