Hyperscale infrastructure is seen by many as the answer to Big Data, web-facing applications, mobile communications and a range of other challenges facing the enterprise, but there is a problem: the disconnect between massive scale and the need for customized performance.
Top-tier enterprises like Google and Facebook have led the hyperscale movement by speccing out their own hardware largely stripped of the bells and whistles that accompany most channel solutions. They make up for this through advanced software management that allows them to provision and repurpose data resources as needs arise.
But since the standard enterprise lacks the clout to purchase optimized hardware in bulk, how are they to leverage hyperscale infrastructure without giving up the performance that they have come to rely upon? Enter the programmable processor.
Intel made a splash recently with word that it is working on a new x86 Xeon that contains a field programmable gate array (FPGA) that would allow core logic to be reprogrammed at will to suit specific workloads. In this way, the enterprise would deploy a hyperscale environment, and then figure out what to do with it as data requirements emerge. And best of all, it means that hyperscale does not have to devote itself exclusively to one or two key applications, but can be parceled out to any number of enterprise functions, and then reconfigured as new challenges arise, or even leased out to others cloud-style.
Microsoft, for example, is working with FPGAs to devise a reconfigurable fabric that would greatly enhance performance of key web-facing applications like the Bing search engine. Project Catapult would place the fabric on a board level solution with associated local DRAM that would sit behind a 48-server half rack. The idea is to bring hardware acceleration to the service’s search algorithms, backed by a high-bandwidth interconnect to enable broad resource allocation. The company is currently testing the system in configurations that scale past 1,600 servers.
At the same time, Oracle is getting some special lovin’ from Intel in support of its Exascale database platform. The company has apparently persuaded Intel to provide a specialized Xeon – the E7-8895 v2 – that enables performance characteristics to be altered on the fly according to the workloads it encounters. In this way, the device can be tailored toward the heavy number crunching of traditional database applications, or the parallel processing that normally accompanies Big Data functions. The device is likely to enter the normal sales channel soon, but since it was Oracle’s idea to begin with, they’ll get first crack at it.
Even “off the radar” applications like image resizing are getting the customized hardware treatment, according to Enterprise Tech’s Timothy Prickett Morgan. Dell and Convey Computer have teamed up on an FPGA that accelerates image processing for web-scale applications. With client devices now ranging from desktops and laptops to tablets and smartphones, organizations are under pressure to repurpose image content to a high degree. Any delay in resizing those images results in poor user performance and potentially lost sales. Using Convey’s Wolverine FPGA, Dell says it can significantly accelerate the performance of Memcached content processing to reconfigure images for multiple displays. This also cuts down on the data loads of hyperscale operations because it eliminates the need to store multiple image files of different sizes.
Programmable processors, then, provide the best of both worlds for the enterprise. They can be made available across a variety of commodity platforms, but still provide the customization needed to enable efficient, effective delivery of scale-out applications. The enterprise will probably need to retain some in-house knowledge as to the best way to implement this programmability, but that is a far sight cheaper than building multiple dedicated platforms for key workloads, and then having them sit idle while traffic is low.
Arthur Cole writes about infrastructure for IT Business Edge. Cole has been covering the high-tech media and computing industries for more than 20 years, having served as editor of TV Technology, Video Technology News, Internet News and Multimedia Weekly. His contributions have appeared in Communications Today and Enterprise Networking Planet and as web content for numerous high-tech clients like TwinStrata, Carpathia and NetMagic.