Today, IBM attempted once again to try to recover at least some of the ground its has shamefully lost to EMC in Big Blue’s own captive mainframe market as well as in the broader and more lucrative open systems disk markets. Perhaps it was a bad omen when the web page describing the new web caching software – what everyone else in the world calls a proxy server – for its 3466 Network Storage Manager had many of its links and all of its graphics missing. Eventually, IBM got out the word that it is enhancing its enterprise disk subsystems, inching one step closer to its much-discussed and little-understood Seascape disk initiative. But what IBM has delivered of Seascape thus far really doesn’t change IBM’s competitive stance for the better.

By Tim Prickett

What IBM wants – and what is keeps saying that its customers want – is a data storage architecture that is more flexible and network-oriented than the central host or client/server system approaches that most of the world’s data processing operations still adhere to. Under the Seascape architecture, which IBM formally announced last June, IBM will use its core AIX system software, PowerPC processors, UltraStar disk drives and ADSM hierarchical storage management software to build data servers that can field requests for data from IBM mainframes, AS/400s as well as standard NT and Unix servers. Of course, it is not easy to do this without upsetting current product lines, which is one of the reasons why no one has been able to buy a Seascape data server yet. The other reason is that it is very difficult to build a scalable disk array that can serve both midrange shops and high-end mainframe shops all with the same components. It took EMC two years to get the cost of its Symmetrix array low enough to break into the AS/400, and then, the Unix and NT markets. Before it could do that, it had to recoup all of its controller development investments by charging lots of dough for Symmetrix arrays to eager mainframe customers, who lined up to take advantage of that array’s large blocks of disk cache memory – memory that made transactions faster and therefore allowed customers to forego a very expensive mainframe upgrade if they needed more computing power. That it has taken IBM years to try to get Seascape together is therefore not at all surprising. If EMC can’t do it, IBM probably can’t either. For the moment, the main Seascape product is still the 3466 Network Storage Manager, small Unix server running IBM’s Adstar Distributed Storage Manager software – which is used to make backups of data stored on clients and servers over a network. Right now, most ADSM customers – IBM claims that over 200,000 systems are using it today, which sounds high – are running the software on their main production servers. With MIPS prices where they are on mainframes and AS/400s, the cost of using ADSM is very high. That’s why IBM invented the Network Storage Manager, which includes a rack of SSA disks (up to 288 gigabytes) that act as temporary swap space for ADSM and either IBM Magstar MP or DLT tape drives (up to 13 terabytes) to permanently archive the data that ADSM gathers from clients. With the 3466, the central hosts and servers are out of the backup loop completely. People frequently get confused and think that the 3466 is a traditional disk subsystem. It isn’t. It doesn’t attach to a server, except indirectly through a Token Ring, Ethernet or FDDI local area network, and you would not use it to drive online transactions. Prices for the 3466 range from $61,500 to $512,000, and fat configurations can cost well over $750,000, which is three times the price of a Symmetrix array, which is a standard RAID array but which doesn’t include HSM.

Rejiggered the 3466

On February 9, IBM rejiggered the 3466 to take on a new job, that of an internet/intranet proxy server. Proxy servers are used by both internet service providers and corporations to help speed up the flow of net traffic by keeping frequently requested information proximate, meaning in memory or on disk locally rather than out on the internet. IBM is pushing the proxy-enabled 3466 to ISPs and big intranet customers; however, customers using the 3466 as a proxy server cannot use it as a network storage server – the jobs are mutually exclusive. IBM is tossing in the proxy server for free on the smallest 3466, the model A01. As part of the big storage announcements on the 9th, IBM also dropped its current Ramac Scalable Arrays, the disk subsystems formerly known as StorageTek Kodiaks before IBM became the sole sales agent for that company last year, and replaced the line with a new product that uses 18 gigabyte UltraStar disk drives rather than the 9 gigabyte drives IBM made StorageTek move to when it took over marketing of STK’s Iceberg and Kodiak arrays. (IBM and STK didn’t have Iceberg enhancements ready for prime time, but did talk about some enhancements due soon to give them Unix and NT support; IBM is also expected to tweak the Icebergs to support its SSA protocol to help bring it up to snuff with forthcoming Fibre Channel disks from other vendors.) The new 9396 Ramac Scalable Array 3 can have over two terabytes of RAID-5 protected capacity in a single box. The new subsystem emulates 3380 and 3390 disk profiles and is supported under OS/390, MVS, VSE and VM operating systems. The improved Kodiak can also support an AIX partition suitable for an RS/6000 server if that server is running AIX 4.1.5 or 4.2.1 or higher. The base RSA-3, model 300, comes with the Kodiak controller, 629 gigabytes of disk capacity and 256 megabytes of mirrored cache memory. Customers can add up to 4,096 megabytes of cache memory on the RSA-3, the same maximum as on the older RSA-2 disks, but in this case, it is not just theory, it will be practice. Because the RSA-3 has half as many disk arms per unit of capacity as the RSA- 2 disks did, response times on many interactive jobs can suffer. IBM says that because of improvements it added in the RSA-3’s microcode and in the internals of the new UltraStar 18XP disks, the maximum I/O throughput of the fatter disks are about 15 percent better than the RSA-2s. But IBM’s tests are for old and new Ramacs both with 4 gigabyte caches – no one bought a Kodiak with that much cache, mainly because they didn’t need to because the RSA-2s had lots of disk arms. Mainframe shops rarely fill their disk subsystems above 60 or 65 percent capacity (for performance reasons) and this will not likely change here, either. Although at $703,360 for 315 gigabytes of space – $2.25 per megabyte list price, not included expanded cache memory – they will be sorely tempted. The RSA-3 will be available May 29.

Revised Magstar code

IBM also tweaked its 3590 Magstar software so its fastest and most capacious tape subsystems can attach to HP 9000 and standard Windows NT 4.0 servers. IBM has rewritten its Magstar microcode so it can control an HP 28696A Fast/Wide SCSI adapter on 9000 systems; Windows NT servers must have an Adaptec 2944 SCSI adapter. The revised Magstar code will be available March 20. In addition to the real Seascape products mentioned above, IBM also previewed its Unix and NT support on the Ramac Virtual Array (Iceberg) arrays as well as some new Seascape technology that it is buying from outside network experts to aid in data sharing between mainframes and Unix or NT servers. IBM’s Cross Platform Extension microcode enhancements for the Iceberg will finally put them sort-of on par with EMC’s Symmetrix. Iceberg’s SnapShot incremental backup and Remote Copy disaster recovery software is roughly analogous to EMC’s Symmetrix Remote Data Facility extensions for Symmetrix. The outside technology IBM is previewing under its own InfoSpeed label comes from Computer Network Technology, a high-speed networking company from Minneapolis. The company provides hardware and software that allows customers to move data from mainframes to Unix and NT servers at much faster rates than are possible using clunky SNA or TCP/IP connections. To get the best speeds, customers will have to buy a special ESCON-to-SCSI gateway that can boost data transfers as high as 28 megabytes per second. InfoSpeed consists of the exPress Data Mover software (CNT’s FileSpeed) and Data Gateway hardware (CNT’s UltraNet Storage Gateway). The hardware and software allow bulk data transfers between S/390 servers and IBM RS/6000s, HP 9000s, Sun Sparcs, Sequent Dynix servers and the usual gang of NT servers. Suffice it to say, EMC’s Symmetrix can long since support MVS, NT and Unix partitions and data sharing between those partitions. When IBM delivers CPE for the RVAs (Icebergs) – Big Blue isn’t saying exactly when it will be available – it will unfortunately be available for IBM’s slowest and skinniest disk arrays. The new Kodiaks are considerably faster and fatter. EMC, it seems, won’t lose much business on this one, and Hitachi, which is beginning to understand the open systems disk market, will have a lot to gain as well from IBM’s weaknesses. รก