The networking market is at an exciting pivot point, evolving away from legacy enterprise networking to the cloud. While the public cloud providers (“titans”) may take a “do it yourself” approach to engineering cloud network designs, mainstream enterprises demand a “cloudified” turnkey solution and want to emulate cloud operators. The increasingly massive scale of address tables, devices, flooding, broadcast traffic from discovery protocols, subnets and routing protocols have accelerated the need for disruption in networking workflows, making Arista a unique and welcome pioneer for customers ready to make SDN a reality.
To appreciate the need for SDN and cloud solutions one must step back and understand why the cloud network is dramatically different from legacy networking.
First, in a cloud, everything is dynamic. Resources become available and go off-line, users are logging in and out, and workloads are going up or down depending on compute needs. This is a fundamental difference of cloud versus static computing in enterprises.
Second, cloud data centers are much larger than typical enterprise datacenters and can contain tens, even hundreds of thousands of servers. Legacy management practices and policies that are used in smaller datacenters don’t apply to clouds since Continue reading
Scale-out cloud-applications continue to be the most disruptive force to traditional network architectures in the data center. They demand an open and uncompromised cloud network, unshackled by monolithic and the prehistoric proprietary networks.
Our customers and partners are reshaping this traditional networking industry. Previously burdened by monolithic software and underperforming hardware, the new evolving bifurcation of computing architectures for modern cloud-based applications has resulted in key trends and drivers for this evolution including:
1. Unstructured data is everywhere. Whether from users, applications or machines, it is growing exponentially with no vertical or industry being spared. One current obstacle to working with large data sets is the use of relational databases and desktop statistics/visualization packages that require massively parallel software running on hundreds, or even thousands of servers.
2. New workloads are changing the notion of separate SANs. This impacts the role of networking and IP storage, where virtual machine mobility sustains multiple gigabits of throughput by default with multicore processors.
3. Cloud intensive applications such as content distribution, and new infrastructure technologies such as containers and Hadoop clusters are pushing the envelope of what is possible with massively parallel transactions. These new large-scale data analytics have given birth to converged Continue reading
The networking industry is changing rapidly with demand for more dynamic control of big data, and scale-out cloud applications. The inevitable shift to software-defined workloads and workflows is crystal clear. This calls for more “disaggregation” of software models for the network stack.
Should networking move to a more open standards-based approach or continue with legacy stacks? The answer is defined by considering the following three options for networking software stacks, and the actual choice depends on the factors as shown below in Figure 1.
1. Classic OS: The established network vendors tend to develop multi-million lines of complex software code with enterprise or service provider class features for LAN or WAN. This closed monolithic “Blob-OS” model can be based on modified and proprietary versions of a BSD or Linux Kernel. Traditional enterprise support is the hallmark of this model, but innovation is rarely evident. Claims of programmability usually include band-aid APIs, or guest virtual machine access. The classic OS addresses mature markets akin to a mainframe usually for customers with siloed IT stacks that support legacy applications.
2. Cloud OS: At Arista, our software engineers build based on an open Linux Kernel, providing programmable capabilities that legacy switch-based Blob-OSs do Continue reading
SDN (Software Defined Networking) is finally becoming clearer. It is not “Still Don’t Know” nor is it a specific overlay controller. Simply put, it is an open and programmable way to build networks for customers looking at utilizing hybrid combinations of public and private cloud access.
We are witnessing a shift from multi-tier oversubscribed legacy enterprise networks to two-tier leaf-spine or single-tier Spline™ cloud networks with east- west traffic patterns scaling across thousands of servers. Arista was the first to introduce this new architectural “leaf-spine” approach for cloud-based networks and five years later others are still attempting to mimic. Lets review some practical examples.
Facebook: Take an important and familiar social networking application, Facebook. Their public information shows that they deploy a memcache architecture, which allowed them to reduce the user access time to half a millisecond by using fewer network tiers, resulting in lower application latency. As we log into Facebook, the single login request triggers thousands of look-ups on databases and memcache servers. Legacy enterprise multi-tiered networks would result in delayed look-ups and would negatively impact the user experience and interest in a significant way.
Amazon: Shopping couldn’t be easier than online on your favorite site. Have you Continue reading
Every year I reflect upon how my predictions compare to actual outcomes. Once again, that time has come, so let’s take a walk together down 2014’s memory lane, while also looking forward to exciting industry developments in 2015. Clearly innovation in networking is returning as we are seeing venture capitalists once again investing in networking innovation!
Prediction #1: The rise in server virtualization is driving network virtualization deployments.
Evaluation #1: Half True.
One can transcend network boundaries at both L2 and L3, building seamless virtual and physical networks with VXLAN as the key L2 over L3 foundation. The VXLAN specification co-authored by Arista and VMware, and in a similar vein the NVGRE specification co-authored by Arista and Microsoft, were key turning points for network virtualization. Arista’s strategic partnership announced in August 2014 with VMware (NSX, vSphere and vCloud Director) and multivendor interoperability with other controllers from Nuage Networks, OpenStack and the OpenFlow community were key milestones in 2014. New protocols take time to be adopted – usually 3-5 years. VXLAN is at that tipping point for broader implementations in place of the proprietary, vendor-specific options we have seen.
Prediction #2: “SDN” is no more “Still Don’t Know”.
Evaluation #2: Continue reading
The innovation of hundreds of startup companies created the Internet, and the Internet has changed the world. Innovation continues to have a dramatic impact on networking in recent years. These new developments have changed the way applications, workloads and networks interact. Having been involved in this industry for more than three decades, I have witnessed and been part of these transformations from the 1980s to the 2015 era. Each phase of innovation has been characterized by new companies and entrants, as depicted below:
| PHASES OF NETWORK INNOVATION | Epoch | Vendors | Network Technologies | Trends |
|---|---|---|---|
| First 1980–1995 |
AT&T, Sun, 3Com, NET, Proteon, UB, BBN, DEC, IBM |
ARPANET, Circuits, Hubs, SNA, Ethernet, Token Ring, Routers |
Terminal-Mainframes and Minis, Channel Attach |
| Second 1995–2010 |
Cisco, Juniper, Nortel/Bay, Alcatel, Lucent, Avaya |
Switching, Multiprotocol Routing, LAN-WAN, TCP/IP |
PC, Client-Server, Web, North-South traffic |
| Third 2010–present |
Arista, VMware, Facebook, Microsoft, Splunk, Red Hat, Palo Alto, Aruba, many others |
The SDN Era of Open, Programmable Networking, DevOps meets NetOps, Universal Cloud Networks |
Mobile Virtual Workloads and Workflows, Big Data, Hyperscale Web, Virtual Machines / Containers |
Dominant companies often fall by the wayside when they do not anticipate and react to clear market trends as Continue reading
Every year at Arista has been filled with milestones and enriching memories. I want to acknowledge and thank each and every Arista well-wisher for contributing to this incredible experience. So let’s take a walk together down memory lane and look back on our journey from start-up to a now public company. I think of Arista’s first decade as being comprised of three phases:
Funded in a unique fashion, without traditional venture capital investment, Arista (first called Arastra, located on Arastradero road in Palo Alto) was placed in a unique position. Our founders, Andy and David, were our funders, too, and they cared deeply about building the company with the right technology foundation. Ken Duda, also a founder and our EOS software genius, brought a radical, resilient and programmable network-OS for modern, disruptive applications. Bringing some of the best and brightest engineering minds together resulted in an innovative network-wide operating system that challenged legacy enterprise switching vendors. Andy Bechtolsheim and I launched the company officially in October 2008. With just 50 engineers we gained 50 customers by the end of 2008, proving what small focused teams could accomplish. Our early adopters welcomed us as a breath of Continue reading
Legacy networking vendors have often declared that they do not build “boxes” but actually build “systems and architectures”. I have tried to understand what that really means. Undoubtedly, new applications on the Internet have evolved and now depend on a modern infrastructure that outlives any particular static workload or physical machine. Indeed, at Arista, we along with our customers are looking at a technology evolution that goes beyond individual components to a universal cloud architecture.
Building a “best of breed” product is a vital foundation to building a good system. Typical metrics may include feeds and speeds such as latency, power, port density (non-blocking), fabric capacity, throughput and scale combined with a feature-rich network operating system. Examples of this include Cisco’s routers and Catalyst switches (with IOS) in the 1990s, F5’s Big IP and Splunk’s data analytics in the past decade, or even more recently, Arista’s 7000 series Leaf-Spine-Spline products. Vendors with breakthrough products are pioneers and thought-leaders in their markets and often establish trends rather than follow them. Sustained differentiation beyond point features is a common theme. Such products are disruptive in nature, bringing critical business benefits and reducing capex spend within Continue reading
Leading customers and researchers in cloud and data center networking have been promoting the importance of understanding the impact of TCP/IP flow and congestion control, speed mismatch and adequate buffering for many decades. The problem space has not changed during this time, but the increase in the rates of speed by 100X and in storage capacity by 1000X have aggravated the problem of reliable performance under load for data intensive content and for storage applications, in particular. One Arista fan summed it up best by saying:
“Basically the numbers have changed by order of magnitude, but the problem is the same!”
Poor performance and inadequate buffering in a demanding network is a painful reminder that buffering, flow control, and congestion management must be properly designed. TCP/IP was not inherently built for rate-fairness, and packets are intentionally dropped (yes, only window fairness is possible). Yet the effect of these drops can be multiplicative given major speed mismatches of 10-100X inside the data center. In the past, QoS and rate metering were adequate. However, at multi-gigabit and terabit speeds and particularly as more storage moves from Fiber Channel (with buffer credits) to Ethernet, packet loss gets more acute.
The adoption of cloud networking architectures by both the hyper-scale cloud companies and increasingly enterprise networks proves the need for open standards and modern networking software to gain the benefits of agility, programmability and resiliency. These architectures are all driven by the move to standardized topologies and container-scale deployment to achieve cloud economics.
The recent Facebook introduction of a reference design to align to the OCP (Open Compute Platform) server project with a network switch (“Wedge”) based on a Linux OS is a good benchmark for the use of open standards, control and merchant silicon. While many may view this as a threat to legacy proprietary networking, to me it’s a welcome validation of Arista’s approach to building modern software that is open and programmable as opposed to a proprietary, bloated and complex legacy OS. It is also a symbol of Arista’s co-development of APIs offering access for specific application control in Facebook’s network. This is a fitting example of how “white box” technology could be applied to a specific SDN use case. It is not trying to address broad data center use with multiple applications and mobile workloads.
Two factors are driving Continue reading
You may have noticed Arista being uncharacteristically quiet throughout an imposed “quiet period” leading up to our IPO (ANET). While the industry continued to speak out on behalf of, or against, Arista, the company remained true to its focus on solving customer problems through disruptive cloud networking architectures and technology. Enabling innovative applications to take advantage of modern networking through Arista EOS remains a key priority for the company.
The migration from legacy “policy per application” to universal cloud networks is crystal clear and underway. The classical 1990s web, file and database tier in client-server architectures, with north-south traffic, is migrating to universal workflow telemetry and workload automation for east-west, server-to-server traffic of the 21st Century. This is particularly true as networks move to mandates of terabit scale for data, control, and management. Arista invented the concept of “Leaf-Spine” and late last year introduced the “SplineTM” as architectures to support these next generation network requirements. Five years later others are still trying to mimic the approach but lack the fundamental software or scale to achieve it.
Through the years we’ve witnessed many failed attempts to lock-in customers with proprietary fabrics (FabricPath, QFabric, etc.) and once again we are witnessing Continue reading
Enterprises are still a complex mix of legacy and newer cloud applications, yet smart use of universal SDN-based cloud networks is the great equalizer in bringing enterprises and the new applications of the cloud together. Evolutionary migration strategies from a mainframe to a client-server architecture can also be applied to the next phase of cloud and virtual age networking. To appreciate how they apply, one must better understand the diverse definitions of SDN, and its true applicability in next generation networks. Let’s review some of the terminology often used and confused in our industry.
Overlay SDN: The most visibly promoted controller for SDN overlays today is VMware’s NSX (Microsoft System Center, Juniper Contrail and Nuage Networks may also fall into this category). Some networking features and functions are moved into overlays to control the data, flow or forwarding path. This includes:
1. Software overlays to shift management functions from the control plane of the network to servers
2. Specific use-cases such as server virtualization, L4-L7 load balancing, security, Openflow etc.
Functional controllers leverage existing physical networks and apply features and functions such as provisioning that can be used via abstraction, APIs, a CLI and limited scripting.
Underlay SDN: Controllers do Continue reading
Lately, two acronyms have been making the rounds: SDN (Software Defined Networking) and ACI (Application Centric Infrastructure – promoted by Cisco). Both have things in common which equate to great marketing: a delightful vision and being difficult to pin down in terms of a crisp definition. Let me try to clarify as best I can with the disclaimer that this is purely my perspective representing Arista, as we celebrate the deployment of our second million ports of cloud networking.
The common view is that SDN is a controller or a set of network management products based on Virtualization Technologies or OpenFlow. At Arista we have a more pragmatic view. To us, SDN is a programmatic suite of open interfaces that allows applications to drive networking actions. Unlike the misconception that SDN is just a controller, I believe SDN is about scaling the control, management and data plane with programmatic and open interfaces. This means customizing the network with high-level scripting and programmatic languages, structured and machine-readable APIs, and standards-based protocols as well as interoperability with controller-friendly networks.
As we enter 2014, we are witnessing the deployment of SDN via Arista EOS and associated programmable network applications such as Advanced Telemetry, Continue reading