Here’s a roundup of some of this week’s headlines from the HPC (High Performance Computing) industry:
SGI launches InfiniteStorage 5500. SGI introduced InfiniteStorage (IS) 5500, a next generation storage platform. Built with an innovative modular design for extreme density and scalability, the IS5500 offers exceptional performance for high-bandwidth and high-IOPS applications with enterprise-class reliability. It is targeted at big-bandwidth applications such as full-motion video, data analytics and VMware and HyperV environments. The IS5500 can intermix multiple drive types and enclosure densities in a single, scalable system, and features fully redundant I/O paths with automated failover, dual-active, hot-swappable controllers and components, mirrored data cache with battery backup that is de-staged to flash, and hardware-based RAID 6 for maximum performance. SGI also announced that NASA’s Pleiades supercomputer system, built with Altix ICE technology, has achieved more than one petaflop per second in sustained compute performance and moved into the number 7 spot on the June 2011 Top500 list of supercomputer sites in the world. In this video SGI CTO Dr. Eng Lim Goh explains how they are collaborating with Intel to achieve an exaflop by 2018.
Intel equipped to lead path to Exascale computing. At the International Supercomputing Conference (ISC), Kirk Skaugen, Intel (INTC) vice president and general manager of the Data Center Group, outlined the company’s vision to achieve ExaFLOP/s performance by the end of this decade. An ExaFLOP/s is quintillion computer operations per second, hundreds times more than today’s fastest supercomputers. “While Intel Xeon processors are the clear architecture of choice for the current TOP500 list of supercomputers, Intel is further expanding its focus on high-performance computing by enabling the industry for the next frontier with our MIC (Many Integrated Core) architecture for petascale and future exascale workloads,” said Skaugen. “Intel is uniquely equipped with unparalleled manufacturing technologies, new architecture innovations and a familiar software programming environment that will bring us closer to this exciting exascale goal.” Addressing the significant increase in power consumption needs, Skaugen explained that Intel and European researchers have established three labs with three goals: to create a sustained partner presence in Europe; take advantage of the growing relevance of European high-performance computing (HPC) research; and exponentially grow capabilities in computational science, engineering and strategic computing. Intel also products from SGI, Dell, HP, IBM, Colfax and Super Micro, all of which are working with Intel to plan products based on “Knights Corner,” the code name for the first Intel MIC product.
Hardcore and Super Micro demonstrate Liquid Blade Server. Hardcore Computer joined Super Micro Computer at the 2011 International Supercomputing Conference (ISC) in Hamburg, Germany Monday where they showcased its revolutionary Liquid Blade server. Liquid Blade employs Hardcore Computer’s patented total liquid submersion cooling technology.“Hardcore Computer’s Liquid Blade total submersion technology is the future of high-performance computing and general data center cooling.” It is an elegant, extremely energy efficient solution that allows for peak system performance,” said CTO and Co-Founder of Hardcore Computer Chad Attlesey. “Our patented liquid submersion cooling also allows our Detonator workstations to deliver dual 4.8GHz Intel Westmere CPUs on a custom platform engineered for speed and reliability.”
NetApp E5400 benchmarked and enhanced. NetApp (NTAP) announced that the E5400 has been benchmarked for use with the Lustre file system, an open-sourced high-performance file system used by the majority of organizations on the Supercomputer 500 list. In a combined testbed NetApp and Whamcloud used a 4U storage system to achieve 5GB/s for read performance and 3.5GB/s for write performance with the E5400. This combination enables customers with a 40U rack to achieve up to 50GB/s in read performance per rack. “Teaming with NetApp to provide expanded support options for Lustre environments gives a distinct advantage to our HPC customers,” said Brent Gorda, CEO, Whamcloud. “We’re extremely excited with the benchmarking of the E5400 with Lustre in the Lawrence Livermore Hyperion test bed. Lustre is gaining in importance within the HPC community, and working closely with NetApp allows our two companies to deliver outstanding support services, ensuring that customers get the most out of their Lustre environments.” New features available on the E5400 include support for three drive enclosures and multiple drive technologies, as well as scalability to 600 drives per 40U. A New 40Gb InfiniBand interface option is available and can be integrated into InfiniBand infrastructures for flexibility in any high-performance storage network.
A look at the fully-packed racks inside a Facebook data center facility.
How many servers does Facebook have? For some time now, the stock answer has been “more than 10,000 servers,” a number the company began using in April 2008. Facebook has continued to use that number, even as it has soared past 300 million users and dramatically expanded its data center space.
We now have an update: Facebook has 30,000 servers supporting its operations. That number comes from Jeff Rothschild, the vice president of technology at Facebook, who discussed the company’s infrastructure in a presentation last week at UC San Diego (link via High Scalability).
“Today we have somewhere in the neighborhood of 30,000 servers,” Rothschild said during the Q&A session following his talk, adding that the number ”will be different today than it was yesterday” because Facebook is adding capacity on a daily basis.
That places Facebook among the largest Internet companies that have publicly discussed their server counts, but still well behind hosting providers Rackspace, 1&1 Internet and OVH, which each house more than 50,000 servers in their data centers. See Who Has The Most Web Servers for more date on the largest infrastructures.
20,000 Servers In 18 Months
It also suggests that Facebook has added about 20,000 servers since early 2008, which explains why it borrowed $100 million in May 2008 to fund server purchases.
Rothschild also shared some huge numbers associated with Facebook’s photo storage operation, which now stores 80 billion images (20 billion images, each in four sizes). Rothschild said the real challenge isn’t storage, but delivery. ”We serve up 600,000 photos a second,” he said.
25 Terabytes of Log Data – Daily
The amount of log data amassed in Facebook’s operations is staggering. Rothschild said Facebook manages more than 25 terabytes of data per day in logging data, which he said was the equivalent of about 1,000 times the volume of mail delivered daily by the U.S. Postal Service.
Rothschild also discussed the effectiveness of the company’s engineering operations. Facebook currently has about 230 engineers on staff, who manage data for more than 300 million users. Rothschild said that having one engineer for more than 1 million active users has been a consistent historic ratio at Facebook. ”We believe engineers at Facebook have a dramatic impact.”
The 1-hour, 10-minute presentation discusses the company’s commitment to open source technologies and the importance of memcached and Hadoop in their operations.
Here’s a look at some of our past coverage of the growth of Facebook’s infrastructure:
Facebook Makes Big Investment in Data Centers: The company has just locked down a large chunk of data center space in northern Virginia that will provide room for thousands of additional servers for growth in 2011 and beyond.
Facebook: Managing Epic Growth in Real-Time: CTO Jonathan Heiliger presents an overview of Facebook’s back-end operations at the Velocity 2009 conference in June.
Facebook Spending $20 Million a Year on Data Centers: An analysis of Facebook’s data center leases as of May 2009.
A Look Inside Facebook’s Data Center: A recruitment video provides a glimpse inside the server-packed racks and aisles of a Facebook data center.
Facebook Expanding Its Data Centers, Again: In early 2009 the company leased additional space in its East Coast hub in Ashburn, Virginia.
Facebook Pushes Limits on Memcached: Caching is key to massive web scalabilty. Here’s how Facebook is extending a popular caching technology.
