Cluster Computing:

ATIPA SELECTED TO PROVIDE TWO ITANIUM CLUSTERS TO DoE
by Tim Curns, Editor

Atipa, solution provider of high-performance computing systems, has made several significant advances in the cluster computing industry recently. The company successfully deployed two clusters containing 1072 Itanium 2 CPUs for Department of Energy. HPCwire recently spoke with Dana Chang, Vice President of Technology, and Roger Heflin, Atipa's Principal HPC Engineer, to discuss Atipa's recent challenges and successes.

HPCwire: Tell us a little bit about Atipa, please.

Dana Chang: Atipa is an ISO 9001 quality certified manufacture for HPC solutions, from hardware to software. Our kernel engineers are dedicated to implementing value to excel in HPC interconnects, parallel, distributed and grid-level computing. Our clustering engineers are specialized in developing fault-tolerant architectures, services and systems, and our application engineers are focused on understanding customers' requirements, and building a cost-effective infrastructure for their open source solutions.

HPCwire: So what's happening now with Atipa? You recently announced the deployment of two Itanium 2 Linux clusters to the Department of Energy.

Roger Heflin: We delivered two large Itanium clusters to the D.O.E.. The press release documents what we shipped in more detail, but it came down to two clusters totaling 1072 CPUs, each has 536 Intel Itanium 2 CPUs, roughly 234 nodes and 12 terabytes of disk space. We delivered it to them - their agreement was - after receipt of PO, we had 90 days to obtain the hardware, configure the software and hardware, and deliver and install the hardware on their different sites. The new engineering staff was able to implement all of that and install it and put it on their site, on two sites within a week in different cities. We have successfully exceeded the contract's performance requirements using Myrinet interconnect. Right now, both clusters attained a TOP500 rating of 1810 Gflops, placing each cluster around 100th on the current TOP500 list. Atipa engineers are proud to be in this large Itanium project.

DC: Throughout the 90-day configuration time, Atipa's President, Mike Zheng, devoted himself completely onto this project, to make sure all hardware was configured correctly. It's rare that the president of a company would pour his heart into a project and be very determined that it must succeed. He is constantly on the phone or email with clients to confirm every little detail because he understands this project is not just a number to Atipa, it's an honor of customer's trust.

HPCwire: Do you know what the clusters will be used for?

RH: I'm not sure how much I can say about that. The DOE's website would say what can be said about that!

HPCwire: How did you work with Intel on providing the DOE clusters?

DC: Atipa collaborated extensively with Intel to achieve the biggest bang for the buck in providing on-time delivery and superior quality at the best price. During the configuration stage, Intel worked closely with us when we needed assistance on the Itanium IA64 architecture. Intel has allowed us to work with each level within each department to be more efficient in meeting our customer's needs, especially on Intel Server Management (ISM) issues.

HPCwire: Are the clusters performing at the desired level right now?

DC: Both clusters have met customer's requirements and Atipa has fulfilled the contract both in software support and warranty agreement. The support and warranty contract is for three years, and the clusters are performing to customer's satisfaction.

HPCwire: How does the system tolerate failures?

DC: The two clusters need to operate in a manner that it's easy to reconfigure and easy to remove failed components and, most importantly, continue to operate with minimal downtime, since the two clusters are mission-critical with NFS I/O manual fail over requirements. Atipa has designed the I/O nodes with four external Atipa Fiber to SATA Qualifiers, all connected to two Brocade SAN switches. This design will allow NFS I/O nodes to access all disks if necessary during failures of disks.

HPCwire: What do you think is the most important factor for Atipa's continuing success?

DC: Customer support. Customer support is extremely important in the open source community... if customers have trouble running their jobs, they will work with cluster vendors like us who will try to troubleshoot, diagnose and help the customers to get their codes running properly. Of course, when they run their codes, they may know how to troubleshoot, but we need to have the knowledge and to eliminate any hardware or cluster-related problem.

This is an area where our new engineers are extremely strong. We have a staff this year that knows the hardware and the software. We can diagnose with one person, without having to go through 3 or 4 people.

HPCwire: I'm sure that makes things a lot easier for customers....

DC: We will continue to offer non-proprietary hardware for future upgrades. All of Atipa's clusters are expandable. There's nothing proprietary.

HPCwire: You plan on exhibiting at Supercomputing 2004, right? What kind of plans do you have for that, and in the future?

DC: We will exhibit Atipa blade servers and our demonstration clusters, which is what we have in the shop for customers to run their job and compare the performance with different CPU platforms such as Itanium, Xeon64 Nocona or Opteron. We also have a new storage solution with a 5U chassis with 8TB for less than $20,000. In addition, we have a CFD (computational fluid dynamic) demonstration for the for our mechanical engineering clients. As an enterprise financial market, we will have an Oracle rack database server on display as well.

HPCwire: Ok, could you please describe some of the challenges you faced this past year?

DC: Last February, we had two employees from Atipa leave to start to their own company without any notice. But before they left, they destroyed the Atipa Linux Install Server including the backup. The server contained specific developments such as install scripts to automate Atipa cluster configurations, downloaded drivers and patches for products sold since 2000, and customer cluster configuration backups over three years. Fortunately we been in business for many years and have always been prepared for a disaster, therefore we have been able to bounce back quickly and continue to be a strong contender in HPC.

HPCwire: What were their positions within the company?

DC: One was the director of Atipa and one was the CTO for Atipa.

HPCwire: So how did this affect your business?

DC: We ended up doing quite a bit of rebuilding! The servers contained different distributions of Red Hat, SUSE - the open source version - and any other developments we had, including some cluster management software. Also destroyed were all the drivers and patches from the last three years...so it was a lot of work for our new engineers to rebuild. Orders from January, February, and March were delayed.

HPCwire: So you hired a new staff to fix these problems?

DC: Yes, we did. We have recruited and relocated bright and knowledgeable engineers with Master Degrees from several states, Virginia, Okalahoma, Texas. They have worked days and nights to rebuilt Atipa.

HPCwire: The employees that destroyed your servers also sent out messages to customers saying that Atipa was going out of business. How did you set the record straight with your customers?

DC: We did several things. Back in March, we sent out emails to all our Atipa customers. Though a lot of the customer information was gone, we sent out as much we could to the names we had left. Then we sent a letter from our attorney regarding the employee dispute and assured our customers that Atipa was in business and would fulfill all our customer contracts. We put in extra effort in rebuilding the relationship with customers. Atipa will stand behind its warranty for any customer's purchase whether they'll continue to purchase from us or not. HPCwire: Is there anything else you'd like to talk about?

DC: We would like to ensure readers that Atipa, a division of Microtech Computers, Inc. who has been in business since 1986, is very financially well set, even with this employee dispute, the company is financially strong. We would not have been able to recruit our new engineers in such a short time or be able to deliver projects in our most difficult time, if otherwise. Combining over 20 years of hardware expertise and software knowledge, Atipa will continue to be a major supporter in the Linux Open Source community. Atipa Linux Distribution (a x86 platform with 64-bit support) and Cluster Management Suite are ready to launch. We welcome any input from the community.

All of Microtech and Atipa manufacturers understand what we have gone through and have not cut down our credit line, but have expanded it as we've grown. This year, even though we had to do a lot of explanation about the rumors going around, our sales volume have not stopped. For the beginning of fourth quarter, 2004, we have already accumulated $1.7 million of orders to be shipped out. Our total gross sales have exceeded 2003 already.

HPCwire: Thanks for clearing up any misconceptions and telling us about your recent wins!

Catch Atipa at Supercomputing 2004 this year in Pittsburgh, PA. Atipa will be located at booth #2135.