IBM i Shines In Academic Research Data Center
February 15, 2016 Dan Burger
The Power Systems Academic Initiative is more than classes that teach RPG, PHP, and enterprise computing. It’s much more. It’s about research and the role of information technology. And more and more it’s about data and analyses. Where this is headed has caused the Academic Initiative data center to relocate to accommodate growth and the requests from colleges and universities that keeps IBM i in the midst of innovation.
During the past month and a half, the AI data center has found a new home in Tucson, Arizona, in a facility IBM uses primarily for its storage product testing and located near the campus of the University of Arizona. At its previous location, about 100 miles northwest via I-10 in Phoenix, floor space was limited and three-phase electrical was unavailable.
Because expanded systems for storage was a priority, the IBM Tucson facility was a good fit for the three Power 780s and a Power 770 along with approximately 250 TB of storage, a DS8700 storage subsystem and a DS8870 capable of 250,000 IOPS and up to 3.4 GB/sec of bandwidth. The large storage groups are for, you guessed it, data crunching, mostly engineering analytics that come from university research projects.
“There’s increased demand for analytics capabilities,” says Kevin Langston, the principle enterprise system architect at the facility. “We have a lot of tiered environments where the database might be on IBM i, the portal on a Linux system, and a data warehouse and applications on AIX. We put it together so that all systems talk with one another. Where it runs has more to do with where we have the resources available than it is a conscious attempt to run multiple OSes in a single box or split them onto multiple boxes. We are looking for processors and memory allocations and whether we need native disk or SAN-based disk.”
At the university level, there are some with the financial resources to assemble large analytical systems like this, but there are more universities that are without the deep pockets but with an understanding of the potential that systems built for analytical analyses are paramount to research projects. If they have the expertise to participate, IBM is making systems available that can do the number crunching. Where the numbers get crunched, which systems are involved, are of little consequence to the end users, says the guy who architects these systems.
“The AI program has always been a mixed bag of requests, but the requests for analytics is increasing,” Langston says. “What we see are requests that are rising above the issue of platforms and OSes and putting the focus on the productivity of the application base, not what it’s running on. At that level, it’s the skills involved with using the productivity tools and not so much the skills based on operations maintenance. We become the architects of their front ends. We provide distributed services that allow the applications to run with the best performance.”
There is the undeniable lure of big iron, however. It’s like the call of the wild. For some the biggest thrill of a tech conference is hiking through the expo area with the purpose of standing beside serious enterprise computers and the opportunity to lay eyes on a big time data center comes with butterflies in their bellies. So let’s talk numbers. The AI data center server count, give or take a few, stands at 350. Keep in mind the complete transition to the new facility has not been completed yet. By mid-June there will be three fully loaded Power 780s with Power7+ chips, each machine with 2 TB of RAM and 128 cores. By the end of the year, a Power E880 with Power8 processors will be added to the lineup.
Langston explained the planning and configuring of the data center was based on knowing that memory, processors, and storage are critical to the analytical loads that were forecast. Running an analysis that takes 20 hours to process takes a lot of resources sitting on the floor, especially when there are multiple institutions in the same frame.
“The idea was to get systems that are N-1 in terms of current technology. That’s Power7+, which is one processor increment behind Power8,” he says while quickly adding that the DS8870 SAN is current technology.
“These big systems are a heavy lift. Looking down the road at building out the systems, you don’t want to do it a bunch of times. We put enough resource in the frame on the floor that it can scale out based on the anticipated needs of the schools that are in the queue,” he says.
Of the three operating systems that run on Power–IBM i, AIX and Linux–IBM i is the least understood and barely recognized by the academic community. It remains largely behind the scenes in the new environment, but receives kudos from Langston for its database and its ease of management. “We try to take advantage of the specific attributes of all the operating systems that run on Power, Langston says. “In terms of ease of the management components, the one that keeps topping the list is the IBM i. We try to take advantage of what it can do for us in portal and database as much as possible.
“If you build an enterprise database using DB2 on AIX or Linux, you don’t go out and buy DB2 Enterprise Edition. You have to get compilers, and the portal, and build something that will run on these systems. On i there is none of that complexity.”
This gets back to the discussion about whether the data center caretakers care where jobs run or data is collected. From a management aspect, it’s Langston’s favorite. “When a DB2 Enterprise database is requested, we say to them, ‘Go out and get Data Studio and download it as your client, and we’ll put you on i.’ It is fairly transparent to them. They can do all their stuff in a common interface without having to know anything about the operating system.”
“From an operational management aspect, we love it. I’d say that for 80 percent of what we do, my preference would be to put them in an i partition doing database stuff. It is a very efficient way to manage users and the resources they require.
“The productivity of these systems are in the application stacks. Where they run the best may not be on i. It is a choice on our part where to run things, and typically we choose i for the management aspect. I don’t have 20 people here doing stuff for me. There are two architects that run the data center. And the sun never sets on these systems. We have a help desk, but we don’t have a lot of problems.”