A First Look at IBM's New Linux Servers

Today, IBM announces the latest of its Power Systems line of high-end servers. These are the Power Systems S812LC, the Power Systems S822LC (for commercial computing) and the high-performance Power Systems S822LC. All of them are custom-built for Linux.

The S812LC is designed to handle workloads that have high memory and storage requirements; examples include Hadoop and Spark tasks. It features ten cores and 1TB of memory, with a bandwidth of 115GB/sec. It has space for up to 14 disk drives.

The next step up is the S822LC, designed for commercial computing roles, such as on-line transaction processing. This server has 20 cores, 1TB of RAM and a memory bandwidth of 230GB/sec.

The S822LC builds on these already impressive specs by adding two NVIDIA Tesla K80 GPU accelerators for massively parallel processing.

All of these servers are based on IBM's OpenPOWER Architecture.

IBM's POWER Architecture is a direct descendant of its PowerPC range. The technology is based on more than 20 years of continuous research and development. The goal has been improved performance across a wide range of applications, from handheld devices to supercomputers.

In 2013, the OpenPOWER Foundation was formed to open the specification to the community. This project specifically targets the high end of computing, providing solutions that can be freely implemented in data centers all over the world.

Every aspect of the technology has been opened up, from the design of the chipsets through the firmware and essential utilities. The result of opening the specification is that a great many developers were able to become intimately familiar with the capabilities of the system.

At the same time, other top technology providers (such as Mellanox, NVIDIA and many others) have been able to contribute their insights into the platform. The overall effect is a stronger platform and a more productive community.

IBM is not the only technology provider to build servers based on the POWER Architecture specification, but it does have the longest history in developing these systems. IBM continues to develop new hardware and systems software innovations, such as SMP Fabric, which dynamically controls the resources consumed by a service when traffic peaks.

IBM's Power Systems have carved out an enviable niche within the commercial UNIX industry due to the improved performance and greater utilization that its architecture enables. This makes them the current top choice for the UNIX enterprise computing field.

With this latest range of servers, IBM is looking to make a major dent in the real-time analytics and cloud computing markets, fields that currently are dominated by Linux-based stacks.

Linux has proven itself to be extremely well-suited for these fields, where high-performance virtualization and secure networking are essential basic requirements. At the same time, the emerging processing requirements of today's high-end real-time computing tasks are straining the limits of Intel-based commodity servers.

As data centers move from batch processing to complex real-time analytics, new computing models are required. Open-source projects, such as Hadoop and Apache Spark, make real-time analytics faster and easier to develop. But although a large degree of optimization is possible within the software layer, performance ultimately is dependent on the underlying hardware.

Cognitive systems are beginning to play a vital role in big data applications and analytics. Older systems would be able to detect patterns in data through numerical relationships or by recognizing patterns of strings. These systems are unable to detect meaningful data, which is encapsulated within natural language--in text, video or audio. But these data points are valuable and can drive vital business decisions.

Cognitive systems use approaches such as natural language processing and machine learning to discover these vital data points and reveal them to the decision makers. For instance, a system could be set up to detect the early signs of disease epidemics by monitoring social-media broadcasts. Although Google famously demonstrated a statistics-based algorithm for tracking flu outbreaks, its solution was based on a purely statistical analysis of language. It didn't "understand" what the text meant, and so it sometimes generated false positives.

Cognitive systems can uncover these insights from structured data or unstructured data, which greatly reduces the manual "data munging" work that often consumes the majority of an analyst's time.

Cognitive systems offer a deeper level of analysis and more meaningful insights, but it comes at a great cost. Such intensive data processing requires massive computing power and doing it in real time requires an infrastructure that is optimized for extreme speed.

Allegiant Air is one of the first IBM customers to use Linux on the Power Systems architecture. The company uses it to perform real-time analysis of customer behavior and uses this to make real-time offers to individual Web site visitors, increasing conversions and sales.

Of course, these computing tasks go far beyond the requirements of the average blog or lightweight Web application. So the Power Systems LC line of servers has a specific high-end market in mind. That's not to say that small sites and apps won't benefit from these improvements though. Cloud computing providers who switch to IBM's new servers will be able to provide faster processing at a lower cost. And, large Web applications will gain more processing power, enabling them to expose richer functionality through public APIs.

What's more, this new influx of raw computing power does open some exciting doors for the Linux community as a whole. Hardware and software are symbiotic--improving one makes it possible to improve the other. So we can expect to see even more innovation in the high-end and real-time computing field.

______________________