Northwestern University

MISSION CRITICAL DCE/9000

Northwestern University - Evanston, Illinois
Improving Data Access with HP DCE/9000

CHALLENGES

Improve system administration effectiveness and build an open and flexible IT infrastructure that provides an environment for processing social science statistical data while maintaining ability to manipulate data from IBM tape sources.
Migrate data and computational processing from mainframe to UNIX (R).
Maintain ability to process very large files and data sets over two gigabytes in size.
Improve slow response time and provide more cost-effective disk storage.

SOLUTION

Replace obsolete IBM 4381 mainframe with HP 9000/H50 fileserver running HP DCE/9000 and an HP 9000/735 as a general-purpose application server. Implement DCE's Distributed File System to provide uniform name space across the network.
Use DCE to integrate IBM RS6000 as a server for IBM mainframe 3480 tape drives, providing mainframe tape drives to all other application servers on the network.
Use HP Professional Services for hardware, software, and necessary support to implement new infrastructure and tools to enable Northwestern University to take advantage of DCE's services.

RESULTS

Administration effectiveness and flexibility improved through a distributed approach to social science computing that allows individual network segments to be improved depending on faculty requirements.
Improved response time for processing and larger amounts of online storage at lower cost.
Successful shutdown of obsolete mainframe replaced by distributed network of HP business servers running

Hewlett-Packard's DCE. HEWLETT-PACKARD'S DCE/9000 - ENABLING DATA ANALYSIS IN A DISTRIBUTED ENVIRONMENT AT NORTHWESTERN UNIVERSITY

Supporting academic computing for social science research at Northwestern University's Evanston, Illinois, campus can be as exacting a discipline as academic research itself. Chiefly focusing on the manipulation of statistical data, it requires the ability to quickly process very large collections of files while arduously analyzing data sets, some exceeding several gigabytes in size.

Northwestern began to see it would be difficult to provide that same level of exactness across disparate computing resources. Unfortunately, this problem could not be solved by crunching numbers. Instead, real computational control was needed in the form of a set of cross-platform, distributed services. After carefully evaluating its options and testing solutions, Northwestern decided to move to client/server and distributed computing, adopting the Open Software Foundation™'s Distributed Computing Environment (DCE) from Hewlett-Packard.

To help large organizations develop, deploy, and manage applications on disparate heterogeneous networks, DCE provides a common set of services on different types of platforms, offering system administrators and users a consistent approach to a distributed computing environment.

"By exploiting standards such as DCE and UNIX we felt the university would end up with a situation that was competitive, in terms of vendors competing for our computing dollars, and . . . flexible as well. We'd also eliminate a situation where an entirely centralized decision was made every five years to replace the mainframe for a million bucks. Instead, with distributed computing, we'd have a situation where our faculty were spending their own budgets for application servers to improve their own network segments," says Bruce Foster, Northwestern University, Academic Technologies, Research Support.

OVERCOMING MAINFRAME CONSTRAINTS AND IMPROVING PERFORMANCE

In the past, the university had relied on a central mainframe to support social science research computing. The mainframe was an IBM 4381 and ran the VM/CMS operating system. It hosted a variety of statistical software packages - SAS, SPSS, and GAUSS, among others—and was also the home of Northwestern University Library's Data Services archive of data files for secondary analysis.

The same mainframe was an important source of electronic mail services for administrative staff users on the campus administrative SNA network. In addition, incoming data for analysis from government or commercial sources was frequently on magnetic tapes created in older proprietary or non-networked mainframe environments and from IS shops that are sometimes ten years behind current technology.

By 1993 the IBM 4381 had reached the end of its useful lifespan. It no longer could provide a cost-effective means of computing. Due to disk space and CPU speed constraints, the mainframe used for the social sciences had to be replaced at regular intervals. It cost a million dollars every five years, and securing that amount of money on an ongoing basis was a major effort to orchestrate.

Additionally, the mainframe itself was expensive to maintain, and the CPU was slow compared to what a PC or UNIX-based application server could provide. Disk storage was difficult to obtain, so users were feeling constrained. Thus it was clear that the mainframe had to be replaced with a faster, more efficient, fully distributed open systems network of application servers while preserving the capability to manipulate data from tape sources.

As a result of a test-DCE-cell setup consisting of two HP 715-33s and several users, a plan was formulated for a new distributed system consisting of Hewlett-Packard UNIX servers on a network all held together with HP's DCE/9000. DCE was selected as the foundation for core services because it could be accessed from anywhere on the campus network in a transparent manner. DCE's Distributed File Service (DFS) can provide access to a large amount of online disk and optical storage to DCE client systems.

CREATING A NEW SYSTEM ARCHITECTURE USING HP'S DCE/9000

With the test-cell experience in hand and a plan that satisfied the faculty's needs for a file server, a tape service, and application servers, the administration approved the new system to replace the obsolete mainframe. The new architecture would include a file server with 20 gigabytes of storage, an optical jukebox for hierarchical storage management, a tape service that would enable the processing of tapes, and application servers to provide computa-tional function to end users. Users would connect to application servers from their individual PCs or terminals in order to access the tape service and file service, which reside on machines users would never log into. One general-purpose application server would be set up for those users without access to their own departmental application servers. Additional application servers could be added at the departmental level and communicate with the main file server, providing data access, DCE security, and file and other services to 850 users. Tying everything together would be HP's DCE/9000.

Northwestern chose Hewlett-Packard because the university had confidence in HP's understanding of distributed computing and felt that the HP 9000 business servers offered great price/performance. In addition, Northwestern's software was HP-compatible. But the biggest deciding factor was the good track record and favorable relationship that HP had developed with the university over the years. For this requirement, HP invested a lot of upfront time at Northwestern helping with systems configuration.

The new architecture consists of an HP 9000/H50 with 20 GB of fast/wide SCSI II disk storage, a 41-GB optical jukebox providing secondary storage, 192 MB RAM, and a 5-GB DAT drive. It provides DCE security and DFS file services to all the application servers and terminals on the network. The general-purpose application server is an HP 9000/735 with 144 MB of memory and 5 GB of fast/wide SCSI II disk. Tape services at Northwestern were originally shared between the old 4381 mainframe and another campus mainframe, an administrative IBM 3090. Because HP's DCE provides support across different platforms, the IBM tape server could be accessed by any user on the network. Since the tape server could connect to the controller for the 3480 tape drives on the 3090 main-frame, those tape drives could now be shared between the new network and the 3090 as they were once shared between the 3090 and the 4381 mainframes. This eliminated the need to buy 3480-compatible tape drives (at a cost of $20,000 each) for the network. The new UNIX systems provide to 850 accounts all of the important function-ality currently provided by the mainframe, and at a lower cost. These functions include:

The ability to process large files
Access to mainframe-generated tapes
Access to the Data Services archives
Easy access to the wide range of applications used by the social sciences

The new systems are on a 100 Mb/s FDDI ring directly connected to the campus backbone, offering the highest transmission speed possible on the campus network. Because the network backbone is spread between the Evanston campus and the Chicago campus twelve miles away, a wide-area file system like DCE was required. NFS was looked at as a file system, but was rejected for this type of environment. DCE's DFS file system provides further leverage in that in can run across multiple platforms.

One of the benefits from HP DCE/9000 is that there is a uniform name space. One user ID is assigned to each person and applies to all computers that are part of the DCE cell. In addition, the file space is uniform and scalable. Since the file server name is not part of the file space, the addition of file servers doesn't affect the way the file space looks to the end user. Further, as more horsepower is needed, the file service can be replicated as many times as needed - again with transparency to the user. The uniform name space of DCE preserved a substantial investment in tape drives and saved the university more by allowing the tape drive to exist on a smaller, separate server. DCE's uniform name space keeps the password and user names the same on the tape server as on the HP servers. This is important because the tape management software requires a password file that has the same set of user names and UNIX IDs as all the other servers on the network. DCE keeps these files in sync. In addition, the tape software can be installed on only the main DFS fileserver yet be available to all the other servers on the network.

DCE offered the reality of a single network ID and single login situation in a secure environment. This was very important because over the years the university had spent a fortune on account breakins, sharing of accounts, and lost password control.

EXPLOITING THE NETWORK: GROWING THE DCE CELL

The success of HP's DCE at Northwestern University enabled the old main-frame to be shut down. Northwestern now has a wide-area file service that performs faster than the mainframe and provides open distributed computing in a secure environment.

In making the initial investment in hardware and software for this project, Northwestern placed emphasis on more efficient core services - that is, the network, the file server, the tape server, and the infrastructure of hardware and software to support DCE. The general- purpose application server, while adequate to get things started, is not sufficient to handle anticipated growth and demand over the long term. The design anticipates that growth will occur at the departmental level, with faculty purchasing their own workstations to act as application/ computation servers that will be joined into the DCE cell.

This growth has already begun. Individual faculty members and groups of faculty have purchased UNIX workstations and added to the cell. These workstations are primarily HP 712s and 715s, along with some 735s. HP systems are being selected because Northwestern has a critical mass of licensed software and technical expertise, because of favorable pricing, and because Northwestern has confidence in Hewlett-Packard.

More importantly, the system is meeting Northwestern University's objectives for improved CPU speed; adequate, fast, and less-expensive disk storage; and improved system management. The greatest impact is in the system's open, standards-based HP 9000 business servers and the robust HP DCE/9000 software, which gave Northwestern the capability needed to migrate data and application processing from its mainframe to more economical UNIX servers. With Hewlett-Packard as a technology partner, Northwestern is looking to its future. Says Northwestern's Bruce Foster: "Northwestern is at the brink of an exciting new era in computing - for social science research, and for the university as a whole. This is a part of the university's strategy for effective and secure information access in a network environment. Everything that is being done in social science computing is applicable to other computing on campus, both academic and administrative."

Open Software Foundation is a trademark of the Open Software Foundation in the U.S. and other countries. UNIX is a registered trademark in the United States and other countries, licensed exclusively through X/Open Company Limited.

The information contained in this document is subject to change without notice.

Printed in USA M04/96

5965-1415E

MISSION CRITICAL DCE/9000