FLASHBACK!

THE PATH TO THE 150

v2-4/8/96 Archival page

---

It was the emergence of the federal High Performance Computing and Communications (HPCC) Initiative that opened the path to the 150, and a successful proposal from Oak Ridge National Laboratory (ORNL) to the Department of Energy (DOE) Office of Scientific Computing (OSC) that comprised the first step onto this path. The proposal established the ORNL Center for Computational Sciences (CCS) as a DOE High Performance Computing Research Center (HPCRC) and the Intel Supercomputer Systems Division (SSD) as the major computer supplier.

However, the technical path along which the CCS moved initially to reach "today," with its Intel Paragon XP/S 150 centerpiece, had a very different origination - the first machine was a 32 processor Kendall Square KSR1, delivered in September of 1991, prior to the existence of the CCS. At this time there was a Cooperative Research and Development Agreement (CRADA) in place between the Intel SSD and ORNL, which would soon (September 1992) bring to the CCS the first of the Intel Paragons, a 66 node XP/S 5 with the GP architecture (two processors per node - one for computing and the other for message passing).

In June 1992, ORNL and the Intel SSD signed a contract that was to culminate in the Paragon XP/S 150, but this machine was to be preceded by a smaller machine that would return to Intel upon the 150 delivery. This smaller machine, the Paragon XP/S 35, also of the GP architecture, had 512 nodes with 16 MB of memory per node and was also delivered in September 1992. In this same month, our Kendall Square machine was expanded to a 64 processor, two-ring KSR1. This Kendall Square machine has been an important component of our computing capability since.

With the XP/S 35 as our computing "workhorse," we were in a position to appreciate in action the excellent computational environment we had assembled. The small XP/S 5 was the machine for code development, for initiating users into the realities of parallel computing, and for system and computational experiments, while the XP/S 35 was reserved for those who had demonstrated the requisite level of expertise and were prepared for major computational projects.

To preserve this highly effective arrangement and to extend our computing capabilities, we renegotiated the contract with the Intel SSD in early 1994 to increase the memory of the XP/S 35 to 32 MB per node and to transfer ownership of this machine to ORNL. The XP/S 5 would then remain as the "companion" machine for the XP/S 35. Preserving this hierarchical arrangement was a move of major importance for the CCS.

At this point, Intel modified the structure of its forefront Paragon nodes to the MP form, three processors per node, two for computing and one for message passing; this was to be the architecture of the XP/S 150. As part of the aforementioned CRADA, we began receiving an MP-node machine in May of 1994, thereby providing our users an opportunity to familiarize themselves with, and to develop strategies for more effective use of, the 150. This machine has grown and is now a 96 MP-node (the XP/S 14) machine with all nodes having at least 64 MB of memory.

Recognizing the importance of the CCS parallel machines for DOE Grand Challenge and ORNL problems, and the necessity of more computing power for the Laboratory, ORNL scientists, engineers, and management worked together to bring the 16 node IBM SP2 computer to the CCS in October 1994.

And then we were there. On 4 January 1995, the Intel SSD (now the Scalable Systems Division) delivered the Intel Paragon XP/S 150 to ORNL following the successful completion of an arduous set of acceptance tests. We now have a computing power perhaps peerless in the U.S., and anticipate computational achievements on this same scale to extend the often dramatic results from the recent past. We also emphasize that the 150, together with the XP/S 14, provides the highly effective hierarchical MP computing environment that we initiated with the XP/S 5 and XP/S 35 combination.

But, the story does not end here; computing power of this magnitude requires data storage and network systems of corresponding character.

DATA STORAGE SYSTEMS

Working collaboratively with the DOE Atmospheric Radiation Measurement (ARM) Program, we have assembled an extraordinary storage system. Currently based on the NSL-UniTree software, the CCS/ARM storage system is a state-of-the-art, client/server file storage system. System components include:

• Storage Technology 4410 tape robot - capacity 4.8 TB;
• IBM RS/6000 computers (two) as storage system controllers;
• Maximum Strategy RAID (Redundant Array of Inexpensive Disks) system - capacity 108 GB and soon to be 216 GB;
• CREO optical tape storage system (1 TB per Tape!); and
• The Storage System Centerpiece IBM 3495 tape library. Later this year, we will be the first recipient of the new high density tape system with ESCON channels, increasing the capacity to over 100 TB.

COMPUTER ROOM NETWORKS AND
CONNECTIONS TO THE "OUTSIDE" WORLD

Essential to all of our computing are networks to provide for quick and efficient movement of information. The CCS uses many different types of networking hardware to provide access to our machines, as well as to the CCS/ARM storage system.

Within the computer center, High Performance Parallel Interface (HIPPI) networks provide 100 MB/s connections among our Paragon system. This is the current state-of-the-art in high speed networking in a computer center environment.

Connectivity between the computer center and our ORNL users occurs through both FDDI (100 Mb/s) and Ethernet (10 Mb/s). ORNL is connected "to the world" via two networks, both part of the Internet: The DOE Energy Sciences Network (ESnet) provides 45 Mb/s connections to other sites within the DOE community; and the Southeastern Universities Research Association Network (SURAnet) provides 1.5 Mb/s connections to universities in the southeastern part of the country.

THE INDUSTRIAL CONNECTION

The Computational Center for Industrial Innovation (CCII), a key component of the CCS, serves as the centerpiece of ORNL's computationally oriented industrial interactions. Established as DOE National User Facility in August 1994, the major emphasis of the CCII is high-performance computing and its application to industrially relevant problems. High-performance computing for diverse industrial applications is vital to the well being of U.S. enterprises. With 80% of U.S. industrial jobs belonging to small or medium organizations, such as environmental and manufacturing companies, the need for accessible, affordable, technical and computational resources to solve real, large-scale problems in analysis, design and manufacturing is immediate. In CCII, ORNL provides the vehicle through which U.S. industries can collaborate with computational scientists and engineers on a cost-shared basis to solve technical problems.

The CCII's major objectives are to provide U.S. industry with:

• Convenient access to MPP through a high-performance facility;
• Ability to easily interface and collaborate with ORNL scientists and engineers in research and development (R&D); and
• Coordination between high-performance computing and applied scientific R&D to generate timely and innovative technical solutions to real-world problems.

CCII PARTICIPATION

Potential industrial partners should contact the CCII Director. A determination of the nature and extent of an initial project can be made at that time.

If the proposed project is deemed introductory and nonproprietary, the vehicle for project initiation should be a collaborative project with ORNL scientists or engineers. The industrial partner will be designated as a CCII Participant.

If the initial effort is to be extensive or proprietary, one of the following formal Affiliation Categories will be utilized: CCII Associate, CCII Member or CCII Partner.

The CCII utilizes the following facilities within the CCS:

• Intel Paragon XP/S 5, XP/S 35 and XP/S 150;
• Kendall Square Research KSR1-64 supercomputer;
• High-speed workstations and VizLab; and
• Classrooms, office and conference rooms.