ENGINEERING

Atipa Technologies, the high performance computing (HPC) division of Microtech Computers, Inc., which has been successful in designing and customizing high-end workstations and servers since 1986 and cooperates in an ISO-9001 certified manufacturing facility, is an active supporter of the open source community and leading solution provider in high-performance computing. Atipa Technologies successfully deployed two identical Itanium 2 Linux Clusters to the Department of Energy. Each Cluster consists of 536 Intel Itanium2 CPUs, with 12 terabytes of usable fault-tolerant I/O storage. Atipa Technologies successfully met every requirement set by the customer, including a 90-day delivery cycle. A target for this project was to maximize the performance delivered while staying within budget. Atipa Technologies collaborated extensively with Intel to achieve the biggest bang for the customer buck in providing on-time delivery and superior quality at the best price. The company designed each Itanium Cluster with 200 Intel Tiger II 2U chassis installed with dual 1.3GHz Itanium 2 Madison Processors for the compute nodes. Additional compute nodes comprised 34 Intel Tiger IV 4U chassis installed with quad 1.3GHz Itanium 2 Processors for each cluster. The clusters were set up identically and shipped to two different locations within one week. “Working closely with Intel on this project has given us a totally new perspective of Intel. It allowed us to work with each level within each department, to be more efficient in meeting our customer’s needs,” said Mike Zheng, president of Atipa Technologies. “I am also excited about our future projects that will build on the working relationships we have established within Intel.” Zheng added that significant pricing and delivery advantages can be achieved by this cooperative relationship. The project required a system that would support a long useful life cycle with strict guidelines for formal software verification, qualification, and version control. To fulfill these needs, Atipa created a stable, standards-oriented software environment that would minimize future conversion costs of new systems. RedHat Enterprise WS supporting Itanium 64-bit architecture, 64-bit Intel C, and Fortran Compilers were installed on all nodes. The system of four NFS/IO nodes, each able to sustain 70MB/second writes and 90MB/second reads – for a total of 290MB/second on reads, exceeded the required rate by over 40 percent. The two clusters contain both Gigabit networking and Myrinet interconnects. Using the Myrinet interconnect, each cluster attained a TOP500 rating of 1810 GFLOPS, placing each machine around 100th on the current TOP500 list. The system is also required to tolerate failures of disks in such a manner that it is easy to reconfigure and remove failed components and continue operations with minimal changes. The basic Atipa NFS IO setup used for each of these clusters has four NFS I/O nodes and four external Atipa Fiber to SATA Qualifiers with a total of 12TB of useable space using RAID-5. Each Atipa Qualifier is built with sixteen 250GB SATA drives, all hooked to a Brocade SAN switch to allow all NFS I/O nodes to access all disks if necessary. The Software Engineering Support Group at Atipa Technologies has seven full-time engineers who are devoted to supporting customers in assuring quality compliance with all of Atipa’s contracts and specifically trained to emphasize core areas in the HPC market. The kernel engineers are dedicated to implementing value to excel in HPC interconnects, parallel, distributed, and gridlevel computing. The cluster engineers are specialized in developing fault-tolerant architectures including services and systems, and the applications engineers are focused on understanding customers’ application requirements and building a cost-effective infrastructure for their open source solutions. “Our extensive research in the development and design of robust Linux cluster solutions for data-intensive applications shows the necessity of including storage systems that complement the computer cluster's throughput while exceeding the price-performance characteristics of our competitors,” noted by Atipa’s Software Engineering Support Group.