Sometimes when you issue a press release,  you need to blog about it to make sure everyone understands the significance of the moment and the news.   In this case, Calxeda announced  that Fedora had chosen and installed Calxeda-based Boston Viridis clusters (96 nodes) to equip the Fedora community with production ARM servers. This is their shiny new build environment…  So, I was going to blog about it, but then read Charlie Demerjian’s article in SemiAccurate.  ’nuff said.  You nailed it, Charlie. Thanks!

http://semiaccurate.com/2013/05/15/redhat-put-calxeda-based-arm-servers-in-to-production/

Ok, most everyone knows that a “Moonshot” is something grand; something  audacious and bold;  and typically something very, very far away (252,088 miles away to be precise).  But in this case, part of that distance has already been covered by the team that HP and Calxeda assembled over 18 months ago as part of HP’s Pathfinder Innovation Ecosystem initiative.  In case you missed it, HP and its partners announced Moonshot servers on Monday, April 8, in New York City.   While ARM vied with Intel for attention,  many were trying to get a bead on the ARM aspects, since HP had already announced last year that “Gemini” would initially be Intel Atom based..  Here’s a snapshot:

1. As HP has often reiterated,  Moonshot is designed to be processor agnostic. They previewed prototype hardware from Calxeda, Applied Micro, and TI in addition to Intel Atom, which will ship first.
2. Calxeda has fully functional quad-node Moonshot Server cartridges working at 1.4 Ghz in the  labs  (See photos !)
3. Calxeda will be able to support HP’s launch schedule to ensure their customers have access to production-quality Calxeda-based Moonshot servers later this year,   with up to 180 quad-core EnergyCores in the 4U Moonshot server chassis at 1.4 Ghz.  That is 4 times more dense than the Atom server they showed at today’s event.
4. As the Anandtech review concludes, Calxeda can deliver 80% better performance per watt that the best-in-class Xeon processor, even in a relatively CPU-intensive web application. Expect the HP system to enable similar value when it is ready for the test.
5. Unlike Calxeda EnergyCore, Intel’s new Centerton chipset has not been generally available for testing.   We all look forward to seeing how well it will do against ARM parts when it begins shipping.
6. It is interesting to note that Intel felt compelled to announce Centerton’s replacement (Avoton) on the same day that HP announced support for Centerton,

Calxeda powered the very first Moonshot Development Platform,  Redstone, announced in November, 2011, and is proud to be one of the founding members of the Pathfinder Innovation Ecosystem.  In 2012, we began deploying Redstone systems in HP’s Discovery Lab, at ISV partners, and in HP’s customer datacenters.  Our joint testing and validation work in these projects accelerated the ARM server ecosystem and enhanced customer’s understanding and comfort with ARM-based servers.

We proved that you can really build a dense server cluster with ARM processors.  Yes, it is real LINUX, indistinguishable from any other server, except the exceptional performance per watt.  Working together, we have met with scores of datacenter users and independent software vendors, working to understand which applications make sense on the first generation of ARM servers in Moonshot.

We’ve learned a great deal on this journey together.    And now, we hover just above the surface, preparing to touch down with an ARM-based HP Moonshot system in just a few short months; the Calxeda cartridge is running in our labs and will be ready for production later this year.

Ready for touchdown?  So are we!  We’d like to thank HP for being a great partner, and for their courage to try something really new; something that deserves the name “Moonshot”.  And thanks to everyone who has been testing Redstone systems for jumping on board early.   But make no mistake; this trip is not over;  it has just begun.

 I don’t want to spoil the fun of reading the article for you, but it all comes down to what Anandtech calls “The Results That Matter”. We’ve been fairly open that the goal of our product is perf/W or perf/U and in those scenarios I’m quite pleased with what Anandtech found. In their web hosting benchmarks (PHPbb deployment), the team showed that the Calxeda-powered Boston Viridis can provide 44% more performance at 19% lower power than a 2P Xeon E5-2650L system or 16% more performance at 36% lower power for the higher wattage than a 2P Xeon E5-2660 system. This is not a speculative guess at where performance will be, but rather a real number on real hardware that you can go buy today from Boston in the UK, and Penguin, System Fabric Works, Avnet or Scalable Informatics in the US.

I particularly want to call attention to the use of a PHPbb deployment as a real world benchmark. One of the comments on the Anandtech story that caught my eye was that this was a “cherry-picked” benchmark. As presented in the review, a focus area for our first product is LAMP-based web applications and PHPbb provides a good look at that, since it ties together all of the main LAMP components into one widely used application. Yes, it tests our hardware in a use case that we designed for, but it is neither a niche case nor irrelevant to general web application or web hosting scenarios. No benchmark is perfect, but this one can be used as a performance baseline for many LAMP-based applications, like WordPress, which as of this writing is running on over 60 million websites worldwide.

The micro-benchmarks in the Anandtech article do however point out some areas where we have ground to cover. As we look ahead, I see three areas where we will not only improve the benchmarks, but also overall application performance. First of all, the Calxeda engineering team helped Anandtech to do some tuning and drop overall power draw significantly, without affecting performance. Those savings are all based on settings and features that are available to every Calxeda customer, there’s just some learning curve since they don’t exist on x86 systems. We continue to add to that list and have plenty more tricks up our sleeve on current hardware for upcoming firmware updates. One example is that we’re already targeting sub 100W idle power for a later firmware release (compared to 135W by Anandtech).

Second, the maturation of ARM code in the Linux ecosystem has been very rapid.  Not only does a large percentage of Linux code run on ARM-based systems, but the code is now being optimized for ARM. Take a look at the projects going on at Linaro and the LEG group in particular and you can see there’s a lot of goodness coming to the Linux/ARM ecosystem. All of that work is starting to show up in official Ubuntu, Fedora, and now OpenSUSE distributions. On the ECX-1000, Ubuntu 12.10 brought across the board performance improvements compared to 12.04, with some packages, like PHP and Ruby, improving 30% or more. Another example is the STREAM benchmark; we can already improve upon the STREAM result published by Anandtech by another 9% using standard GCC compiler optimizations. There’s a lot more “free” performance to be had out there.

The third area is the ARM roadmap. The Anandtech team has pointed to the rapid evolution of ARM-powered hardware and we’ll be on that curve as well. We have an A15 part coming this year that will increase both memory footprint and raw performance. Beyond that, when ARM v8 products launch the improvement will continue.

This is a first step, of many, for Calxeda and ARM servers in general, so I want to close by reminding you Anandtech’s results are a starting point on our first-gen hardware, and a darn good one at that. As we work with more hardware and software partners, the ARM performance and capability gains that the Anandtech article points out (and better) are entirely within our grasp.

We’ll be at World Hosting Days in Germany next week with our partner Boston Limited, so Anandtech’s results in a web hosting scenario are extra relevant. Drop by booth B10 next week and let’s talk about what we can do for your hosting environment.

Boston releases world’s first commercially available ARM as a ‘Service’

Hannover, Germany (March 5, 2013). Boston Limited, a leading manufacturer of high performance, low-powered server, virtualisation, storage and cloud solutions, is proud to unveil its ARM-as-a-Service (AaaS), powered by Breeze and ARM®, at CeBIT 2013. For more information and a live hardware demonstration please visit Boston at Hall 2, #A39, Hannover Exhibition Grounds.

Powered by one of the industry’s most disruptive technologies, the Boston AaaS is the world’s first commercially available cloud offering, based on the Calxeda® EnergyCore® ARM-based processor technology. The Boston AaaS is a platform which provides developers with all the tools and services required to port and migrate software to the ARM platform. Built specifically to assist in migrating and porting applications from X86 to ARM, Boston have teamed up with software developers Ellexus for their product Breeze, a choice of product also made by ARM to profile and troubleshoot applications on their own HPC cluster.

Breeze is a novel technology for tracing programs as they run in order to monitor file dependencies and environment settings. Its offers a unique visibility into the inner workings of complex scripted flows such as those used in semiconductor design or complex software builds. Breeze helps IT managers, tool vendors and users to communicate and solve problems more quickly.

“We’ve been working with ARM for 18 months and we are now extremely excited to be partnering up with one of the stand-out front-runners in the ARM-based server ecosystem, Boston Ltd. We understand the hardware constraints that many software developers will face in trying to enter the ARM environment and Boston’s cloud offering provides a fantastic opportunity for those wanting to future-proof their applications as more and more users move over to ARM technology.” Says Dr. Rosemary Francis, Managing Director of Ellexus Ltd.

Boston’s ARM-based servers, the industry renowned Viridis Microservers, are built on ultra-low power System-on-Chips from US-based hardware manufacturer, Calxeda. Boston was the first company in the world to launch server grade ARM products when the Viridis was launched in 2012 and this cloud offering continues this momentum, establishing Boston as one of the leaders in ARM server based products and services.

“There is tremendous demand for easy access to ARM server technology, so the time is ripe for AaaS,” says Karl Freund, VP Marketing, Calxeda, Inc. “We are thrilled to see Boston and Ellexus stand up this service to provide cloud-based access to engineers to build and optimize their server codes for ARM.”

Boston’s ‘ARM-as-a-Service’ delivers dedicated physical quad-core nodes as opposed to virtual CPUs like the majority of conventional cloud offerings. Users will be able to develop on single nodes or test scaling capabilities of applications across multiple nodes within the cluster. Users will be able to choose from varying levels of software and professional services to assist in their migration.

“The Boston cloud is the ultimate resource for application and software developers looking to port their software on to ARM. Our platform provides all the tools to facilitate porting software to ARM in one easy to use cloud offering. There are also a range of training videos and professional services available for users looking to fast-track their migration to ARM” says Boston’s Head of HPC, David Power.

Mobile, cloud computing spur tripling of micro server shipments this year

Shawn Kaplan

Advances in multi-core computing have allowed far greater compute densities such that nearly all datacenter racks run out of available power far sooner than physical space.  Traditional High Performance Computing (HPC) X86 clusters can consume upwards of 400W per rack unit (U), this means that a typical data center rack with a 5KW – 8KW circuit can be maxed out in as little as 1/4 or 1/2 of the available space.  Many of today’s forward thinking IT leaders are asking “Why can’t I have both extremely dense computing and better power efficiency?”

I’ve been talking to a number of financial services and digital media customers about the potential that the “ARM Sever Revolution” is showing against X86.  For those of you new to ARM, these are the processors found in the smartphone or tablet you’re carrying around.  As you can imagine, they are powerful yet extremely light on power consumption.  While initially only targeted at mobile computing, chip and board manufacturers are re-applying these technologies to enterprise class server applications.  You can now go to HP and order a server configured with ARM processors.

Calxeda, for example, has some very interesting technology bringing power savings and a density of Linux servers that have the potential to turn this market on its head.  One of their OEMS is talking about 24 Calxeda EnergyCore(TM) server nodes, 96GB of memory, and 6TB storage is measuring 130W “at the wall”.   That is not a typo.  T-w-e-n-t-y F-o-u-r (24) servers in a 2U chassis for 130W. By comparison, this can be as much as a 20-1 reduction in power from traditional x86 architecture. At these numbers, configurations of literally THOUSANDS of cores are now possible into a single cabinet.

What’s held back the adoption of other more exotic solutions like FPGA and GPU has been that they require complete rewrites of applications.  Perhaps the most compelling part of the “ARM Revolution” is that ARM servers run Linux!  This means that ARM servers running Linux offer compute/watt advantages without requiring a complete rewrite. Developers will be able to recompile and retarget applications with a quick recompile.  Better cost, better power, better TCO, with the push of a button.

Keep an eye out for ARM processors and what this revolution can bring to the financial services industry in 2013.

• Networking: Intel’s SOC requires you to add hardware for networking
• Storage: Once again, there is no SATA connectivity included on the Intel SOC, so you must add hardware for that
• Management: Even microservers need remote manageability features, so again with Intel you need to tack that on to the power and price budgets.

Unless you add additional hardware on top of it, Intel’s SOC allows you to boot and not much else. Let’s also consider the fact that you’ve got a total of 8 lanes of PCI Express Gen 2 on each SOC. If you’d like to add the Server-Class items listed above, choose wisely, because those 8 lanes will go fast. Add all of that hardware, plus memory, and 6 W is simply not possible.  And of course these additional components add cost and take space as well.

Let’s expand that thought to an actual Atom S1200 powered system, like the Quanta S900- X31A. Each node includes a Marvell 88SE9130 SATA controller at a TDP of 1W, an Intel i350 1GB controller at 2.8W TDP, an AST2300M estimated at a conservative 1W, and an SODIMM at roughly 1.2W (Using the same number we at Calxeda have used). That adds at least 6 more watts per node, almost doubling the 6.1W TDP of the processor. Multiply that across 48 nodes and you just tacked on 288W to each chassis. In a 42U rack full of them, you just added 4kW to each rack! By no means is that a limitation or shortcoming of the Quanta design, which is actually quite good, but rather an indication of the excess baggage that all vendors will need to deal with in putting together an S1200 powered system.

The currently shipping Calxeda ECX-1000 Server-Class SOC ships with SATA, Ethernet fabric links, IPMI-based management, and 8 lanes of PCI Express Gen 2, standard at 3.8W (5W including 4GB DDR3). It’s also worth pointing out that Calxeda’s integrated fabric switch provides more than just the Ethernet ports missing on the Atom S1200.  Applied at the system and rack level, it can dramatically reduce Top of Rack Switch ports and cabling complexity, while increasing internode bandwidth by 10-fold.  You can have all of that in a 5W server. Not 5W + additional components. Why not take that 12W budget you need for each S1200 node and get two Calxeda nodes with all of the Server-Class features included?

In the end, Intel may simply be claiming 64-bit as the main benchmark for Server-Class. When matching microservers to the appropriate workloads, we’ve found that there is surely a place for 32-bit in the datacenter. We’ll be providing a blog post on that very topic in the near future.

So, while the Centerton announcement indicates that Intel takes “microservers” seriously after all, it falls short of the ARM competition. It DOES have 64-bits and Intel ISA compatibility, however. Most workloads targeting ARM are interpreted code (PHP, LAMP, Java, etc), so this is not as big a deal as some would have you believe!Intel did not specify the additional chips required to deliver a real “Server Class” solution like Calxeda’s, but our analysis indicates this could add  10 additional watts PLUS the cost. That would imply the real comparison is between ECX and S1200 is ~3.8 vs ~16 watts. So roughly 3-4 times more power for Intel’s new S1200, again, comparing 2 cores to 4. Internal Calxeda benchmarks indicate that Calxeda’s four cores and larger cache delivery 50% more performance compared to the 2 hyper-threaded Atom cores. This translates to a Calxeda advantage of 4.5 to 6 times better performance per watt, depending on the nature of the application.

So, what’s a little core like ARM doing in a place like this? Its all about the data. “Data Intensive Computing” in HPC is pronounced “Big Data” in the enterprise.   And the two communities have another thing in common: both are seeking more energy efficient solutions to large computations challenges. So naturally, they are turning to ARM with great hopes for the future.

While the ARM cores do not provide tremendous computational horsepower today, Calxeda’s massive fabric and performance roadmap show a lot of promise as the industry prepares for the age of “Exascale” computing (1,000 petaflops).   Of course most HPC applications will require more floating point performance than the roughly 3 Gigaflops of an A9, or perhaps 6-10 GFlops of an A15, as well as 64-bit addressing.  However there are workloads such as Big Data processing and genomic analysis which can be effectively hosted on ARM today.

Calxeda’s booth at SC’12 was crammed with working systems from 4 partners

Calxeda had a small booth with lots of hardware from Boston Ltd, HP, Penguin Computing, and Scalable Fabric Works, purring quietly and producing almost no heat, of course.  There was a constant throng of traffic from attendees from around the world coming to see for themselves the promise that ARM holds:  a 24 node server, with 24 SSD, consuming only 160 watts under load.  The power meter helped convince the incredulous that this was no joke, but demonstrated a true breakthrough in power efficiency.  (Oh, and of course we had very cool t-shirts to give away!)

In addition to the many Calxeda-based systems and projects, Barcelona Supercomputing center issued a press release announcing that their next generation green supercomputer (Mont Blanc) is being built using mobile parts from Samsung (dual core A15 with Mali GPUs.  Although this part was not designed for servers,  this is an early example of  how the ARM ecosystem can tailor SOCs for specific applications and is a peak into the future of Supercomputing.

So, while this little core stands humbly at the entrance to the world of massive computing challenges, the neighborhood is welcoming ARM with open, well ,… arms.