Thursday, 29 December 2016

Open Systems for Business – Unlimited Performance

Once upon a time, computer systems was often a combination of specific Hardware and Software packaged and delivered from one vendor. The word Open was not associated with IT and all manufacturers tried to do it all themselves. Once upon a time IBM was one of those companies.

Today, the world has changed and so has IBM. The application landscape evolve in a pace not seen before and the majority of new code is based on Open Source. Almost every modern application also claim to run on any hardware from any manufacturer and infrastructure choice is secondary, so they say.

But is it really true? Infrastructure has no impact on application performance, availability, the user experience and the business value? If you need unlimited performance and always on access, infrastructure really do matter. It always has and it always will, even more so going forward.

Open Systems for Business, Unlimited Performance, IBM

Any server design start with the brain, the CPU. Everything else is designed based on the characteristics of that CPU architecture. The server needs memory and some connectivity to talk to the external world, like your data storage and applications residing on other servers. All servers have parts from multiple vendors and you have a choice of who to buy from. Typically these servers are x86 based and many consider them as industry standard or even Open since there are many suppliers out there.

But does the expression “industry standard” make x86 open? Today’s x86 servers all have one thing in common. The CPU and associated chipsets are developed and manufactured by Intel. The companies that are selling x86 servers do very little innovation themselves and they completely rely on Intel to drive the CPU design, memory pipelines and available I/O bandwidth. Ok, someone will say AMD also make x86 CPU’s and yes they do but almost exclusively for PC’s today, not for servers.

Intel is one company and based on that fact, I would argue that x86 today is not an open server design. X86 do not take advantage of open collaboration or joint development. Smart people talking to other smart people do increase innovation, speed up time to market and lower associated development costs. Spreading work to those best suited have higher possibility of adding value the more people you have in “your team”, in-house or extended. This is one of the reasons why Open Source is so successful, the organized work by a multitude of people, all adding their knowledge and working together to solve a specific problem. It’s fast and efficient and create innovation. Hardware is no different, just much more expensive to manufacture and harder to change if you get it wrong.

One of the emerging IT challenges is the explosion of unstructured data and the compute power needed to extract insight and drive new business value out of that data. The x86 strategy to increase compute capacity seems to be adding more and more commodity servers. The result is what people like to call server sprawl, more network, more software, more complexity, I/O bottlenecks and soaring costs.

OpenPOWER was created in 2014 to develop new infrastructure solutions that deliver unlimited performance at massive scale. Only by including all the experts responsible for all the individual components and technologies can you build a complete system solution, from compute, to memory, I/O and storage and ensure you innovate in the area that have the biggest system impact. It doesn’t matter if you have the fastest CPU if data moves slow to memory and even slower between servers. Your workload will not perform and insight won’t be instant.

Let’s give a few examples on how OpenPOWER drive faster innovation:
  1. The use of FPGA accelerators. A growing number of workloads can take advantage of accelerators. The bottleneck for acceleration is the PCI-E interconnect between CPU and the accelerator card. OpenCAPI is a new interconnect that increase bandwidth, minimize latency and remove that PCI-E bottleneck.
  2. The use of GPUs. NVlink is increasing the bandwidth between GPUs and the CPU providing superior throughput and system performance.
  3. Communicating with other systems. OpenCAPI enables the use of external Flash storage as main memory. You can run a 40TB in-memory database on a single server with that Flash storage attached over OpenCAPI.
  4. Open roadmaps. Providing insight and possibility to team up in areas of interest. Again speeding up innovation and enabling a whole ecosystem to take advantage of new technologies. An example is PCI Gen 4 that upcoming POWER9 will utilize and where members already are developing cards, firmware and solutions.
OpenPOWER is delivering truly Open servers and the benefits are unlimited performance achieved by Systems designed for massive throughput with no bottlenecks. OpenPOWER servers outperform any x86 server and deliver more work for less money.