Slowly but surely, IBM is getting its Seascape disk storage product line together. Although it will still be many months before IBM’s Seascape arrays – code-named Tarpon and announced on June 2 as the Versatile Storage Server – are on parity with EMC Corp’s flagship Symmetrix arrays, Big Blue can finally start to make a […]
Slowly but surely, IBM is getting its Seascape disk storage product line together. Although it will still be many months before IBM’s Seascape arrays – code-named Tarpon and announced on June 2 as the Versatile Storage Server – are on parity with EMC Corp’s flagship Symmetrix arrays, Big Blue can finally start to make a convincing case that it will provide products that can and will compete against those from EMC.
By Timothy Prickett Morgan
While IBM will sell much simpler RAID 5 arrays for its AS/400, RS/6000 and NetFinity lines for years to come and will resell StorageTek Iceberg and Kodiak RAID 5 arrays at least until 2000 (that’s when the when the current deal with STK runs out, but negotiations are underway right now to extend that deal), there is no longer any question that Tarpon is the first in what will likely be a long line of disk storage servers. Tarpon, or rather what Tarpon will evolve into, is IBM’s best weapon to date (other than FUD and price cuts) against EMC’s Symmetrix, which is the undisputed pace setter and market leader in mainframe and open system storage. The Symmetrix design is simple enough in concept: use banks of inexpensive SCSI disks and sophisticated electronics to mimic the expensive monolithic 3380 and 3390 disks used with IBM mainframes. EMC went one step further, putting giant blocks of cache memory on the Symmetrix controller, which had the effect of lowering response times for online and batch jobs, often allowing mainframe customers to forgo an expensive processor upgrade while at the same time offering customers cheaper mainframe storage. Over the past several years, EMC has tweaked its Symmetrix mainframe disk array to work with AS/400s and then Unix and NT servers. Up until recently, EMC marketed two distinct Symmetrix lines – the 3000s for open systems and AS/400s and the 5000s for mainframes, but these days both lines are identical except for software options. Customers with 3000s can add microcode to turn them into 5000s. Symmetrix ranges in size between 70 gigabytes and 6 terabytes of disk space, with cache memories ranging from 512 megabytes to 4 gigabytes. IBM often calls EMC a one trick pony, but some of that is probably jealousy. EMC’s Symmetrix has created a $3bn business. Last year, EMC generated more revenue from the Unix business than it did from the mainframe side – the first time that has ever happened. Indeed, its reseller partnerships with Hewlett-Packard, Bull, Silicon Graphics, Sequent and Unisys (all of which cannot compete with EMC so have chosen to join them) brought in 30 percent of EMC’s sales for the year.
Beefed up controller
EMC moved its mainframe disks into the open systems world; IBM is going the other way, moving its Unix arrays to PC platforms and then to mainframes. In 1996, IBM enhanced its 7133 Serial Storage Architecture Unix disk subsystems to support OS/2, NetWare and NT. The Tarpon is a cluster of 7133s with a beefed up controller that supports multiple incompatible hosts, multiple data partitions and data sharing between like hosts (many servers can share the same copy of data with the Tarpon). Each Tarpon includes two four-way PowerPC SMP servers running a baby version of AIX; the controller uses 332 megahertz 604e processors, which provides lot of MIPS for advanced functions that will eventually be added to the box. (EMC doesn’t divulge what’s in the Symmetrix controller these days, saying only that there are 3000 MIPS of power, more than enough to make the disk array take over jobs once handled exclusively by central hosts.) RAID 5 data protection is not managed by the Tarpon controllers, but is implemented in a reworked 7133 adapter card (which also includes 4 megabytes of mirrored, battery-back non-volatile cache memory). The base Tarpon array comes with 230 gigabytes of SSA disk capacity, 512 megabytes of cache memory and redundant controllers and power. It can be expanded up to 6 gigabytes of cache memory and 2 terabytes of capacity. Eventually, Tarpon will be enhanced with functions now only available in IBM’s RAMAC mainframe disks arrays – namely, Remote Copy and SnapShot. Remote Copy allows customers to keep a copy of live data mirrored on a second set of remote RAMACs; SnapShot allows customers to create a backup of live data without taking the server or the data it uses offline. These provide functions that are similar to those provided by EMC with its Symmetrix Remote Data Facility and TimeFinder software. Both vendors argue incessantly over whose code is better since the differences between the underlying array hardware in an IBM Seascape and an EMC Symmetrix array – disk drives, cache memory, ESCON and SCSI connections and so forth – are becoming less important over time. Increasingly, software is the differentiator in the storage business. IBM, for instance, plans to eventually add its Adstar Distributed Storage Manager (ADSM) software to Tarpon, as well as a Java virtual machine which will allow IBM and third parties to write Java applets to provide even more functions for Tarpon.
EMC still holds a few significant advantages. All of its advanced software is available for all of its disk arrays and available for all hosts that connect to them. IBM’s advanced software (SnapShot and Remote Copy) is only available on RAMAC Virtual Arrays (actually StorageTek Icebergs). The Seascape arrays do not, as yet, support these functions, nor can they handle plain old data serving for mainframes. But they will. IBM says that by the second quarter of 1999, it will be able to connect Tarpon arrays to its mainframes and that by the second half of 1999 it will add clones of the SnapShot, Remote Copy and other software that is currently only available in its high-end RAMAC mainframe arrays. EMC has also been successful in selling its Symmetrix add-on software. The company claims that it has over 1,000 TimeFinder customers and well over 2,000 SRDF customers. (IBM disputes these numbers.) EMC has already provided Fibre Channel connectivity (for HP servers), and IBM won’t have that ready until next year, either. That said, IBM has a few advantages of its own, of course. For one thing, the SSA guts inside Tarpon are faster than the Ultra-Wide SCSI electronics inside Symmetrix. (This does not necessarily guarantee better response times, however.) IBM has been as successful as EMC in selling storage software – it has sold over 2,000 SnapShot licenses and has given away almost as many Remote Copy licenses. Moreover, Big Blue’s Tarpon array offers much more cache memory and the ability to easily add computing power if software add-ons for the array start eating MIPS. IBM makes the PowerPC chips and raw disks it uses, too, so it can afford to cut prices a lot lower than EMC can. And it did on Tarpon announcement day. A base Tarpon with 512 megabytes of memory and 230 gigabytes of storage lists for $250,000. EMC’s Symmetrix with the same amount of cache memory but only 70 gigabytes of storage sells for the same amount – or rather, it used to at open systems accounts before Tarpon came on the scene. Pricing is, of course, all over the map on data storage arrays. On the street, Symmetrix has been able to command a price premium compared to IBM mainframe arrays, and this isn’t going to change until IBM delivers mainframe support for Tarpon next year. But in midrange accounts, EMC just lost its price edge. Analysts we spoke to expect street bids for both Tarpon and Symmetrix to be between $.80 and $.90 per megabyte with reasonable amounts of cache memory on the controllers – a far cry from the $1.25 per megabyte that EMC had enjoyed last year. In hotly contested sites where IBM is pressing to regain account control or expand into EMC’s partner’s bases – mainly HP, Data General, SGI and Compaq – prices could hit the $.50 per megabyte level. IBM has had to cut prices to this level on its Multiprise internal disks (RAID 1 only) to compete with Symmetrix. The question is, can EMC go that low without losing its shirt? Up until now, it has been able to walk away from deals with low prices, but EMC may not be able to do that soon, and it certainly won’t be able to do that next year. Perhaps that’s why EMC has committed to spending $1bn of R&D funds over the next three years to enhancing Symmetrix. The only way EMC will be able to compete will be to shift the argument away from price to features. The good news for both EMC and IBM is that the open system storage business is growing so fast that both vendors will be able to grow, most likely at the expense of Sun, Compaq and other midrange vendors who cannot commit the resources that either of these giants can to developing advanced storage servers.