Tuesday, September 29th, 2015

Time	Event
12:00p	Mega Data Center Design: What DuPont Fabros Learned from Building Its Biggest Facility Yet About one year ago, DuPont Fabros Technology, whose business is building massive data centers and leasing wholesale space to companies on a long-term basis, brought online its biggest facility yet: ACC7 in Ashburn, Virginia. It was the first facility to use the company’s newest data center design, whose core concepts it said it would replicate in future projects. ACC7, with its total power capacity of 41.6 megawatts, was an exercise in mega data center design driven to a great extent by technology and business trends, Scott Davis, executive VP of operations at DuPont, said. Davis talked about designing the facility and what his team learned as a result in a presentation at last week’s Data Center World conference in National Harbor, Maryland. The four most important new design elements at ACC7 were medium-voltage power distribution, medium-voltage isolated-parallel UPS topology, a high-efficiency cooling plant, and a new approach to laying out the computing space. Medium-Voltage Power Distribution DuPont decided to use medium-voltage power distribution because of recent advances in medium-voltage circuit breakers, switchgear, and power distribution units, Davis said. The approach gives the data center design more reliability and flexibility but also costs a lot less than a more traditional low-voltage design, primarily because it requires a lot less copper, he said. It reduces the total length of feeder wire required by a factor of seven. To get a duct bank to a 2500 kVA rating using 480V power, the design would need eight sets of four-wire ducts with about 2,700 feet of copper wire. The 4160V system at ACC7 achieves the same rating with only 180 feet of shielded cable in a single three-wire duct. The design also requires a lot less concrete and PVC piping. Here’s the cost comparison in detail from Davis’s presentation: The medium voltage system is more reliable because there are fewer terminations and because the duct banks run cooler, according to Davis. The layout is more flexible because you can have longer feeders without the huge cost. According to Davis, the new medium voltage PDU could be a disruptive technology. The system separates transformers from distribution panels, replacing 10 conventional 300kVA PDUs with a pair of oil-filled PDU transformers. The system is 99.6 percent efficient and takes up a lot less space, which for a data center provider means more room for customer cabinets. The choice to go with medium voltage is not without its drawbacks, however. One challenge is there are fewer quality products on the market. Medium voltage is also a much bigger electrocution hazard, and construction workers are less familiar with methods and best practices for working with medium voltage. Other drawbacks are labor intensity of cable handling and terminations, tighter area for termination work, and a more complex relay protection scheme. Audits, Smaller Deals Call for More Physical Segregation DuPont has changed the way it lays out computer rooms at ACC7. The facility has 28 rooms, yielding nearly 250,000 square feet of computing space, with 1.48MW per room. Each room has the dual-transformer PDU for 2N electrical redundancy and 18 to 22 CRAH units. There are no raised floors. Another first at ACC7 is steel mesh fencing to separate IT racks within a single room. More customers demand such separation, which the mesh enables without taking up a lot of floor space and without modifications to electrical and mechanical infrastructure. Mesh walls can separate DuPont’s cooling and power equipment in a room from the customer’s IT racks or separate sections of rows of equipment or individual rows from each other. There are two trends at play. One is more stringent audit requirements. Companies providing services to especially privacy-sensitive customers, such as healthcare organizations, increasingly need to have physical segregation of some of their clients’ equipment, Davis said. The other trend is customers who would traditionally use retail colocation space taking space at DuPont facilities. It used to be that the smallest customer at the provider’s facilities would have 500kW. Today DuPont is making more deals with customers who want from 100kW to 200kW and who want their equipment physically isolated from neighbors. Warmer Temperature Means More Efficient Cooling Because servers can now run warmer, DuPont is able to use more free cooling. The company went with a water-side economization plant, assisted by a chiller in Ashburn. The heat exchangers provide 65F to 70F water, running year-round and using the mechanical chiller as needed. Higher operating temperatures allow the operator to use much less fan power on CRAH units – 60 percent less than traditional units – and to use much smaller chillers with less pumping power. The caveat is more involved management of hot and cold air and mandatory containment. More details about the ACC7 data center design here. (Comment on this)
3:00p	Preparing for a Successful Data Migration Wai T. Lam is Co-founder and CTO of Cirrus Data Solutions. When it comes to data migration, most of the advice you read from the “experts” sounds like it came from a fortune cookie you might receive at a restaurant run by Confucius himself. It’s easy to give obvious advice when talking about data migration, such as “find the right people who can be trusted,” or “avoid downtime and minimize risks,” or “find the right tool that can do everything centrally and automatically.” However, concrete and specific suggestions are probably more helpful. Confucius was a sagacious man, but there are no records of him ever performing a data migration. Therefore, we’re on our own to discover the wisdom of the ages. When analyzing enterprise-level migration projects, the success stories share some common characteristics. Most of them can be summed up in the answers to the following questions: What Type of Data is Being Migrated? It is very important to determine the type of data being migrated since there are disparate migration types. They are: NAS-based – network shares are migrated from one system to another. LUN-based – disks (LUNs) are migrated from one storage to another at block level. VM-based – the entire virtual hosts are migrated from one hypervisor to another. Application-specific – one set of database or objects, is migrated from one host to another. Also, what type of overall migration is this? Local migration – within the same data center. Remote migration – from one data center to another. Cloud migration – to cloud provider (e.g., AWS, SoftLayer, Azure, Google Cloud). What is the Current Storage Environment? It is equally important to have a reasonable understanding of the current storage environment. Using SAN as an example, what hosts – in terms of OS and applications – are involved? What are the FC switches and RAIDs? Most of all, how large and inter-dependent is the labyrinth of the SAN? Operational complexity could be exponential and not linear relative to the size of the system. An intimate understanding of the entire picture avoids multiplication of complexity and obviates disasters. How Much Data is Being Migrated? It cannot be stressed enough that an accurate and clear estimation of the amount of data must be made. Furthermore, a thorough accounting of the number of hosts, LUNs, and storage systems must be performed. Often, the required effort cannot be judged merely by the number of terabytes that need migrating. Even knowing the total number of LUNs is only half of the story since the number of hosts may be the biggest factor in determining the time and effort required for cutover after the data is migrated. Of course, the amount of data and data rate will determine the absolute time needed for migration, but there are other factors involved as well. How Much Time Overall Will be Allocated to the Project? This is almost obvious, but it may not be that simple a question. The amount of data is usually predetermined, so the only other parameter is the data migration rate. This again highlights the importance of understanding the data, application, and environment because knowing if the data can be migrated in a divide-and-conquer manner helps to formulate the strategy – not only to meet the timeframe, but also to control operational complexity, as well as potential cost. How Active is the Data and How Much Downtime is Available? How active the data is determines if the migration can be completed in time, or at all. This immediately raises the question: How much impact to production will be tolerated during the migration? Furthermore, during cutover, how much production downtime will be allowed? Knowing this ahead of time allows agreements and strategies to be worked out well before migration starts, and also helps to ameliorate stress during cutover. What is the Right Tool for the Migration? The selection of a data migration tool can make or break the project, but what are the evaluation criteria specifically? Again, using the SAN migration as example, a migration tool should be able to: Minimize the need to look up every WWPN, host, storage target, and GUID of every LUN, thereby saving hundreds of hours of painstaking and error-prone work. Install seamlessly with no disruption to the production environment, which significantly reduces risk during preparation and execution of the migration. Provide a clear and intuitive display for the automatically discovered SAN configuration, complete with identities, status, and I/O statistics for all hosts and LUNs to be migrated (as well as those not to be touched), so the migration operation can be organized correctly and accurately with high confidence. Intelligently and flexibly allow minimizing, or even eliminating impact to production during migration by, for example, migrating at non-production hours, or automatically yielding to production traffic, with adjustable intensity/throttling. Monitor, report, and alert on the migration operation in a centralized manner, while also including the production traffic and activities in general with valuable status and statistics. Perform data integrity checking of the intermediate images before cutover occurs, allowing any issues to be discovered before production is switched over to the new storage. Allow scrubbing of the data disks after the data is successfully migrated out, especially when storage is to be returned after lease. The scrub level should be variable to allow an efficient mode for speed, or a secure mode with multi-pass scrub to ensure data is securely erased. Provide a clear, precise and complete post-migration report detailing exactly what happened during the migration process every step of the way, down to individual LUN activities. Demonstrate a solid track record of success and satisfied customers for the given type of migration (through case studies and reference accounts, for example). Who are the Right People for the Job? Now we circle right back to the Confucius wisdom: Above all, find the right people – whether using internal personnel or external professional services partners – who can be trusted and who have experience with the given type of migration. How do you know if they are truly qualified? Well, it is actually very simple: Do they articulate the questions posed above with concrete answers, or do they just give you a bunch of fortune cookies? 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)
5:19p	MapR Adds Native JSON Support to Enterprise Hadoop Distribution MapR, one of the more prominent providers of commercial distributions of Apache Hadoop, has added native support for JSON documents to MapR-DB, its NoSQL database. The addition gives MapR’s enterprise Hadoop platform the ability to run analytics on data stored in one of the most used data-transfer formats today. MapR’s distribution provides both Hadoop, for batch analytics, and Spark for real-time, or stream, analytics. JSON, or JavaScript Object Notation, is primarily a standard browsers and servers use to talk to each other. It is an alternative data-transfer format to XML. Like XML, it is text-only, but it’s simpler and easier to use. It generally takes fewer lines of code to describe an object in JSON than it does in XML. JSON is also a more elegant alternative to storing data in a relational database. For example, it can aggregate data from a row that spans 20 relational database tables into a single document, or object, according to a white paper by Couchbase. One big practical advantage of adding native JSON support to MapR’s enterprise Hadoop platform is the ability to run analytics on data wherever it resides instead of extracting it to one place, where an analytics application can crunch through it, Jack Norris, chief marketing officer at MapR, said. “You can do that now in an integrated environment,” he said. MapR can run analytics on JSON documents without big data transfers and as the documents themselves change. “The types of applications that are possible are basically moving from a reporting and a ‘what happened’ perspective to integrating analytics and impacting the business as it’s happening,” Norris said. Those applications can be things like making a split-second user identity decision after a credit card is swiped, making personalized shopping recommendations in real time, or customizing ads for millions of cable subscribers based on demographics and viewing habits. Another advantage is scalability. According to Norris, a scalable system using JSON has traditionally been an “oxymoron,” but because MapR has put a lot of time and resources into making sure its platform scales well in one or across multiple data centers, the document standard’s scalability problems have been addressed, he said. MapR promises scalability across thousands of nodes per cluster, with clusters deployed in data centers around the world. The company is targeting a general-availability release of the new features in the next quarter. A developer preview is available for download now. (Comment on this)
5:36p	Intel Unveils Its Fastest Data Center SSDs Yet Giving a big boost to its Solid State Drive data center family Intel launched the P3608 series SSDs, engineered specifically for enterprise and high-performance computing workloads. The new line transcends the previous Intel generation across many metrics. The P3608 puts two SSDs on a single card and moves data in parallel to achieve higher transfer speeds, albeit at the cost of greater power consumption. The metric that differentiates the P3608 is it registers transfer speeds over 5 GB/s and up to 850,000 random read IOPS. This is achieved with the help of a NVMe (Non Volatile Memory Express) dual-controller architecture and a move from four lanes to eight lanes of PCIe 3.0. With a dual-controller architecture paired with multi-core Xeon processors, Intel said, the IO can be evenly distributed. Intel notes that this is also more efficient than SATA or SAS, providing more IOPS at a lower CPU utilization. The software RAID-0 load balancing is assisted by a new version of Intel’s Rapid Storage Toolkit for Enterprise (RSTe). The new P3608 family has MLC NAND flash memory and includes capacities of 1.6TB, 3.2TB and 4TB in a low-profile add-in card. The P3608 has a 21.9 Petabytes written endurance rating and 1 million hours mean time before failure for the new SSDs, according to Intel. Intel’s Frank Ober talks about querying very large databases and fast analytics with the new P3608 series, and how it is “fun to run very large TPC-H like queries against this type of drive. To put it in perspective, he said that with “4 x8 cards and 16TB, you’d be able to support over 40TB of compressed SQL Server data using the Fast Track Data Warehouse architectures, because of the excellent compression available with this architecture.” Pricing reflects the power delivered for these new cards, yet it is competitive for the HPC, enterprise and big data markets it is targeting. Intel P3608 cards run approximately $2.19 per GB. (Comment on this)
6:51p	Coping With Data Deluge: Data Center Power, Cooling, Management Trends Your data center is supporting more users, more business use cases, and a lot more data. Digital content and the Internet of Things are generating ever more data and will continue to do so. All of this places resource challenges on the data center facility. The latest Cisco Cloud Index report shows that globally, the data created by IoE devices (Cisco calls it the Internet of Everything) will be 277 times higher than the amount of data being transmitted to data centers from end-user devices and 47 times higher than total data center traffic by 2018. So, how are data center operators responding? What are they doing to create better environmental controls? What’s being done to better support this growing dependency on the data center and the resources it provides? In the latest State of the Data Center Survey by Data Center Knowledge sister organization AFCOM we asked respondents about their data centers and how they are managing their most critical components. Because of the big dependency on data center services, redundancy and uptime are big concerns. We saw fairly steady trends around redundant power levels today and over the next three years. For example, at least 55 percent already have, and will continue to have, N+1 redundancy. Similarly, no more than 5 percent of respondents have today or plan to have in the future 2(N+1) redundant power systems. For the most part, data center managers are using at least one level of redundancy for power. Like power, cooling must be a big consideration in the cloud and digital age. Data centers are increasing density, and cooling is critical to keep operations running efficiently. When we look at cooling, more than 58 percent indicated that they currently run and will continue to run at least N+1 redundant cooling systems. Both today and three years from now, 18 percent said they will operate an N+2 cooling redundancy architecture. Seventy percent of respondents indicated that power density (per rack) has increased over the past three years. Twenty-six percent indicated that this increase was significant. This has forced managers to look at new and creative ways to power their data centers. There is also an increased interest in renewable energy for data centers. For example, 34 percent have either deployed or are planning to deploy within the next 18 months a renewable energy source for their data center. Those who already use alternative energy sources as part of their their data center energy mix split in the following way: Solar: 70 percent Natural Gas: 50 percent Wind: 50 percent Hydro: 27 percent Geo-Thermal: 10 percent So, not only are we seeing growth in density, we’re also seeing more creativity in how the modern data center is supported. Organizations are thinking outside the box when it comes to proper data center and infrastructure management. With that in mind, they are also investing in better DCIM software tools. When it comes to implementation plans and current deployments, the survey found that DCIM is a critical factor when it relates to data center management. The top five currently implemented DCIM software functions include: Environmental (temperature and humidity): 63 percent Facility management: 50 percent Power/energy management: 50 percent Security: 49 percent Cable management: 47 percent With so much increased focus around the data center, it’s clear that redundant systems and good management strategies will continue to be big planning and consideration points. Be sure to download the entire report to learn about the big industry trends, where data center management focus is shifting, and how new requirements will impacts your business. (Comment on this)
7:00p	Private Links to Office 365 Now Live at Equinix Data Centers Equinix and Microsoft have partnered to offer performance- and privacy-sensitive enterprises the option to consume Microsoft’s cloud productivity software suite Office 365 over private network links out of Equinix data centers around the world. Microsoft is the first in what will likely be a string of major Software-as-a-Service companies whose services will be accessible via private links by Equinix colocation customers. Equinix’s then VP of cloud innovation Chris Sharp told Data Center Knowledge in July the company was in talks with all of the big SaaS companies, including the top cloud CRM, or Customer Relationship Management, players. Enabling enterprises to connect their servers in Equinix data centers directly to the public cloud infrastructure services, such as Microsoft Azure, Google Cloud Platform, Amazon Web Services, and IBM SoftLayer, without relying on the public internet, has been one of the fastest-growing business segments for the data center provider. Doing the same for SaaS is the next strategic step for Equinix as it grows its role in the enterprise cloud market. Equinix first announced it would provide ExpressRoute Office 365 services in May. It is now available at 16 of its data centers in the US, Latin America, Europe, and Asia Pacific. ExpressRoute for Office 365 will not bypass the public internet entirely, but it will enable most 365 network traffic to avoid it, Bill Long, VP of interconnection services at Equinix, wrote in a blog post. ExpressRoute for Office 365 has been widely requested by Equinix customers, Sean Iraca, senior director of cloud innovation at Equinix, said in an interview. The size of overall opportunity for Equinix in SaaS is “hard to quantify, but it’s massive,” he said. “SaaS is in demand across the board.” A lot of the demand for private connectivity to SaaS is driven by hybrid deployments, where a customer wants to keep part of their infrastructure under their own control, colocated at an Equinix data center, while taking advantage of the SaaS application front end for global user access. SharePoint, Exchange, or Active Directory are the primary examples of Microsoft apps that can be set up this way, Iraca said. Equinix also announced another partnership with Microsoft around providing ExpressRoute for Azure for government agencies. The launch of Azure Government over ExpressRoute is in its early phases, Long said, with initial availability expected through Equinix data centers in Ashburn, Virginia, and Chicago. ExpressRoute for Microsoft Office 365 is now available at Equinix data centers in: Atlanta Chicago Dallas Los Angeles New York Seattle Silicon Valley Washington, D.C. Sao Paulo Amsterdam London Hong Kong Singapore Tokyo Melbourne Sydney (Comment on this)
9:13p	Microsoft Launches Three Cloud Data Centers in India Microsoft has brought online three cloud data centers in India to improve performance of Azure services for customers in the country. The Redmond, Washington, company becomes first of the top US public cloud providers to offer services in India from local data centers. Amazon Web Services does not have an availability zone in India – although it is currently building a cloud data center there, which it plans to bring online next year – and neither does Google Cloud Platform. IBM SoftLayer also has an India cloud data center in the works. Global reach is one of the attributes that make a public cloud provider better or worse than competitors. India is one of the fastest growing markets for cloud services, and having local infrastructure to serve local customers can be a major advantage in taking advantage of its growth. Gartner expects India’s public cloud services market to reach $838 million by the end of this year – up nearly 33 percent from 2014. The figure includes revenue from Infrastructure-as-a-Service, Business Process-as-a-Service, and Software-as-a-Service. Microsoft first announced plans to establish cloud data centers in India about one year ago. The three new sites are in Pune, Chennai, and Mumbai, serving Central India, South India, and West India Azure cloud regions respectively, Jason Zander, corporate VP for Microsoft Azure, wrote in a blog post. Azure services are now available out of those regions, but Microsoft expects to add support for Office 365 out of the new data centers in October and for Dynamics CRM in the first half of 2016. Microsoft launched a cloud data center in Australia last year. (Comment on this)

<< Previous Day

2015/09/29 [Calendar]

Next Day >>