|The best of both worlds|
"The structure and management processes are very sophisticated in the mainframe environment, whereas open systems are still in their infancy and trying to catch up," Iñiguez says (for more on Marriott's project, see "Marriott stays with mainframe storage").
John Herbert is another IT manager who never got the message that the mainframe is dead. As storage management supervisor with New Jersey's Office of Information Technology, Trenton, NJ, Herbert directs the strategy for the more than 11TB of storage used by the IBM mainframes supporting most of the state's government departments. Mainframes aren't only alive and well under Herbert's watch--their use is growing as many applications are moved from open systems, back onto more reliable multiple virtual storage (MVS).
That goes against the conventional wisdom mainframes are simply a legacy platform and mainframe-only storage should be folded into omnipresent, multiplatform storage area networks (SANs). Call him an old fuddy duddy, but Herbert's three-person team maintains a separate 7TB compressed mainframe and 4TB open systems SANs, and so far he remains unconvinced that consolidation is a step in the right direction.
"You manage [mainframe and open systems storage] differently and maintain them differently," he says. "I can see the advantages of consolidation, but I don't see us taking advantage of it. Some of the basic stuff is a problem, like dynamic sparing and the nightmare of root security [on Unix]. The mainframe is still a long way from the server as far as disaster recovery capabilities."
Rather than pushing storage onto what are effectively open systems storage arrays with mainframe interconnects, Herbert is using the mainframe as the state's primary data management tool. Open systems servers may stick Web interfaces on top of key Customer Information Control System (CICS) databases, but the data remains safely stored on a Hitachi Data Systems (HDS) 9960 storage array that was recently installed to consolidate several slower, smaller direct-access storage device (DASD) systems.
Everywhere mainframes are used, storage administrators are finding out that despite years of grandstanding by open systems vendors, the advancing pace of technology has not left the mainframe behind. Their market penetration may be dropping--from 13% in 2001 to 8% in 2006, according to IDC--but the downward trend may be more reflective of a growing market than of mainframes' lack of popularity.
Indeed, mainframe-attached storage still accounts for a major proportion of the sales of EMC, HDS, and IBM, which together account for the bulk of mainframe-attached storage systems. Mainframe users are clearly making the most of their investments, and are continuing to add to their storage pools as their mission-critical applications grow.
Aiming to bring those mainframes up to scratch with their fast-moving open systems cousins, mainframe storage vendors have breathed new life into mainframe storage by bridging the gaps between flexible open system SANs and legendarily robust and mature mainframes.
The new ESCON
Foremost among these improvements is FICON, a mainframe interconnect introduced by IBM in 1998, but still in the planning stages at most companies. FICON bundles the equivalent of eight ESCON channels--each of which can support up to 4,000 I/O operations per second--into a standard fiber connection. The result is full-duplex data transfers of up to 100MB/s, compared with just 17MB/s half-duplex data transmission in ESCON environments and 4.5MB/s using old parallel technology.
Architecturally, FICON is similar to Fibre Channel (FC), which has become the standard method for interconnecting storage and servers within open system SANs. Four out of five layers in FICON and FC are identical--the only differences are in their top layer signaling. This allows compatibility with existing FC equipment, which is a significant benefit because FICON allows mainframes to be easily linked with a broad range of FC-capable servers, storage arrays, tape drives and other SAN-related equipment.
FICON also offers better error correction than ESCON, and supports up to 16,384 unit addresses per channel compared with ESCON's 1,024 unit addresses per channel.
Seamless connectivity is an underlying requirement for the integration of mainframes into evolving SANs, yet it's not the only reason FICON is winning supporters. For many companies, FICON's most appealing trait is its ability to maintain data integrity over much longer distances than was possible with ESCON (over 20 km without repeaters and 100 km with repeaters, as opposed to ESCON's maximum of three km without repeaters and 43 km with repeaters). Boosted by EMC's Multi-Hop technology, FICON has been extended thousands of miles across the Atlantic.
The extra distance provides much-needed flexibility in the design of disaster recovery strategies, allowing companies to set up continual data replication policies between primary and secondary data centers. Using remote replication features of the mainframe-attached storage, data consistency and integrity become far easier to guarantee.
FICON has been a boon for Albertsons Inc., of Dublin, CA, a multistate grocer with 2,300 stores in 31 states and sales of nearly $40 billion. With approximately 30TB of mainframe transaction storage and a 35TB Teradata data warehouse for data analysis, Albertsons recently combined its open systems and mainframe administrative groups and has been working to bring its IBM Shark and EMC Symmetrix storage under a single management banner.
The opportunity came last year, when Albertsons built a new data center and upgraded its mainframe connections to FICON. In the new data center, the use of FICON has reduced the volume of cabling by a factor of four and--via an Inrange Fibre Channel Director--provided a common connectivity platform for both mainframe and open systems storage. It's also allowed both types of servers to share Albertsons' StorageTek 3494 automated tape library and 3590 tape drives, and has sped the mirroring of data between primary and backup data centers 12 km apart.
Once 2Gb/s FC technology is available, Bob Lembo, the manager of enterprise storage with Albertsons is confident FICON's similarities will allow it to smoothly follow FC to speed data's movement around the systems. And while it had previously used proprietary IBM and EMC software, Albertsons recently implemented Computer Associates' BrightStor to manage both environments from a single interface.
"Since most of our core applications remained on the mainframe, we had to install the mainframes in the new data center," Lembo says. "Open systems still have a way to go, so we try to deploy technologies where we can leverage one technology for both platforms. We will take advantage of new technology when we need it."
| Marriott stays with
Marriott International, of Washington, DC, has been using mainframes for years. With a considerable investment already put into the systems, the worldwide hospitality giant--which manages over 2,400 properties in 64 countries--is continuing to extend its life with a series of upgrades that reinforce the mainframe's role as the center of Marriott's business. Next year, Marriott will move from its current ESCON to a FICON interconnect that will speed data flow between the mainframe, tape drive and tape silo.
The company's 2TB of mainframe data--supporting a pair of IBM 9672 mainframes--are stored on an IBM Shark disk system that replaced an IBM RAMAC virtual array several years ago. The Shark system also provides around 6TB of data for Marriott's Windows NT environment.
Backup is naturally a major concern for the company, and the need to improve backup speed recently led to an upgrade from a StorageTek 4490 tape silo to a StorageTek 9840 tape drive. The new system's higher-density tapes have reduced the number of backup tapes by a ratio of 23:1, freeing up a large number of slots that will soon be used to back up the company's open systems storage.
In Marriott's configuration, the open systems servers will be piggybacking on the mainframe environment. Data coming from the Veritas NetBackup system, which backs up the company's open systems, will be routed to the mainframe for management and transfer to the tape silo. This process will be restricted to a standalone backup SAN so that other environments are not affected.
Ultimately, Iñiguez sees the mainframe and open systems worlds "converging slowly into an asset that can be used cross platform." But, she adds, "I don't think it's something that's going to happen quickly."
Divide or conquer?
The different experiences of Albertsons and New Jersey's Office of Information Technology reflect the conflict that's facing administrators of mainframe storage. On the one hand, compelling cost equations and the need to upgrade aging equipment is pressuring them to combine mainframe and open systems storage on the same storage array. But on the other hand, doing so exposes mainframes to the still-evolving management of open systems SANs. For this reason, many storage experts are finding it's easier to let the mainframe people do what they know best, and let open systems storage managers watch and learn.
Over the next year, vendors will continue closing the longstanding gaps between mainframe and open systems storage. But their real challenge lies in easing the management of multiplatform storage: FICON provides the connectivity, and the upcoming will see vendors filling in the rest of the equation.
"We've seen a real change in attitude to storage from a mainframe perspective over the last year, but the roadblock is still the management of the storage environment," says Ted Newman, a practice leader in the Systems and Storage Practice of consulting firm Greenwich Technology Partners, White Plains, NY.
Newman believes improving management and technological integration will drive the creation of a centrally managed storage utility model. "Whereas companies previously had only a small number of people knowing how to administer storage, now they can focus on functionality, policies and procedures associated with how you delegate storage to hosts," he says.
Although applications such as Veritas' SANPoint Control and CA's BrightStor are feeling their way towards cross-platform storage management, standardization of management protocols such as Common Information Management (CIM), Bluefin, and EMC's WideSky universal translator will drag mainframe and open system storage onto common ground. That ensures storage administrators can continue to benefit from mainframe reliability while supporting the proliferation of open systems common in most enterprises.
Not all mainframes will make the transition with equal ease, however. Older VSAM-based data structures--which have little resemblance to FC's fixed-block architecture--will have trouble moving into more modern environments. So, too, will less-mainstream mainframe platforms from vendors such as Unisys, Blue Bell, PA. Ninety percent of the storage Unisys ships is directly connected to its MCP- and OS2200-based mainframes. Although those systems can be hooked into SANs, they require a dedicated path through the SAN switches and "don't like to lose their dedicated ports," says Fred Hanhauser, Unisys' director of marketing for storage products, who concedes that continuing support for the two systems is a ponderous effort.
"Any time we add in a device to mainframes--even a new tape device--it takes us a long time to update the drivers," he says. "Generally speaking, it's a year's worth of work. We go to a lot of [effort] to prove the value of applications currently running on our mainframes and I think we'll see these products continue well into the future. But for new applications, customers are buying open systems," such as Unisys' Intel and Windows 2000-based ES7000 servers.
Customers moving away from mainframe environments inevitably face the question of whether it's better to maintain separate, monolithic mainframe storage or move everything to a mixed mainframe and open systems storage array. This approach is more than simply one of convenience: stringent caching requirements of the mainframe tend to favor dedicated storage tuned for mainframe environments, and you can take a performance hit by configuring the box to support open systems as well.
The key to making this decision is to weigh the management benefits of having a single platform against the performance benefits that may be gained by retaining separate mainframe storage. Assessing both options in terms of cost--both directly (acquisition and staffing costs) and indirectly (the potential impact on business performance if transactional systems are slowed)--is one way to compare them.
As open systems servers continue their unstoppable push towards high-end legitimacy, even mainframes are beginning to take a few tips from their less-expensive cousins. It's only recently, for example, that they've gained support for RAID-10, a longtime standby of open systems environments that allows data striping across aggregated volumes.
Some mainframe storage managers are warming to another open systems technique: using inexpensive arrays of disks to provide nearline storage as an alternative to tape. Storage management independent software vendor (ISV) Fujitsu Softek, for one, plans to launch a product letting mainframe storage managers virtualize pools of mainframe and open systems storage across JBOD arrays.
Properly managed, the JBODs can slash storage TCO and boost performance up to eight times compared with using FICON to prop up aging disk arrays, according to Scott Kennedy, Softek's vice president of strategic business development. Such arrays can then be used in place of tape to increase mainframe storage efficiency by speeding access to data stored in hierarchical storage management (HSM) systems.
The Linux factor
Even those changes pale by comparison with the architectural revolution that IBM has wrought by combining the open systems Linux OS with its traditionally closed zSeries mainframes. The significant traction gained by this combination during 2002 confirms that the OS--and not just standalone management platforms--will play an increasing role in the future of mainframe storage.
The Linux zSeries combination saves storage managers from the limitations of conventional mainframe storage. For example, traditional ESCON-attached storage volumes are limited to 27GB capacity, while a SCSI-based FC-attached device can address up to 2TB per volume. This is important for the Linux mainframe's advertised role in running thousands of virtual machines simultaneously.
Server consolidation isn't the only benefit of Linux on the mainframe. IBM's zOS and OS/390 mainframe operating systems were designed for ESCON-only attachment, limiting their relevance in the fiber-based architectures of the future. Linux-based zSeries servers include native Linux drivers that convert between SCSI and ESCON signaling.
Simply using Linux and FICON can improve performance by eliminating signaling overheads, according to Tony Pearson, senior software architect in the Storage Systems Group of IBM, Armonk, NY. In such a configuration, the Linux systems remain oblivious that they're reading from and writing to mainframe storage. The gradual blurring of the lines between mainframe and open systems will eventually facilitate a more open approach to mainframe storage management.
"Storage is taking on a bigger role as the center of the universe for a data center," says IBM's Pearson. "The mainframe is now becoming just one of the many servers that attach to the SAN. People will choose the mainframe as being the best place to run their apps for the mainframe's capabilities. They'll choose Unix or Windows for their advantages, and they'll be able to move applications from one platform from another without having to change their storage approach."