Some time ago we have helped to set up an interview with Arista’s Technical Director John Peach for Czech IT magazine IT Systems. We believe it’d be a shame not to publish also the full English version of it. So here you go.
Openness, simplification, automation, and visibility are the key factors for efficient and flexible network operation… says John Peach, Technical Director for Arista Networks.
John Peach is the Technical Director for Arista Networks in EMEA. Since joining the company in 2010, John has established a technical team across the region, focusing on cloud and web-scale platforms, HPC, data centre and campus environments. In his current role, he works closely with customers and technology partners, engaging with R&D teams to define innovative products that address customer needs and develop new markets. John brings more than 15 years’ experience to Arista Networks, having previously worked for networking industry innovators.
John, first of all, can you please introduce Arista to the readers, that have not heard about your company yet?
Arista is a pioneer in software-driven, cognitive cloud networking for large-scale data centre, high performance computing and campus environments.
Founded in 2004, launched in 2008 by a team of industry veterans, Arista brings together a rich and extensive history in networking and innovation to create market-leading open and programmable solutions for our customers, building public and private cloud computing systems.
Arista’s market-leading switching and routing platforms, as well as our WIFI solutions, redefine scalability, agility and reliability and are supported by two advanced software solutions – EOS, an advanced network operating system; and CloudVision, our automation, telemetry and operations platform.
EOS (Extensible Operating System) is a ground-breaking network operating system with single-image consistency across all hardware platforms, and modern core architecture providing rock-solid reliability, open standards functionality and complete openness towards customization, extension, and integration.
What are, in your opinion, the current trends in data networks sector?
In the merchant silicon world, speed and density increases continue to follow Moore’s Law and therefore provide higher capacities and faster infrastructures for those that are at the head of the curve. This is important for some customers. However, there are some broader trends which I feel dominate.
I’d start with openness. This is as crucial to Arista as it is to our customers. At a high level, openness means avoiding the traditional lock-in to vendor or silicon specific capabilities that often limit customer choice in terms of product attributes, compatibility or lifecycles.
This means not only leveraging open standard protocols but also providing open APIs and the ability for customers to build a right-sized architecture using components that meet the particular price/performance/capability requirements for each project as well as to build and operate using the customer’s chosen set of tools.
The second major trend is simplification. Many infrastructures are unnecessarily complex in terms of topology or protocol choice, usually due to product scaling limitations, proprietary solutions or traditional segmentation of administrative domains leading to inflexibility, poor reliability, poor application performance, and high operational costs.
Simpler designs are clearly easier to operate with multiple vendors, allow more choice, improve stability and cost far less to run.
Finally automation and visibility. These are clearly critical to reducing reliance on error-prone and time consuming manual tasks involved in both the build and operate phases of any infrastructure. Automation allows for faster turn up and maintenance procedures while streaming real-time telemetry provides a full view of prevailing network conditions and drastically reduces the time taken to resolve issues.
At Arista we call this general trend the PINs to PICs transition; traditional ‘Places in the Network’ approaches are where each use case has an individual, very tightly defined set of devices and software tools tied to that specific silo and based on legacy assumptions. For example, using a different set of networking products for the data centre compared to campus, each with different operating systems, management tools and protocols that are either incompatible or very hard to seamlessly integrate.
Modern ‘Places in the Cloud’ approaches use a consistent, simple and highly automated build and operate paradigm for everything from Branch to WAN to Data Centre and the customer is only tasked with selecting the right hardware attributes for each use case.
Sometimes I think of this being similar to how a driver might select between a petrol or diesel car engine and 2- or 4-Wheel Drive, depending on how the driver wants to use their car; Without fundamentally changing the way they purchase, operate and maintain the car and without needing to build a special set of customized roads on which to drive it.
On which technologies from the data network sector should IT managers focus today?
IT is the backbone of any modern enterprise and all enterprise systems are becoming heavily influenced by the enormous progress in cloud-based technologies. Given the pace of change in the IT space, I think the single biggest lesson we can learn from the cloud is how to move from the legacy ‘big-bang’ lifecycle of IT projects towards flexible continuous development and integration.
Avoiding technology dead-ends is critical for both business agility and cost management, so the key attributes of Openness, Simplification, Automation and Visibility are the critical factors that allow a business to change direction rapidly and operate their environments efficiently and proactively.
Contemporary data centres have to face rapidly growing demands on performance and scalability.
At the same time they are forced to lower their expenses and power consumption. How do you think these basically contradictory demands can be satisfied?
From a networking standpoint, there are three principal factors that drive the ability to keep scaling compute resources.
Keep up with developments in compute and storage by providing faster interfaces
Drive down the cost of purchase and operation (power) per Gbit (or Tbit) with each new product generation both in terms of the device and the connectivity (optical components etc.)
Ensure that the density of connectivity continues to increase to avoid building inefficient infrastructures or introducing too much latency.
Along with the product life cycles, minimizing barriers to adoption of new products and ensuring they are extremely reliable and highly automatable are critical to enabling large scale customers to continue to innovate rapidly.
Arista’s approach of working closely with high end merchant silicon vendors and abstracting the specifics of the product using a single consistent operating system (EOS) that is designed from the ground up for highest reliability and simple automation ensures that each product generation is instantly accessible and does not require a complete paradigm shift in software, hardware or knowledge transfer.
In the last year, you have acquired Mojo Networks – a wireless network company. How did this acquisition affect your portfolio? If I am not mistaken a primary intention of the acquisition was not access to a wireless network segment but a strengthening of possibilities of your CloudVision solution.
As you are aware, Arista has been tremendously successful in the Data Centre and in many wide-area infrastructure deployments with customers realizing significant benefits in terms of reliability and simplicity of operations.
For a number of years, customers have also been using our products in various campus use cases to bring some of the advantages of EOS and CloudVision to other areas of their infrastructures.
This drive to unify the approach to IT led to our 2018 announcement of Cognitive Cloud Networking, designed to bring operational consistency and modern cloud principles to the enterprise campus, Arista extends this architecture with the addition of Mojo Networks which eliminates the traditional, closed wireless controller model, to bring radical improvements in scale and economics.
Secure, high-performance cognitive WiFi at cloud scale is critical for modern BYOD and IOT intensive environments and the integration of Wifi into the Arista portfolio allows CloudVision to become a single pane of glass for Data Centre, WAN, Campus and Branch infrastructure.
In connection with a network infrastructure, there has been a vivid discussion about security and technologies of certain companies that are not, according to regulatory bureaus for cybernetic security in some countries (including the Czech Republic), trustworthy. How does Arista face these rising fears of various companies and organizations of cybernetic risks?
There are multiple aspects to this question – the perspective of how Arista engineers its products on one hand and how we help customers with security concerns on the other.
On the first question, I’d say one of the largest issues faced is a lack of transparency in the design and build of operating systems deployed across all types of IT products. Commonly appliance-type products run closed, custom operating systems that are not open to global scrutiny and may therefore contain unintentional or malicious backdoors or exploits.
At Arista we are proud to build our software on an open Linux distribution; this means that the core operating system is both open for code scrutiny and, as importantly, is put to work in many millions of devices in many use cases beyond networking devices where there are different types of threats. Working with the Linux community we both ensure we maintain knowledge with prevailing threats and feedback our own discoveries.
On the hardware side, many products are based on underlying OEM platforms that have not been developed in-house and therefore are also candidates for exploitation. At Arista, we design our systems in-house and include specific integrity mechanisms to ensure our devices are not tampered with.
To briefly answer the second part of the question, many customers recognize the need for strong segmentation and encryption technologies for data crossing public networks or across 3rd party fiber between sites. Historically external encryption devices have been extremely expensive and lacking in the required performance for modern applications.
This is a focus area for Arista – driving down the cost of implementing security at Data Centre speeds by integrating encryption directly into our products at line rate.
Your company has recently introduced a new 7360X series. What innovations does this series offer? What functions and attributes of the new series are in your opinion the most crucial?
Modern cloud networks and intensive HPC, Machine Learning/AI environments require extremely large leaf-spine networks with a high degree of multi-pathing in order to scale compute clusters efficiently without losing performance or introducing excessive costs.
The maximum network diameter is constrained by the radix, or total port count, of the spine tier, the capacity of a single system and the scale of the multi-pathing architectures.
The 7368X4 provides the industry’s highest single-chip radix at 128 ports of 100G which doubles the maximum network diameter and dramatically reduces power, cost and latency by reducing the number of switching chips in the network path which in turn means more resources available to scale compute clusters further.
Outside of specialist infrastructure, the ability to support a mixture of 100,200 and 400G also makes the switch relevant for many general enterprise use cases and the common EOS operating system ensures that the 7368X4 can easily slot into any open and programmable infrastructure.