So to start the ball rolling, the first consistent piece of feedback we hear from multiple points in the ecosystem is that one size no longer fits all. In my home country of the UK, the expression “is horses for courses” (I can explain off-line). Fundamentally, server workloads do not all look the same…The mix of CPU performance, memory and IO bandwidth requirements vary by application. There are a broad set of studies on the topic, here is one such example in this paper from Christos Kozyrakis from Stanford University. For cloud based servers, where the box itself is the revenue generating asset of the company, there are strong economic reasons why it makes sense to tailor the platform for a specific task or set of tasks. For many of these workloads, we believe the critical decision criteria is moving away from pure performance and instead viewing performance (of that particular workload) / watt / dollar
ARM has been looking at this area for a while. James Hamilton’s (excellent) blog picked out some of our early experimentation regarding a web server about two years ago. This demonstrator, constructed from off-the-shelf hardware and software modules showed that it was perfectly possible to construct a particular type of server using alternative technology. Of course, much more work would need to be performed, from both hardware and software perspectives to construct a value proposition that was commercially viable. This has been the focus of my team inside ARM and a number of external individuals/companies since then.
A number of other pieces need to be pulled together in order to develop a compelling platform that end users can evaluate and ultimately deploy. Calxeda, an ARM silicon licensee, has announced its trailblazer initiative as it engages with a number of middleware software partners. With a number of applications associated with Web 2.0 applications including a Java component, I am particularly excited at the technology demonstration at Oracle’s JavaOne conference last week of the C2 just-in-time (JIT) compiler running on ARM at I plan to cover this topic in more detail in subsequent blogs. ARM has had a long standing relationship with Oracle and the ARM architecture is supported with their C1 JIT compiler as part of Java Standard Edition for Embedded 7. The C1 JIT minimizes latency but, as a result, does not focus on performance optimization. The C2 JIT improves performance at the expense of latency. Oracle showed a side by side comparison and claimed a 20-40% performance improvement depending on workload. This is important for applications like Hadoop, an application many people are seeing as a good fit for ARM Powered servers. No release schedule was announced for this compiler on an ARM target. For those of you that missed JavaOne, here is a good write-up of the broader set of announcements made during the show.
Software is a massive topic and merits many more paragraphs in future writings. I also plan to outline how ARM's business model, detailed in Ian Thornton’s relatively recent blog, will be utilized to enable a new category of server platforms to be developed. Of course, I welcome your feedback as to areas you would like me to explore in future blogs.
Pulling an ecosystem together does not happen overnight – it will take some time. This initiative will provide the market place with a range of options and allow the market to select the solution that best aligns to their particular requirement. So it will be more than one size and more than one winner. We look forward to the road ahead.
Ian Ferguson, VP Segment Marketing, ARM, has spent years fighting from the corner of the underdog. Most of those scars are healing nicely. Ian is particularly passionate about taking ARM technology into new types of applications that do not exist or are at the very formative stages. After driving ARM’s server program for five years, Ian now leads ARM’s vertical marketing organizations, that supports ARM’s partners to grow their business across a wide range of applications with a view to reinvent the way the server function is implemented in networks as opposed to simply replacing incumbent platforms.
0 Comments On This Entry
Please log in above to add a comment or register for an account
Market Disruption and Innovation From Sensors to Servers
on Yesterday, 09:33 AM
Fortune Brainstorm Green
on May 13 2013 10:58 AM
Moonshot - a shot in the ARM for the 21st century data center
on Apr 09 2013 01:22 PM
Bringing the Benefits of the Smartphone to Pay-TV
on Mar 14 2013 05:34 PM
2013 - A Lucky Year For All Smartphone Consumers
on Mar 13 2013 06:58 PM