Wednesday, June 18th, 2014

Time	Event
11:00a	Software Defined Storage Startup Brings OpenStack Swift Object Storage Into the Enterprise SwiftStack launched release 2.0 of its software defined object storage platform that comes with Swift, the object storage portion of OpenStack, an open source cloud architecture. SwiftStack wants to help bring object storage to enterprises that want to have storage capacity on any device, anytime, anywhere available to any application. It aims at enterprises involved in scaling for the new web-based world. With 2.0 SwiftStack is coming out of stealth mode, announcing a number of flagship customers, including eBay and HP Helion. Object-based storage is a storage architecture that manages data as objects as opposed to file systems and block storage, which manages data as blocks within sectors and tracks. There’s been a lot of investment in object storage and several vendors, ranging from smaller startups to the likes of EMC, building solutions for it. SwiftStack founder and CEO Joe Arnold said all enterprise applications will eventually rely on object storage to keep up with growth of data and access points required by users. ”It’s the only way enterprises will be able to compete today and in the future,” he said SwiftStack focuses on OpenStack, providing control, integration and distribution capabilities for the open source cloud. The hardware-agnostic SwiftStack platform simplifies manageability, integration and scalability for customers using commodity hardware or legacy hardware. Launched in 2011, the company has been relatively quiet despite signing up several Fortune 500 customers.  Besides eBay and HP, they include Pac-12 Networks, which produces more than 750 live sporting events annually. Seamless integration with legacy SwiftStack 2.0 enables seamless integration with legacy infrastructure, allowing enterprises to tie in to existing on-premises systems. The startup says its solution lets IT build an object storage system for any application, and lets CIOs deliver customer-centric infrastructure at lower cost and with fewer resources. The new features include a Filesystem Gateway that integrates object storage with existing file-based applications. It supports CIFS/NFS and installs on any Linux distribution. Data comes in as files and objects and comes out the same way. SwiftStack 2.0 also has a software defined storage controller that has improved visibility and control into a storage system, even when it spans multiple data centers. It provides a single pane of glass for operators to manage and scale their storage environment. This creates a cohesive SDS platform to orchestrate upgrades, utilization reporting and gain insights into how the storage systems are performing. There are enhancements to enterprise authentication with plug-and-play integration with enterprise management systems, such as LDAP and Active Directory, providing seamless access to their object storage. Object storage at eBay and HP “Software-defined storage powered by open source technology provides the best value by leveraging commodity infrastructure,” Russ McElroy, manager of storage engineering and strategy at eBay, said. ”SwiftStack and Swift are key to our object storage roadmap. “SwiftStack’s products and expertise have been key to our ability to scale eBay to the multi-petabyte level. Not only can we control how our storage is configured, but our storage platforms can be consolidated, so it’s easier to manage and scale our data.” HP Helion, the newly formed unit for all things cloud at at the company, uses SwiftStack for a sync-and-share solution and to create a static web pages for the company’s various business units. “To do this rapidly and efficiently, we turned to SwiftStack,” said Mike Gasaway, senior solutions architect for HP Helion. (Comment on this)
11:30a	Hosting World’s Big Trends Mulled Over at HostingCon 2014 There are a few constants to HostingCon each year. The first is how the hosting industry will tackle Amazon Web Services’ dominance. The second is the continuing convergence of the industry – the distinct lines of shared, dedicated and managed hosting of yesteryear have all but blurred. There’s always buzz about a major hosting acquisition or two. At this week’s HostingCon in Miami (a 10-year anniversary of the conference), it was GoDaddy’s purchase of innovative hoster Media Temple and Peak 10’s acquisition by GI Partners. The Media Temple deal spoke about the importance of matching cultures and transparency with employees for a successful merger, while Peak 10’s valuation answered some big question many people have had about the industry. There’s a ton of mixed service providers in the mid market that are hard to gauge due to diversity of the business. Peak 10’s acquisition put a multiple on this type of business and potentially will spur more acquisitions in similar vein. An attendee who wished to remain anonymous noted that with GI Partners owning ViaWest and Peak 10 and ViaWest’s gradual expansion eastward, it’s worth asking whether there a ViaEast could be in the cards? Maybe not. But it will be interesting to see how far GI Partners will take the potential for synergy. Peak 10 recently disclosed it was looking at two specific new markets. Where’s the growth? There was general sense that the middle market is a major growth area at the moment, the section of the market populated with companies that have until now sat back and waited to see what worked with cloud and what didn’t. Startups and born-on-cloud companies have been driving cloud revenues, and enterprises have the resources to tackle cloud in meaningful ways, while the middle market was in wait-and-see mode. Opinions diverged on which sectors within this market had the most potential. Phil Shih, of Structure Research, said he was seeing strong managed services growth, while GigaOm’s Ashar Baig said he expects massive consolidation and closures in the managed services world. Cloud revenue continues to grow well ahead of other services. Shih noted that colocation growth has been slowing, from high teens it was once gaining down to low-to-mid teens. He argues that it is the industry’s maturity that is causing the deceleration rather than adverse pressure from outside players. There are also new themes that have gained traction, namely data security and how the hosting business will fare after the NSA surveillance disclosures. “The next cold war is the encryption economy versus the surveillance state,” said Christian Dawson of ServInt and the Internet Infrastructure coalition.  Several tips on how to fight back were provided in a number of sessions. The industry clearly sees the surveillance disclosures as damaging to the U.S. hosting businesses. Amazon: from threat to opportunity In years past, AWS was seen as the biggest threat to the hosting industry. This year Antonio Piraino of ScienceLogic held a session on how to build a business providing managed AWS services. “Talking about working with AWS was almost taboo three-five years ago,” Piraino said. “Now you need an AWS strategy.” Piraino doesn’t suggest going all-AWS for infrastructure, but rather offering a managed option for AWS for customers who are looking for hybrid infrastructures. Companies like Datapipe offer managed AWS services alongside their own infrastructure offerings. Piraino argues that while many enterprises are looking to AWS, the majority of them don’t have the capabilities in house and look to a trusted advisor. This is an opportunity for hosting providers who can step in to help. Piraino has an interest in this, as ScienceLogic is a platform for advanced monitoring and could be a valuable tool in building these services. The tools that AWS provides are fairly cut and dry and don’t provide the majority of what enterprise customers need, according to Piraino. A new(ish) way for colos to look at cloud A company called Atlantic.net has gone from being a dial-up provider to colo provider to cloud provider. A relatively small firm out of Florida, its reps spoke about how it began using cloud for temporary capacity for colocation customers, and how cloud grew into a product of its own. The company began offering its self-developed cloud to customers and saw tremendous uptake by colo customers, domestically and worldwide. What started as a service to add value to colo customers is now a major focus. Adam Weissmuller, director of cloud solutions Internap, had a session on bringing cloud into the colo era. Internap is seeing two types of buyers: traditional enterprise IT customers looking to buy infrastructure for the entire organization and cloud buyers for a single application. Cloud buyers are typically end users, business units and developers. By offering both cloud and colo, it addresses both of these needs. There is a disconnect between the IT side and the business side, but through hybrid infrastructure in a single portal service providers can remote this disconnect. While colocation customers still like to put servers close-by, Internap is finding that these same customers still care about location when it comes to cloud. Cloud also does not have much room for customization. Colocation is cheaper on the whole for multi-year engagements and allows for high customization, but means a slower speed to market. Cloud lets those in an organization, typically referred to as “shadow IT,” get the agility they need within the confines of the IT contract instead of going outside. Other providers, such as Interxion, maintained that they will not get into cloud, instead providing those services to customers via partnerships and building ecosystems. Growth in Europe Interxion said the colo business in Europe is doing extremely well, growing faster than growth in the low teens projected for the U.S. JF van der Zwet, global marketing manager at the company, said that enterprises are the big opportunity going forward. All of the on-premises legacy deployments are undergoing tech refreshes and looking to colocation as the next evolution. There isn’t a clear cloud leader in Europe because each country is its own unique market, but individual local cloud providers have been driving growth for Interxion. He mentioned a joint partnership between Interxion, Dell and OnApp that allowed a customer to simply drop a cloud into an Interxion data center, build on Dell hardware and the OnApp platform. The product looks even more promising following OnApp’s federated cloud announcement, as it now provides a common data footprint. Interxion also wants to help U.S. companies establish presence in Europe by introducing them to the right people and giving insight into the political landscape and tax regulations. Disclosure: HostingCon is a property of iNet Interactive, which also owns Data Center Knowledge. (Comment on this)
12:30p	Building Blocks – Predictable and Simple Storage for Virtual Environments Saradhi Sreegiriraju is the senior director of product management at Tintri, a provider of smart storage for cloud and virtualized environments. Today, virtualization is commonplace in the enterprise data center and its benefits are indisputable. However, several challenges must be addressed to ensure the success of any virtualization initiative. To accommodate rapid growth and fluctuations in demand, the infrastructure needs to easily support the scaling of individual VMs, as well as the addition of new virtualized workloads. Although network scaling is becoming simplified with software-defined network architectures, scalability continues to be one of the top challenges when increasing the level of virtualization in the modern data center. Virtualized environments are complex by nature. However, today’s virtualization solutions have been designed for easy scaling, simply by adding virtualization hosts with a building block approach. By using this approach, traditional storage frequently becomes the bottleneck when scaling virtualized environments. Let’s review some of the challenges enterprises face when using general-purpose storage solutions for virtualization. Scaling adds complexity Traditional disk-bound storage architectures can be scaled for either capacity or performance by adding disks, but this doesn’t make for an efficient environment. IT organizations usually over-provision traditional storage to obtain the performance needed by their virtualized environments resulting in excess, unused capacity and unnecessary operational complexity. However, there are only so many disks traditional storage controllers can support. When the maximum number of disks is reached, enterprises have to either add another storage controller and manage it separately or perform a forklift upgrade to the entire storage system. In either scenario, scaling traditional storage to meet the growing needs of virtualized environments is an ongoing battle. Lack of control and increased overhead Although storage performance and the costs of scaling are top concerns for IT administrators, the increased management burden is the most significant issue for most organizations. From complex operations needed for provisioning, to mapping storage in virtual environments, administrators are forced to plan how they want their storage environments far in advance to encompass the different groups of VMs they will be creating. This practice can overwhelm IT administrators as they manually track the assignments of virtual servers and VMs to storage arrays and volumes. Many organizations are still using extensive spreadsheets for storage LUN or volume-to-VM mapping, which is an incredibly inefficient and error-prone way to manage storage. Additionally, traditional storage systems make it extremely difficult to predict how scaling will affect existing virtualized workloads. With traditional storage solutions, all VMs are thrown into one big pot of storage. There is little to no insight into the performance of individual VMs, and no way to easily apply or measure the quality of service for specific VMs. The problem is further exacerbated in the cases when environments with related VMs are forced to span multiple storage systems, as this leads to more performance issues. The lack of insight into individual VMs across all storage makes troubleshooting time consuming for administrators, who are forced to investigate all possible issues across multiple dimensions and workloads. With traditional storage, finding the source of any performance issue requires multiple layers of management software. The lack of insight and control over storage leads to unpredictable behavior in the virtualized environment. This can have serious consequences for organizations, especially when combining different types of critical workloads with dissimilar performance requirements onto the same storage system. Predictable performance is simply out of reach for organizations using traditional storage platforms, since it is impossible to know if new VMs will fit on the existing storage or cause any performance issues with the applications or services. Building blocks: A better approach for scalable virtualization storage There is an entirely different approach to scaling storage in virtualized environments. This new approach believes that storage should be scaled the same efficient way as the building block approach in virtualized environments is used for the compute layer. By understanding and operating at the VM level the new storage is able to leverage the management constructs from the virtualization layer, making scaling storage extremely simple and predictable. This simple building block approach to scaling storage in virtualized environments means that each building block appears as a single datastore in the virtualization layer. This is in contrast to traditional storage, where administrators must spend time creating separate LUNs or volumes for each new group of VMs, then create and export the corresponding datastores. With the building block approach, there is no longer a need to create and manage complex webs of LUNs and volumes. As an added bonus, building blocks can provide administrators with a clear and comprehensive view of system performance and capacity utilization. IT no longer needs to interpret piles of complex capacity and performance data, or worry about how much spare performance they have to work with. A building block can also ensure there is never any impact from “noisy neighbors” on the same system. With VM-level QoS functionality, storage performance is predictable. Furthermore, different types of VMs such as IO-intensive OLTP workloads and latency- sensitive end-user desktops, can reside on the same building block. If administrators need to scale beyond the performance and capacity of a single building block, it is as simple as adding an additional system to the virtualization layer − a task that takes less than five minutes. This effectively adds another datastore that can be managed by the virtualization layer. The ability to monitor and control the virtual environment is important when scaling storage–building blocks need a unified, intuitive control platform that lets administrators administer multiple building blocks as one. The best choice Virtualization requires storage to scale, and the easiest way to scale is by using a building block approach to adding performance and capacity. Traditional storage systems designed before the advent of virtualization are difficult to administer and scale as an enterprise’s virtualization storage needs change. IT needs storage that understands virtualization by design in order to step up to the demands of dynamically changing virtualized environments. They also need simple, intuitive tools that can provide comprehensive visibility and control into the entire storage environment while empowering them to monitor and operate storage at the VM- level. A building block approach paired with a centralized administration and control can offer organizations the best choice for building and scaling virtualized environments. Industry Perspectives is a content channel at Data Center Knowledge highlighting thought leadership in the data center arena. See our guidelines and submission process for information on participating. View previously published Industry Perspectives in our Knowledge Library. (Comment on this)
1:00p	CenturyLink Launches Utility-Style Managed Services for the Rest of Us CenturyLink has kicked off the roll-out of on-demand managed services customers can order through the same portal they use to deploy cloud infrastructure resources and consume and pay for them by the hour, like they do for cloud compute or storage. The move brings managed services – something that has been reserved for enterprise-scale customers that deploy substantial colocation assets on a long-term basis – accessible to individual developers, startups and small and medium businesses. The company made the announcement in conjunction with this week’s GigaOm Structure conference in San Francisco. Unlike traditional managed services offerings, CenturyLink is not asking for a minimum commitment from a customer who wants to use the cloud-based version, Richard Seroter, director of product management at CenturyLink, said. Traditional managed services usually come with at least a 12-month commitment and cost thousands of dollars. Managed services in the IT world are a form of outsourcing, where a customer pays a provider to manage specific layers of their infrastructure stack. It can be operating system management, database management or middleware management, to name just a few. CenturyLink is starting by offering eight types of managed services in the cloud, but there are dozens more that will eventually follow, Seroter said. CenturyLink’s first eight cloud managed services: Windows operating system Red Hat Enterprise Linux operating system Active Directory Apache HTTP Server Apache Tomcat Microsoft Internet Information Services Microsoft SQL Server MySQL A bread-and-butter business CenturyLink gained its managed services capabilities almost entirely through its acquisition of Savvis in 2011. These offerings have become one of its bread-and-butter businesses. The company lumps cloud and managed hosting revenue into one bucket, which contributed about 40 percent of revenue in the first quarter. CenturyLink does not break out the managed hosting revenue from this bucket, but a spokesman said cloud was responsible for a much smaller portion of the total. On-demand experts, any time One of its biggest competitors in the managed services space is Rackspace, which has turned hands-on management and customer service into a mantra. But, Seroter said, Rackspace’s managed services are still only available with a minimum commitment. The strength of CenturyLink’s cloud-based offerings is that a developer can outsource just the management of an operating system, or just a MySQL database for any period of time, however short. They can request a service as a configuration option through CenturyLink’s cloud interface and have one of the company’s 1,200 or so engineers dedicated to managed services provide it within an hour, Seroter said. Ultimately, the company wants to “give you what you like in the cloud, but have a team of experts behind it, tuning it, managing it, licensing it and so forth,” he said. (Comment on this)
4:30p	Cisco Acquires Swedish SDN Firm Tail-f Systems For $175 Million Cisco has agreed to acquire privately held Tail-f Systems, a provider of multi-vendor network service orchestration solutions for traditional and virtualized networks for about $175 million in cash and incentives. Stockholm, Sweden-based Tail-f Systems was one of the first to take a new approach to the networking Operations Support system, where it decoupled network services from specific network components in an open, standards-based and model-driven service orchestration technology.  Its Network Control System (NCS) orchestration server software enables network operators to provision their services and network devices with a single software tool. The company lists AT&T and Deutsche Telecom as customers with SDN initiatives under way. Tail-f products will join Cisco’s Evolved Services Platform (ESP). The acquisition accelerates Cisco’s cloud virtualization strategy of delivering software that increases value to our customers’ applications and services, while supporting open standards, architectures and multi-vendor environments. With Tail-f’s network service orchestration technology, Cisco’s service provider cloud and virtualization portfolio will simplify and automate the provisioning and management of both physical and virtual networks, the company said. Tail-f’s solutions are equally applicable for today’s network problems, such as layer 2 or layer 3 VPN provisioning and next-generation networking based on network function virtualization (NFV) and network programmability. Tail-f was named a 2014 Gartner Cool Vendor for having SDN control for elastic network services and NFV service orchestration and emerging analytics services. “With a rapidly increasing number of people, devices and sensors connecting across the Internet of Everything (IoE), service providers require new capabilities to deliver value-added, cloud-based services and applications,” said Hilton Romanski, senior vice president, Cisco Corporate Development. “Our goal is to help to eliminate the bottleneck caused by operational complexity within the network. “The acquisition of Tail-f’s network services configuration and orchestration technology will extend Cisco’s innovation in network function virtualization, helping service providers reduce operating costs and the time it takes to deploy new services, making agile service provisioning a reality.” Tail-f employees will join Cisco’s Cloud and Virtualization Group led by Gee Rittenhouse, vice president and general manager. (Comment on this)
6:00p	Amazon’s Vogels: Thank Data Center Efficiency for Cloud Price Cuts Since Amazon launched its cloud infrastructure services, the company has cut the amount of money it charges for them more than 40 times. The most recent one came just yesterday, when Amazon Web Services’ Provisioned IOPS (solid state storage in the cloud) went down in price by 35 percent. Speaking at today’s GigaOm Structure conference, Amazon CTO Werner Vogels said it was the company’s focus on data center efficiency that enabled the company to make such drastic price cuts continuously. “Efficiency of data centers is really important for us because it helps us drive the overall cost of the operation down, which benefits our customers.” Until recently, Amazon has been providing Infrastructure-as-a-Service for years without any substantial challenger. Now, however, giants like Google and Microsoft are going after this market. Because of their scale and substantial in-house data center engineering capabilities they are able to compete with Amazon on price, performance and features. In May, for example, Google reduced the rates of its on-demand virtual server business called Compute Engine by about 32 percent. Amazon followed almost immediately by announcing IaaS price cuts that ranged from 10 percent to 40 percent, depending on the type of service. Speaking at the AWS summit in San Francisco, Andy Jassy, senior vice president of the cloud business, said that was its 42nd consecutive rate reduction. About one week later Microsoft slashed the rates for rentable virtual compute and storage infrastructure on its Azure cloud by up to 35 percent. Commenting on this entirely new competitive landscape, Vogels said it was a good thing for Amazon. More players means more customer using cloud services, which means a “bigger pie” for everybody. Few people outside Amazon’s infrastructure team actually know how the company designs and runs its data centers and the hardware they house. It has been extremely tight-lipped about that aspect of the operation. Understandably so, given the competitive advantage of being able to constantly reduce the price of its services. Google and Microsoft have been more open. Google data center execs regularly present at industry conferences and talk about data center best practices at the company. Microsoft has done the same. The company has been very public about its ITPAC data center modules, for example. Earlier this year Microsoft open sourced the newest server design that will underpin all of its cloud services through the Open Compute Project, a Facebook-led open source hardware and data center design initiative. (Comment on this)
9:30p	Applied Math Upgrades its Cloudy Data Center CFD Modeling Tool Applied Math Modeling has released the latest version of its Software-as-a-Service data center modeling tool CoolSim that uses cloud-based high performance computing resources to process Computational Fluid Dynamics workloads. Data center modeling allows users to test changes in the physical environment using software simulation before they actually implement those changes. CoolSim can be used to perform “what if” scenarios to determine the optimal placement of existing equipment, evaluate new or alternative designs, or visualize the effect of adding new equipment. Data center modeling is complex and CoolSim aims to make it easy and helpful. Once built, the model is automatically submitted to a hosted high-performance computing (HPC) cluster for processing using computational fluid dynamics technology. The output consists of reports and 3D images. Both existing and new users will benefit from the features added in the latest release, the company said. The object component library system was updated to include the latest and greatest from data center equipment manufacturers, and users can now also create their own library items. The entry cost of CoolSim is $8,500 and includes a “virtual server” based on advanced cloud-based HPC clusters. The company says competing solutions begin at $20,000 and are software-only solutions. Because it is a SaaS solution, the company can roll out upgrades easily to everyone at the same time. “CoolSim was designed from the ground up to meet the needs of users who don’t have the time to climb a steep learning curve,” Paul Bemis, CEO of Applied Math, said. “CoolSim 4.3 continues this tradition by adding a significant amount of features aimed at speeding up the model building and simulation process, allowing users to analyze many more design alternatives prior to selecting the optimal design for a given data center environment.” Most of the new features are tweaks to make the CoolSim easier to use. They include: The ability to quickly define rack level details including gaps and servers to the 1U level of detail The ability to group geometry and replicate groups multiple times with offsets in any dimension for rapid creation of data center layouts  A “power view” allowing users to visually represent rack power density in 3D The addition of cylindrical fans to any cooling unit for enhanced modeling fidelity Exposure of fan curve support for modeling high-pressure containment designs An enhanced object component library system including updates from the leading data center manufacturers, including Emerson, Stulz, Tate and Schneieder Electric The ability for users to create their own library items and store them for later use A Pre-Simulation Report that allows users to review model input data in one report, allowing users to make changes to input parameters from within the report “Using CoolSim 4.3, users are now able to build models faster, and with greater fidelity, than ever before while leveraging the cost effectiveness of Cloud Computing to drive down simulation time and cost,” Bemis said. Applied Math’s software isn’t solely for data centers. Other possible application areas include modeling airflow in atriums, hospital operating rooms, clean rooms and smoke modeling. (Comment on this)
11:00p	RagingWire to Build Third Massive Data Center on Sacramento Campus RagingWire, operator of what it says is the largest data center campus in California, is making that data center campus even larger. The company announced it has kicked off construction of a 180,000 square foot building on its property in the state’s capital Sacramento, which will be the third building there. Douglas Adams, RagingWire’s senior vice president and chief revenue officer, said the provider was planning to bring the first 4MW of capacity in CA3 online in November. The expansion comes earlier than management originally planned, but the recent acquisition of an 80-percent stake in RagingWire by the Japanese telecommunications giant NTT Communications was not the reason for the accelerated construction schedule. Space in CA1 and CA2 is simply going fast, and the company wants to stay ahead of demand, Adams said. Started and marketed originally as a disaster recovery destination for Silicon Valley companies, the Sacramento campus has over time grown into a popular primary data center location. While it takes a couple of hours to drive to Sacramento from the Bay Area, it is a much more seismically stable zone than San Francisco and Silicon Valley, both of which are located near six faults. The provider won’t confirm this, but sources have told Data Center Knowledge that one of its biggest customers in Sacramento is Twitter. Another high-profile tenant is StumbleUpon. RagingWire announced the expansion at this week’s GigaOm Structure conference in San Francisco. NTT makes expansion easier to fund While NTT investment was not what enabled the expansion, it certainly didn’t hurt. “It made the financing of it a lot easier,” Adams said. NTT paid $350 million for its 80 percent ownership stake in the data center firm. The company launched and grew on founders’ own dime and its own revenue until about five years ago, when it began to dip into outside credit sources. The NTT deal “gives us access to very, very inexpensive capital very quickly and very easily,” Adams said. Easy access to capital will come in handy. Announcement of the expansion in California comes as the company continues construction on its campus in Ashburn, Virginia, and evaluates other markets around the U.S. for further expansion. Adams declined to say which metros it was looking at specifically, saying only that the focus was on top-tier markets. Those would be places like New York, Atlanta, Dallas or Chicago. CA3 close to being half sold Market dynamics for RagingWire in Sacramento are drastically different than they were when the company started. Before, the company was forced to take risks and build before it could secure customers, which simply doesn’t happen today. “In the old days, we couldn’t presell for the life of us,” Adams said. Upwards of 50 percent of everything RagingWire has built recently was leased before it was finished. At this point, CA3 is close to being half sold. Sticking to what works RagingWire has invested a lot in engineering its own power backup infrastructure, which it says is a lot more resilient than the norm, and has used that as a way to differentiate. It is generally sticking to the same power and cooling design for CA3. The company has used the same overall design for about 10 years now, and it has been “issue-free,” Adams said. Every new build improves slightly on the previous one, however, and CA3 will be able to use more airside economization for cooling than existing facilities do, he added. (Comment on this)
11:30p	Facebook Breaks the Network Switch Up Into Modules After it customized servers and storage to optimize for its applications and to enable its developers to roll out new software features at lightning speeds, networking switches were the remaining component of Facebook’s infrastructure that was a “black box,” with tightly coupled vendor-designed proprietary software and hardware. This will no longer be the case at Facebook. The company has designed its own top-of-rack switch and a Linux-based operating system for it, both of which Jay Parikh, its vice president of infrastructure engineering, previewed at GigaOm Structure this morning. Companies with data center infrastructure of Facebook’s scale, such as Google, Amazon and Microsoft, design their own hardware because the approach results in better performance of their specific applications than off-the-shelf products offer. It also saves them a lot of money since they can source the hardware from multiple competing manufacturers and don’t have to pay the high margins incumbent hardware vendors tack onto their products. Last year, Facebook’s open source hardware design initiative Open Compute Project started work on a switch that would work with any operating system. Earlier this month, Facebook’s director of network engineering Najam Ahmad told Data Center Knowledge that the company was already testing a handful of OCP switches in production in its data centers. The switch Parikh announced this morning was a separate effort, outside of OCP. Facebook does, however, plan to open source the design through the initiative, he said. Disaggregation of the network The main idea behind the box, nicknamed “Wedge,” is disaggregation of individual components. Facebook has pursued disaggregation across its entire infrastructure stack. It enables the company to upgrade individual server components, such as CPUs or network interface cards, instead of ripping and replacing the entire box. It also allows for easy configuration of machines to optimize for different purposes. Wedge gives the company’s networking hardware the same chameleon properties. Both the hardware and the software are split into modules that can be mixed and matched, and Facebook can innovate and upgrade each of these components one at a time. The central modular hardware feature of Wedge is the OCP “Group Hug” motherboard, originally developed for microservers. A single Group Hug board can accommodate Intel, AMD or ARM processors. Besides configurability, using a server board makes the box behave less like a switch and more like a server. “It basically takes the switch and it turns the switch into … another server,” Parikh said. The advantage of that is the ability for Facebook to provision and manage them using the same Linux-based operating environment it uses to provision and manage its server fleet. “This enables us to deploy, monitor and control these systems alongside our servers and storage, which in turn allows our engineers to focus more on bringing new capabilities to our network and less on managing the existing systems,” Facebook’s Yuval Bachar and Adam Simpkins wrote in a blog post. OS turns software engineers into network engineers The company designed FBOSS, the operating system for Wedge, to leverage the software libraries and systems it uses for managing the server fleet. These include provisioning, decommissioning, upgrades, downgrades, draining and undraining. Because it can program switch hardware Facebook can implement its own forwarding software faster. The company’s Software Defined Network controller makes routing decisions centrally to ensure optimal data delivery around the world. FBOSS also puts an abstraction layer on top of the ASIC APIs of the switch which enables its engineers to treat it like they treat any other service in Facebook. “With FBOSS, all our infrastructure software engineers instantly become network engineers,” Bachar and Simpkins wrote. (Comment on this)

<< Previous Day

2014/06/18 [Calendar]

Next Day >>