Netflix Deploys Arista’s 115Tbps Switch
Arista, provider of Netflix’s switching and routing platform, has announced the introduction of their 7500R Series switch. The 7500R Series is specialized for cloud service providers and offers a fabric that can provide speeds up to 115 Tbps with 100GbE density and supports Go programming language and OpenConfig APIs. It also offers 288 Gbps of deep and smart packet memory, supports MPLS, Segment Routing and EVPN protocols, has programmable traffic engineering and is highly flexible with numerous configurations.
The 7500R also introduces Arista’s two-tier active Leaf-Spine option for cloud network design and path customization with the Network-wide Database NetDB and Arista’s Extensible Operating System (EOS). With Netflix’s subscriber count reaching 81.5 worldwide in Q1 2016 and more than 40 million subscribers in the U.S. with no signs of slowing, Arista will need to continue innovating to support their astronomical popularity.
Chinese-Based LeEco Invest $10M To Build CDN in India
Chinese-based LeEco has decided to tap into India’s enormous and rapidly growing Internet user base. With a more than $10 million investment in building CDN infrastructure in 10 Indian cities including New Delhi, Bangalore and Mumbai, LeEco hopes to reach nearly 10 million users in India. LeEco has 650 CDNs locations globally and launched a smartphone in India in early 2016. Providing CDN services for linear cable, live events and video on demand, LeEco will reach users both on their mobile devices and through their TVs.
LeEco has its own original content and will partner with content providers Eros and Yupp TV. LeEco is also integrating its own content transcoding product, compression and clip slicing tools for live streaming along with video resource and content management systems and streaming control. Integration of these features via its CDN cloud will drastically enhance the end-to-end content consumption experience of users in China and in India, which will be its second largest market.
Riverbed Launches SD-WAN Service and Prepares to Go Public Again
Approximately one year ago and nine years after its original IPO launch, activist investor Elliot pressured Riverbed to return to the private sector. Riverbed cited the difficulties of making changes to the company under the microscope of investors and Wall Street and Elliot’s frustration with stagnant performance by the $1 billion company as the causes of this move. Sale of the company to Teacher’s Private Capital and Thoma Bravo raised $3.6 billion. Since returning to the private sector, Riverbed has reduced its workforce from 2,600 to 1,900 and reported in December that they have increased their profit margins to 35%, up from 20% the previous year.
Riverbed does not intend to remain a private company, however, and is now preparing for its second IPO offering. In addition to restructuring the company, Riverbed acquired the German SD-WAN company, Ocedo in February and has now released Riverbed SteelConnect, Riverbed’s “application-defined SD-WAN solution for a cloud-centric world.”
With the SD-WAN sector expected to increase from 1% of the enterprise market in 2016 to 30% by the end of 2019 ($225 million to $6 billion), adoption of this technology was a necessity for Riverbed’s relevance in the market. SteelConnect offers a management portal, point-to-point and full-mesh configuration with automated VPN management, a next-generation firewall and remote LAN switches and Wi-Fi access points. SteelConnect will be available to the general public in Fall 2016.
Facebook Redesigns their Front-End Server
Intel has provided Facebook with its servers for many years. As the servers increased in compute/storage capacity, the demand for power also increased per server. With a set rack architecture power limit of 11kW, the productivity of these Intel server designs were beginning to plateau. Instead of allowing Intel to dictate the design of the rack architecture, Facebook teamed up with them to redesign the front-end servers for data-center workloads.
The QPI links were removed from he servers and all servers became one-socket-based. They also used a system-on-a-chip (SOC) to simplify design, and Facebook redesigned its infrastructure to allow 120 CPUs per rack with 65W processors (increased from 60 CPUs with a 120W processor). Though this switch slightly decreases performance of each processor, they were able to double the total number of CPUs and drastically increase the overall performance. Intel has named this processor configuration Xeon-D or Broadwell-D.
Facebook’s new servers were also designed with 32GB RAM and 128GB storage (named Mono Lake), and Yosemite incorporates the Mono Lake boards and the NIC, each server having its own IP address. This allows CPU communication via a simple NIC switch as opposed to a TOR switch. The Xeon-D processor, however, is limited by having only two memory channels.
Though this is sufficient for Facbeook’s current processing needs, it will need to be increased in later generations to accommodate increased demands on bandwidth and memory. In the meantime, Facebook is working on LLC cache partitioning to allocate space for hot instructions and frequently accessed data to delay the need for these additions and to improve performance.
Google Cloud-Optimized Rendering for Autodesk 3D Animation
Autodesk, specializing 3D animation software, has partnered with Google’s Cloud Platform ZYNC Reader (specializing in cloud rendering) to provide a 3D rendering option that is optimized for use in the cloud. Rendering within the cloud can improve both the speed and the security of the process.
Companies often have to purchase supplemental hardware and software or outsource the rendering process. Investment in datacenter infrastructure and on-premise software is eliminated with ZYNC offered on the Autodesk Maya 2016 software. Increased cloud capacity can always be incorporated if needed for minimal costs and rendering through Arnold, V-Ray and Renderman will all be supported on ZYNC.
Google Cloud Vision API Beta Is Out
Google’s Cloud Vision API, which had a limited Alpha release in December 2015, is now available in Beta to all users. The Cloud Vision API detects objects, texts and logos in images. Label detection is now available on images for $2 per 1,000 images and Optical Character Recognition is priced at $0.60 per 1,000 images with a quota of 20 million images/month. The Cloud Vision API is especially useful for inappropriate content detection (using Google SafeSearch).
It can also analyze the emotions of people in photos, perform text detection and language identification and detection of object through Google photos technologies. The location of an image can also now be obtained from images stored in Google Cloud Storage. Yik Yak, a location-based social networking application, and PhotoFy, a social photo editing and branding app, were prominent users of Alpha version of the Cloud Vision API. The Cloud Vision API is not designed for real-time mission-critical applications and can be used with Python, Java and sampled for Android and iOS applications.
FireEye Expands Security-As-A-Service Offering
Since 2014, FireEye has been promoting their FireEye-as-a-service and has now expanded the coverage of their Advanced Threat Response Centers (now 7 centers on 4 continents available 24x7x365) which respond to an average of 250 security compromises per month and spend an average of 743 hours proactively hunting for threats in customer’s networks.
FireEye also now offers varied coverage options, tiers of service, a simplified pricing model and simplified reporting methods.
Additionally, FireEye has launched its Service Delivery Partner program in which the Partner provides front-end threat detection and FireEye performs back-end monitoring, investigations and proactive hunting for threats. The Herjavec Group is the first FireEye Certified TAM, a new accreditation to certify the validity of the partnership between the companies.