Arvind Krishna has a big job at IBM. Krisha is the General Manager for Development and Manufacturing in the Systems and Technology Group at IBM. His responsibilities include semi-conductor research and development as well as production.
In an exclusive video interview with ServerWatch, Krishna detailed how IBM is differentiating its x86 offerings and what the future might hold for Power.
When it comes to x86, while IBM is buying the same silicon from Intel as other server vendors, Krishna stressed that there are still additional value adds provided by IBM.
How systems are put together, inclusive of the x86 silicon, is a key differentiator for IBM. For example, how a server system connects to memory, as well as aggregate memory bandwidth, matters a whole lot. System resiliency and fault tolerance are also key points for x86 server differentiation.
Power Operating Systems
While x86 is a volume hardware platform that IBM uses to support Linux and Windows, the Power architecture can support multiple operating systems as well. Currently Power supports the AIX Unix platform, Linux and the i5 OS, which is what was once known as AS-400.
When it comes to deciding when to move from one operating system to another, one of the biggest factors is the current installed base of applications.
"If you already have a set of applications running on what was AS-400 / the i5 OS and there is no particular economic benefit for you to go spend the energy to go port them to something else, you'll keep running that OS," Krishna said.
When it comes to determining whether to use x86 or Power, Krishna stressed that IBM will give clients what they want. He added that in most cases customers already have an idea of what they want and need in terms of server architecture. IBM then competes against other server vendors and not within its own portfolio.
The current generation Power 7+ servers were expanded in February of this year.
Work is now ongoing on future innovations that will improve Power over time. Krishna noted that IBM engineers are working in multiple areas and the expectation is that there will be more compute available per unit area.
"It's typically anywhere from 30 to 50 percent from generation to generation, and generations are typically two to three years apart," Krishna said.
From a workload perspective, social, mobile and Big Data will also have an impact on future architectures. The threading model as well as the ability to more effectively handle interpreted languages are other key areas.
"I'd assert that what you can expect to see in the next generation are systems that are the best in the end to make it simple for Big Data and cloud workloads," Krishna said. "We'll be able to quantify that as the systems come out."
When it comes to quantifying server speed claims, there is no love lost between IBM and Oracle. During Oracle's recent SPARC T5 rollout claims were made by both vendors about who was faster than the other.
Overall, Krishna isn't worried about Oracle's SPARC at all.
"I'd compare a SPARC to a Yugo and I'd compare Power to, pick your mainstream car: Toyota, GM, Mercedes," Krishna said.
Krishna argued that the boxes that Oracle talks the most about are the exa-class engineered systems. Those systems are not powered by SPARC, rather they are x86 powered.
"Why the hell is exadata on x86?" Krishna said. "That means to me they themselves don't believe in SPARC; otherwise they would build their premium box on SPARC."
Watch the full video interview with Arvind Krishna below:
Sean Michael Kerner is a senior editor at ServerWatch and InternetNews.com, the news service of the IT Business Edge Network, the network for technology professionals. Follow him on Twitter @TechJournalist.