Open Systems Connectivity To Mainframe Storage Networks.

This article, excerpted from Marc Farley 's book: "Building Storage Networks ", was reproduced with the permission of The McGraw-Hill Companies. Copyright 2000. Osborne/McGraw-Hill.

IT organizations with both mainframes and large open systems computers have tried to improve the ability to share, transfer, and exchange data between these platforms for several years. The ability to do these things supports new applications that cross organizational boundaries and help integrate corporate policies and decisions. Corporate restructuring around the concepts of ERP (Enterprise Resource Planning) systems such as SAP, Peoplesoft, Baan, and others depends on the ability to share or exchange data among the various parts of the organizations. Doing this sooner, rather than later, is a constant goal to help attain a consistent execution of strategic initiatives.

Historically, mainframes and open systems computing environments have been islands of information with relatively slow links between them using file transfer technology over SNA (Systems Network Architecture) and IP data networks. Functions such as NFS have been available on mainframes but historically have not delivered the performance and capabilities one would expect from them due to fundamental differences in machine and network architectures between mainframes and open systems machines.

The phenomenon of the Internet has helped change this scenario as mainframe systems have seen rapid development in Internet protocol handling technology. It remains to be seen how effective the results of these development efforts will be. System, software, and hardware vendors have been touting the ability to integrate mainframe and open systems technologies for many years, but the fact is there is a long way to go. Storage networks in the form of Fibre Channel and FICON may be able to play key roles in improving the current situation. However, the method of sharing and exchanging data between mainframe and open systems I/O channels is still not clearly defined. This article explores some of the technologies and techniques for transferring data through storage I/O mechanisms between mainframe and open systems computers.

Exchanging Data Across Platforms

The ability to exchange data between mainframe and open systems platforms today is somewhat limited. The discussion in this article does not cover data transfers that occur over TCP/IP and SNA (Systems Network Architecture) networks. As an article about storage networking, such methods of transferring data are not really applicable. Nor do they work very well, due to the fact that they involve both processors and lots of inefficient data network processes.

Instead, we focus here on data transfers that occur over storage conduits-- ESCON and SCSI. While it would be great to say that a lot has been done between ESCON and SANs, the truth is, that has not happened yet. Nonetheless, this article treats the subject as if ESCON to SAN is already a reality, and presents much of the information that way in anticipation of the strong currents in the market in this direction. Three methods that have been developed to exchange data between mainframes and open systems platforms are:

* Storage pooling

* Device emulation

* Data sharing 

Storage Pooling Between Open Systems And Mainframes

Storage pooling is a storage management application that aggregates and subdivides disk storage in order to share the raw capacity across multiple systems. Storage pooling can be used to provide storage capacity to both open systems and mainframe computers even though the data structure is considerably different for both machines. A disk subsystem with many devices or the ability to virtual drives can allocate some of its resources for mainframe storage, some for open systems storage, and leave unused resources to be determined. Fig 1 shows a storage subsystem being shared between a mainframe system over ESCON and Unix and NetWare systems over a SAN.

In general, storage pooling does not facilitate data exchanges between virtual devices in a disk subsystem. However, EMC, Hitachi Data Systems, and other companies have developed internal data transfer functions that copy data between mainframe devices and open systems devices within a storage subsystem. This demonstrates some of the potential of intelligent storage subsystems in providing higher levels of data management.

Device Emulation

Device emulation, or virtualization as it sometimes turns out to be, can be used between mainframe and open systems just like it can be used between practically any other combination of systems and devices.

Typically, mainframe storage devices are more expensive than their open systems cousins. Therefore, the ability to implement less expensive open systems devices in place of mainframe devices is appealing to mainframe systems managers who want to implement lower cost products. Not only do open systems devices cost less, but the maintenance and supplies for them also cost less. For instance, magnetic tapes for open systems tape drives tend to be less expensive on a cost-per-megabyte basis than mainframe tape drives.

Another benefit of device virtualization is the ability to substitute software functions running in a system for hardware products. This is a well-known and often-used technique in many parts of the computer industry to simplify installation, configuration, and management efforts.

The use of device virtualization to substitute cheaper open systems devices or subsystems for mainframe devices would probably be widespread if it weren't for the fact that mainframe device I/O controllers pose a few more business and technology problems than their open systems cousins. Emulation of device controller functions in ESCON environments more or less mandates the implementation of ESCON chip sets and driver software that adds cost to the configuration. Furthermore, the rigid structure of mainframe storage requires all devices to work within close tolerances of their expected performance and behavior. In other words, an emulated device has little room for error in mimicking its mainframe counterpart. Testing with these large systems also involves considerable expense for any company wanting to develop emulation products for the mainframe market.

Emulating Mainframe Devices With Open Systems, Subsystems And Systems

For that reason, the emulation of mainframe products tends to involve larger subsystems, or even systems that can provide the functionality desired and spread the development and component costs over multiple real or virtual devices. As mentioned previously, this has already been done with SCSI disk drives in storage subsystems. As a result, the disk drives used in many large mainframe disk subsystems are SCSI disk drives. The controllers for these products have to be able to correctly respond to all the mainframe device commands, including cache requests as well as providing the mapping of mainframe data formats SCSI fixed block devices.

Virtual Tape And The Storage Gateway

Virtual tape is an implementation of device virtualization that substitutes a system and disk storage for a mainframe tape device. Several companies are marketing sophisticated virtual tape systems that run on open system computer platforms and storage. An important component of the virtual tape server is a storage gateway, which connects open systems disk storage resources to an ESCON network.

The idea of the virtual tape server is shown in Fig 2. The virtual tape server presents multiple device images and hundreds, if not thousands, of media images to storage applications in the mainframe. This approach makes it very easy to expand the capacity of mainframe tape storage without having to physically install new hardware--as long as the required disk capacity is available in the SAN. The storage gateway function provides protocol translation between ESCON and SCSI-3. It also provides mappings between mainframe data formats and their respective SCSI device formats.

The storage gateway function could someday become an independent technology that would allow real open systems tape products to be used on the SAN side, instead of disks and disk subsystems. However, this is not likely to occur until the performance of open systems tape devices becomes roughly equivalent to the performance of mainframe tape devices.

Device Emulation: Hot Data Sharing

While the data on emulated devices may be written by an open systems operating system and device driver, it does not mean it is usable by open systems servers. For starters, mainframe data is stored using several different character sets instead of the ASCII character set used by open systems machines. In addition, logical device formats like count key data are foreign to open systems file systems, database systems, and device drivers. For that reason, device emulation is not a realistic technology for providing exchanges of data between mainframes and open systems.

Data Sharing Between Mainframes And Open Systems

The challenges of data sharing are many. The problems of access control, organizational structure, and operating coordination are tough problems that are being addressed by several companies, but could go unsolved for several years in the market. Data sharing becomes even more difficult when it is attempted between such dissimilar systems as mainframes and open systems machines.

Despite the challenges, a mainframe to open systems data sharing solution was developed by Encore Technology, which was acquired by Sun Microsystems. The product that resulted from this was the Sun 7000 storage subsystem. While the product was a commercial flop, it had several interesting internal architectural components; the most interesting feature was its ability to present mainframe data to open systems machines as a virtual CD-ROM drive. The idea of using emulated removable media as a mechanism for sharing data makes a certain amount of sense. Each time the virtual CD-ROM drive was accessed, the data would be read in as if it were removable media. This circumvents the problems of needing to read the data through a file or database system--instead it is mounted and read each time as a piece of standard CD-ROM media. Fig 3 shows the relative functions of this approach.

In retrospect, the whole thing was a bit of a science project that never caught on. Whether it was lack of performance, understanding, or promotion doesn't matter anymore. The product was withdrawn for lack of interest from customers who would pay for it. This might actually have something to say about the requirement for data sharing across platform boundaries.

Marc Farley is the vice president of marketing at SanCastle Technologies (Huntington, NY).