Content Delivery Infrastructure Updates #2


Aryaka Grows 150% in Q2 2016

Aryaka, the global SD-WAN solutions provider, has announced its stunning fourth consecutive quarter of growth. In Q2 of FY 2016, Aryaka is reporting that it has grown over 150% year-over-year compared to Q2 of 2015.

As Aryaka states in its press release, the company has enjoyed strong partner contributions, growth in customer bookings (with over 70 added), and continued expansion attributed to its install base. This growth was due in large part to broader growth in demand for connectivity, which Aryaka provides in a cost efficient, flexible, and low-latency manner.

Aryaka also touted the fact that 40% of its total bookings came from its pre-existing user base, who sought further services from the company. Many of its bookings growth came from China, the APAC region, and Europe.

“I’m excited about another blockbuster quarter whereby we delivered solid execution and tremendous growth across all geographies,” said Shawn Farshchi, President and CEO of Aryaka, in the press release. “Aryaka is providing our customers with global connectivity, application acceleration, and ability to scale their businesses. Our results validate our leadership position in the global SD-WAN space.”

Facebook Is Introducing Open-Source AI Hardware Design

The Facebook AI Research team (FAIR) is touting its advances in the development of AI training hardware, which it attributes to a combination of strategic investments and partnerships with vendors, as well as hardware expertise. FAIR is now announcing that it will contribute its research in GPU hardware to the Open Compute Project.

The press release notes that while machine learning and artificial intelligence have been subjects of research for some time now, larger strides have been made in recent years due to the advances in computing capabilities as well as the public availability of larger and more numerous data sets. In particular, as computational capabilities increase, researchers are empowered to delve more deeply into complex systems.

However, larger scale developments in such software would require the development of a commensurate system: “But we realized that truly tackling these problems at scale would require us to design our own systems. Today, we’re unveiling our next-generation GPU-based systems for training neural networks, which we’ve code-named “Big Sur.”

Big Sur is an Open Rack-compatible piece of hardware, which incorporates eight high performance GPU’s of up to 300 watts to attain a speed that is two times as fast as its predecessor. Distributing the training load over eight GPU’s allows the team to increase the size and speed by a factor of two. In addition, Big Sur is built on NVIDIA’s Tesla Accelerated Computing Platform and can support multiple PCI-e cards. Finally, unlike comparable systems, Big Sur is a highly efficient and can operate in free-air cooled Open Computing data centers. The design materials will be freely available via the Open Compute Project.

Facebook Introduces the 6-Pack Open Modular Switch

For the past few years, Facebook has been working on a new, disaggregated and more flexible network in order to avoid some of the problems encountered in closed and iterative traditional networking technologies. The result of their efforts is an open modular switch platform, dubbed “6-pack.”

Facebook details the process of developing 6-pack in a blog post. The team began by developing a top-of-rack network switch (called “Wedge”) and a Linux-based OS (“FBOSS”) for the switch. In addition, they developed a scalable modular network architecture, a data center fabric that has 6-pack at its core. It is a “full-mesh non-blocking two-stage switch” with 12 unique switching components, each of which is capable of switching 1.28 Tbps.

As this description implies, it comes with two configurations, the first of which exposes 16X40 GE ports to the front and back. The second configuration opens all 1.28 T capacity to the back. Because the switching elements are independent and each contain a full local control plane on a microserver that communicates with the central controller, they can easily be modified without disrupting the system at large. The result is an open and modular switch platform that is highly scalable and built on off-the-shelf building blocks.

Facebook is currently testing 6-pack, FBOSS, and Wedge, and is planning on contributing them to the Open Compute Project in the hopes that they will be used by the community as the foundation of an entire industry.

Instart Logic Introduces Multi-Page Predictive Prefetching

Instart Logic is introducing Multi-Page Predictive Prefetching, as a part of its Appspeed Portfolio. This development promises to improve UX on a broad range of apps. In an era where customers increasingly use mobile devices to reach businesses, mobile UX is an important part of shaping a company’s perception.

The end-point aware application delivery solution addresses an area that is rarely addressed by developers, who expend their focus and resources on optimizing the primary landing pages and main access point. Multi-Page Predictive Prefetching refrains from simply guessing where the user will seek to go next and loading the page in the background, which wastes bandwidth  and resources. Instead the solution identifies static elements shared by across potential pages and models common flows in order to optimally push common resources to the browser, resulting in a more seamless browsing experience.

“Using the tremendous amount of user behavior data we analyze across millions of sessions with our machine learning algorithms in the cloud, we can apply additional performance optimizations that improve the experience throughout a user’s journey,” explains Shailesh Shukla, CMO and VP of Products at Instart Logic. “Multi-page Predictive Prefetching leverages our proprietary endpoint-aware Nanovisor technology to identify the best content to prefetch.”

Cisco’s Visual Networking Index Predicts 3X Increase in IP Traffic By 2020

Cisco has recently released the findings of its Visual Networking Index (VNI) forecast for 2015 to 2020. According to the VNI, global IP traffic will triple and grow at CAGR 22% over the five years leading up to 2020.

The global Internet-using population is also expected to increase by more than one billion in that time, reaching a total of 4.1 billion by 2020. The burgeoning of M2M connections and the growth of the Internet of Things will result in a massive increase in new devices and connections, totaling 26.3 billion by 2020.

As smart cities, virtualized health services, and other M2M connections proliferate (nearly tripling by 2020), global internet traffic will continue to balloon, which will in turn generate growth in the market.

Video traffic will drive the lionshare of global internet traffic in 2020, accounting for a staggering 79%, and reaching 3 trillion video minutes per month. HD and Ultra HD video will account for the highest share of internet video traffic (82%).

On a more somber note, the growth in Internet traffic will also lead to an increase in security threats. Specifically, DDoS breaches are expected to increase to 17 million by 2020. Cisco’s analysis of DDoS attacks found that they can account for as much as 10% of a nation’s internet traffic. Regionally, APAC will be a major driver of growth over the next five years, at a rate of 22% CAGR, followed by North America (19% CAGR).

Is Airtel Checking CloudFlare Traffic in India?

Karthik Balakrishnan suggested recently that Airtel was sniffing and censoring unencrypted data packets that were traveling between CloudFlare’s data centers and the websites connected to it. CloudFlare uses Airtel’s services to connect to host sites, which ostensibly gave Airtel the access to check each packet of data to see if they violated orders set forth by India’s Dept. of Telecoms.

Balakrishnan suggested that each connection between CloudFlare and content servers could be blocked by Airtel inserting an iframe, which he encountered when attempting to access The conclusion he reached was the Airtel was monitoring CloudFlare’s upstream traffic without its awareness, thereby snooping on traffic from all Indian users who access any one of the 2MM+ websites that use CloudFlare. Airtel denied the accusations in response to an inquiry by Medianama:

“This is completely baseless and incorrect. As a policy, Airtel does not block/sniff any content. Only in the case of instructions/orders from the Government or the Courts, specified URLs are blocked. Blocking of any page [as per instructions from relevant authorities] is done at the URL level and not whether it is http/https. This also has nothing to do with the validity of any certificate.”

In an interview with Medianama, CloudFlare CEO Matthew Prince said that the company had reached out Airtel regarding the interference with Pirate Bay’s traffic, which blocked their ability to connect to the customers origin. Airtel reportedly told CloudFlare that this particular customer had had access blocked due to a specific government request, which affected requests connecting to CloudFlare, and not requests from CloudFlare to the customers origin. While Airtel’s agreement with CloudFlare prohibits it from interfering with traffic, the only exception is in the case of government order.

Prince also notes that Pirate Bay opted to send traffic over an unencrypted connection, even though CloudFlare has the ability to pass it over encryption. Using encryption prevents Airtel from pulling the host header and redirecting traffic.

That being said, Barakrishnan’s concerns about user privacy remain open, as Airtel would have had to access the header of every unencrypted packet in order to block the site. While Prince acknowledged that this would problematic, he suggested that there was very little CloudFlare could do outside of providing the option of encrypted connections. Governments in various countries have circumvented this by blocking ISP’s traffic to CloudFlare in order to block certain websites.

Scroll to Top