ARM核心服务器与新型闪存，Facebook都想要

在日前于美国举行的年度开放运算高峰会(Open Compute Summit)上，由Facebook高层主导的开放运算计划(Open Compute Project，OCP)为 ARM核心服务器处理器打开了进驻资料中心的一扇门。 该计划针对采用各种 ARM核心或x86架构服务器处理器的插板(plug-in board)发表一套规格标准，而包括Applied Micro Circuits与Calxeda等厂商，展示了符合新规格并采用自家ARM核心处理器的电路板级设计。 但Facebook也清楚表示，目前该类芯片性能还达不到要求水准。此外他们也指出，闪存供货商需要重新拟定其产品计划，以因应今日资料中心对储存的爆炸性需求。 Facebook是第一家向ARM服务器张开双臂的 基于以上消息，Facebook现在成为第一家向ARM核心处理器服务器张开双臂的大型资料中心业者；该公司的一位高层在去年底就曾透露，Facebook可能会为 32位ARM处理器找到一些少量应用，但在64位处理器问世前恐怕还看不到该架构处理器在主机服务器领域的大量运用，可能得等到2014年之后。 在近日于美国加州举行的开放运算高峰会上(Open Compute Summit)，AMD也宣布了该公司有一位金融服务客户正在测试其去年五月提出给OCP的电路板设计；板卡制造商Mellanox则展示了一款资料中心应用的整合式连网产品。 透 过OCP，Facebook鼓励大大小小的资料中心以及相关供货商，在服务器与其它资料中心设备上采用一套共通标准以降低成本；而Facebook的竞争 对手如Amazon、Google与Microsoft还是采用特制服务器主机板以及其它资料中心相关设备，并没有公开他们的设计细节。 Applied Micro宣布，该公司开发了一款采用自家X-Gene系列64位ARM核心服务器处理器、符合新OCP规格标准的主机板设计；该Common Slot规格可支持各种不同架构的处理器。此外Calxeda也展示了采用 32位ARM核心处理器的主机板设计，可作为类似Facebook的Open Vault磁盘阵列系统的储存控制器。

Facebook的Open Vault磁盘阵列系统
D17esmc

Applied Micro表示，该公司可在2013年第一季结束之前，为关键客户提供X-Gene系列处理器样品，而采用64位ARM处理器的服务器主机版，可望在今 年底进驻资料中心。Calxeda除了展示采用32位ARM处理器之Common Slot 规格主机板，也与Avnet Embedded 合作展示其资料中心设计，预计今年秋天可上市。 Calxeda表示，该公司的64位ARM核心处理器计划在2014年推出。至于AMD与Intel也将推出支持Common Slot规格的 x86 服务器芯片。 本文授权编译自EE Times，版权所有，谢绝转载 第2页：希望服务器和处理器厂商多考虑使用者的意见 第3页：存储方案也需要创新

相关阅读：
• 把握云服务衍生的市场新机遇
• 对话ARM：IC世界的竞争是场拉力赛
• 移动终端引领存储器需求，NOR和DRAM各有困扰D17esmc

{pagination} 在开放运算高峰会的专题演说与采访中，Facebook高层表示，他们希望服务器处理器除了多核心之外别加入太多其它东西。 担任开放运算基金会主席的Facebook硬件设计暨供应链部门副总裁Frank Frankovsky表示，Facebook要的是“可分解(disaggregation)”的服务器，这类服务器能每年升级CPU，且无需同时抽换可能只要每五年左右再更新一次的内存、网络与I/O芯片。 目前至少有五、六家厂商计划推出64位ARM核心服务器处理器，目标是为大型资料中心节省能源，但所有这些方案或是相关产品发展蓝图，都是聚焦于打造以太网络、专属架构或其它功能的高度整合组件，而那些功能是Frankovsky不希望加进处理器的。 “我不敢说已经有人表示可以确切按照我们的要求方式打造处理器，”Frankovsky在高峰会专题演说后接受EETimes美国版编辑采访时指出：“不过当他们听到这样的“可分解”消息，他们可以开始改变产品策略方向。” 他补充指出：“今天是这样的概念第一次被公开讨论，所以处理器设计业者可能还需要一些时间来改变想法，考量他们要整合多少功能或是纳入不同选项组合。” Frankovsky 在演讲台上展示了一款符合开放运算新规格的 Common Slot 主机板，采用x8 PCI Express 连接器来将任何一种处理器主机板连结至其它服务器零组件，包括x86架构英特尔(Intel)处理器与Applied Micro的ARM核心处理器。他表示，Facebook可能会在今年底以前首度布建使用该种 Common Slot 规格主机板的服务器。 根据Facebook表示，设计一款全新的服务器大概需要一年时间，透过采用开放运算规格零组件，该公司最近只花了六个月时间完成新设计；不过这还是跟不上软件更新的速度。 “这是软件与硬件发展速度之间的“阻抗不匹配(impedance mismatch)”问题；”Frankovsky在专题演说中表示：“当所有的东西都被固定在PCB上，或是被金属薄片包裹起来，我们就无法取得较佳的弹性。” 本文授权编译自EE Times，版权所有，谢绝转载 第3页：存储方案也需要创新

相关阅读：
• 把握云服务衍生的市场新机遇
• 对话ARM：IC世界的竞争是场拉力赛
• 移动终端引领存储器需求，NOR和DRAM各有困扰D17esmc

{pagination} 存储方案也需要创新 Facebook 基础架构部门副总裁Jay Parikh则是在另一场专题演说中幽默地表示，如果以汽车经销商来形容储存方案供货商，他们应该要考虑在跑车(闪存)以及货车(硬盘)之间提供折 衷的车款选项；而他想要的是一台混合动力车，也就是具备大储存容量的闪存，可允许读写次数耐受度较低，但价格也要低一点。 “当我们得把不那么理想的东西硬塞进来，对创新实在是一大阻碍；”Parikh表示：“底层的储存系统是今日的一大问题。”他在演说后接受采访时指出，低成本NAND闪存可能开创许多资料中心新应用，他正在定义特殊相关需求并将透过开放运算基金会公开。 Parikh 表示，打造该类大容量、低成本的产品可能会打乱闪存供货商目前的产品计划，但有助于支持资料中心领域快速扩展的新应用。在高峰会上，Fusion IO宣布该公司将推出每GB约3.89美元的ioScale系列快闪记忆卡；至于硬盘机厂商则可提供每GB低于1角美元的的储存方案。 根据Parikh分享的数据，在过去两年使用者制造了约2.8 Zettabytes的新数据量，其中包括数以百万计的、贴在Facebook页面的图片；为了因应巨量资料，他的团队甚至认真考虑采用蓝光光盘做为新的储存媒介。 Parikh还介绍了一种Facebook最近设计的、名为“Cold Storage”的客制化服务器，可让储存在单一机架的2 petabyte信息被快速存取；他表示：“为了因应庞大的资料潮，我们必须投入所拥有的一切资源来加速储存设备层面的创新。” 本文授权编译自EE Times，版权所有，谢绝转载 编译：Judith Cheng 参考英文原文：ARM backers jump on Facebook’s server bandwagon；Facebook covets core-heavy ARM SoCs，by Rick Merritt

相关阅读：
• 把握云服务衍生的市场新机遇
• 对话ARM：IC世界的竞争是场拉力赛
• 移动终端引领存储器需求，NOR和DRAM各有困扰D17esmc

{pagination} ARM backers jump on Facebook’s server bandwagon Rick Merritt Applied Micro and Calxeda are among ARM server SoC vendors supporting a new CPU-agnostic board spec that is as part of the Facecbook-led Open Compute Project. SANTA CLARA, Calif. – The Facebook-led Open Compute Project (OCP) announced a specification for a plug-in board that can accommodate a variety of ARM- and x86-based server SoCs. Applied Micro Circuits Corp. and Calxeda are among SoC vendors contributing board-level designs that meet the spec and use their ARM SoCs. With the news, Facebook becomes the first major data center to open the door to ARM SoCs in servers. An executive for the social networking giant told EE Times late last year that Facebook might find some low volume roles for 32-bit ARM SoCs, but that it sees no widespread use of the architecture in host server processors until 64-bit parts are available, probably in 2014 or beyond. Separately, Advanced Micro Devices Inc. (AMD) announced it has financial services customers testing board designs it submitted to OCP last May. In addition, Mellanox is showing an integrated networking product for data centers at the Open Compute Summit here. With OCP, Facebook is encouraging large and small data centers and their vendors to set common specs for servers and other data center gear to lower costs. Facebook competitors such as Amazon, Google and Microsoft specify custom server boards and other data center gear but don’t openly share details of those designs. Applied Micro announced it developed a board design that uses its X-Gene 64-bit ARM server SoC and complies with the new OCP spec. The so-called Common Slot specification announced at the summit can accommodate all SoC architecture types. Calxeda will show this week a 32-bit ARM SoC board that could be used as a storage controller for a disk drive array like this controller board in Facebook’s Open Vault. (To view a slideshow detailing Facebook's Open Compute project, click on the image). Applied said it is on track to sample silicon for X-Gene to key customers before the end of the quarter. “As the first to deliver silicon based on the ARM 64-bit architecture, Applied Micro gives consumers an opportunity to evaluate the benefits of this compelling processor architecture,” said Frank Frankovsky, chairman of the Open Compute Foundation and vice president of hardware design and supply chain at Facebook, in a prepared statement. “An alternative processor architecture such as ARM, coupled with open source software, has the potential to radically increase the amount of compute power we can get from the energy we consume and the money we spend,” Frankovsky said. “An ARM 64-bit server motherboard design has the potential to reach the data center by the end of this year,” Paramesh Gopi, president and CEO of Applied Micro, said in the statement. Separately, Calxeda showed a Common Slot board at the summit using its 32-bit ARM-based SoC. It also demoed Project Knockout, an ARM-based board that can be used as a controller for disk arrays in the OCP Open Vault storage spec. In addition, Calxeda partnered with Avnet Embedded to show other data center designs it will release in the fall. “Partners like Calxeda are critical to bringing creative new design options to the Open Compute Project community, and we applaud their technical contributions to the project,” Frankovsky said in a Calxeda statement. Calxeda is currently shipping a 32-bit ARM server SoC. It has announced plans for a 64-bit version that will ship in 2014. AMD and Intel will also support the Common Slot spec with x86 server chips. AMD shows boards, Mellanox net subsystem AMD aims to leverage the OCP process for a wide range of server buyers. It showed at the summit the first boards based on its Open 3.0 platform, a spec contributed to OCP in May. The boards are based on AMD’s 6300 processors and were developed in collaboration with 14 financial services companies including Fidelity. “A lot of customers I talk to like the OCP model,” said Bob Ogrey, an AMD fellow and cloud evangelist. “After the first OCP summit in New York in October 2011, I sat down with some of the financial services people,” said Ogrey. “We met again in January 2012 with a draft spec and had multiple conference calls to review and revise the spec which was released at the OCP summit in San Antonio in May,” he added. AMD will consider a similar board design for its 4300 series. It is also approaching server customers in other markets such as oil and gas with the designs. The approach aims to lower the cost and simplify the management of servers by letting multiple vendors tag on to a single high volume design. The boards currently are made by two Taiwan ODMs, Qunta and Tyan, a division of Mitac. The move effectively cuts branded server makers such as Dell and Hewlett-Packard out of the loop. They could offer to sell such boards, but it’s not clear what value they would add since their boards are also made by third-parties such as Quanta and Mitac. Click on image to enlarge. AMD's Open Compute 3.0 board is geared for financial services with a variant for Facebook. The Open 3.0 motherboard measures 16 x 16.5 inches, uses two 6300 CPUs and comes in versions that fit into 1U, 2U and 3U high racks, depending on how many hard drives, DRAM cards and adapter boards they use. One variant of the board is geared for Facebook’s Open Rack servers. Ogrey declined to comment on the Common Slot spec. AMD announced late last year it will ship ARM-based server SoCs in 2014. Separately, Mellanox showed sat the summit CoolBox, an integrated networking subsystem geared for data centers. It includes a top-of-rack switch and a server network interface and supports 10-40 Gbit Ethernet and 10-56 Gbit Infiniband links. Facebook covets core-heavy ARM SoCs Rick Merritt Vendors need to design simpler server SoCs and lower cost flash if they want to fill the many sockets emerging in data centers for the social networking giant. SANTA CLARA, Calif. – At the same event where Facebook managers opened the door for ARM server SoCs in the data center, they also made it clear that today’s chips fall short of what they seek, In addition, NAND flash vendors need to refocus their plans to fill exploding storage needs in the data center, they said. In keynotes and interviews at the Open Compute Summit, Facebook executives said they want server SoCs that pack lots of cores and not much else. They also called for a spectrum of NAND flash products in between the extremes of today’s consumer chips in USB drives and the premium products in solid-state drives. Facebook wants disaggregated servers, said Frank Frankovsky, the chairman of the Open Compute Foundation and vice president of hardware design and supply chain at the social networking giant. Such servers should accommodate upgraded CPUs when they arrive every year or so without needing to swap out memory, networking and I/O chips that might only change once every five years or so. At least a half dozen companies are planning 64-bit ARM server SoCs geared to save power for large data centers. But all those with existing chips and road maps are focused on highly integrated parts that build Ethernet, proprietary fabrics and other features Frankovsky doesn’t want into the SoCs. “I can’t say anyone has come saying we will build it exactly the way you are asking for it,” said Frankovsky in an interview with EE Times after his keynote. “When they hear this disaggregation message, they can start changing their direction,” he said. “Today is the first day that a lot of this has been discussed publically, so it may take time for SoC providers to think differently about how much they integrate and have different sets of options,” he added. He showed on stage a so-called Common Slot board based on a new Open Compute specification. It uses a x8 PCI Express connector to link any processor board into other server components. Intel x86 and Applied Micro ARM chips populated a board shown on stage Facebook may deploy its first servers using such Common Slot boards before the end of the year, he said. Shaving design time, flash costs It once took Facebook at least a year to design a new server, but using Open Compute components it completed a recent design in six months. That’s still far behind the pace of software which changes by the hour. “There’s an impedance mismatch now between the speed at which software and hardware moves,” said Frankovsky in his keynote. “Everything is bound to a pcb and wrapped in sheet metal, and that doesn’t allow us good flexibility,” he said. Separately Jay Parikh, vice president of infrastructure at Facebook, quipped in a keynote that if storage vendors were car dealers they would be offering a choice between a sports car (NAND flash) and a van (hard drives). He joked that what he wants is a Prius, large volumes of flash chips with relatively low write endurance and lower prices. “It really puts a crimp on innovation when we have to shoehorn into things that are suboptimal,” he said. “The underlying storage systems are a big problem today,” he said. Parikh said low cost NAND could open up many new flash uses in the data center. He is working to define his specific requirements and share them publicly through Open Compute, he said in a brief interview with EE Times. Creating such high volume, lower cost products could disrupt the current business models for flash vendors, he said. At the summit, Fusion IO announced it would supply its ioScale flash cards for as little as $3.89 per Gbyte. By contrast, hard disks provide storage for less than a dime per Gbyte. The disruption would help serve a rapidly expanding market. Users generated an estimated 2.8 Zettabytes of new data in the last two years, including millions of pictures posted to Facebook, he said. Parikh’s team is even contemplating the use of Blu-ray drives as a new storage medium to keep pace with the flood. He also described a custom server called Cold Storage that Facebook recently designed to pack into a single rack two petabytes of information that can be readily accessed. “There is a huge data deluge, and we have to throw everything we have at it to innovate faster at the storage device layer,” he said.