Networking industry evolved at a snail’s pace in the post Y2K era compared to the rest of the tech industry. Industry experts attribute that trend primarily to some of the mis-steps that Cisco, the networking bellwether, took during the post Y2K era. Cisco dug itself into a chasm by getting into the Consumer Electronics Business. It rebranded itself as a household name for consumer electronics products with a new logo, acquired Scientific Atlanta – the set top box company, made it’s own version of iPad named Cius, acquired Flip camera business, rebranded Linksys wifi routers as Cisco wifi routers and many more. It clearly missed to foresee, listen and take advantage of the “cloud” opportunity that was knocking on its front door.
During the same period, web2.0 companies such as Google, Amazon, eBay and Facebook grew exponentially. These companies started adding more services to their offerings which brought in additional traffic to their websites. YouTube, for instance, was one such offering. So they added more servers in their datacenters. Additional servers required additional networking equipments to interconnect them. So they bought more networking gears. That’s when they started noticing the disconnect between the trends in the server industry and the networking industry.
The cost of the servers was going down significantly due to the commoditization of the server hardware through open server architectures. Also server side technologies were evolving rapidly – thanks to Vmware, which brought in the server virtualization technology that maximized and optimized the server utilization. Overall the trend was healthy – server cost going down and performance/utilization going up.
To the contrary, the networking industry trend was quite the opposite: closed proprietary architectures, premium pricing model, low link utilization – thanks to the spanning tree protocol that placed redundant links in blocking state, and more importantly crappy softwares that came with more and more unwanted features packaged in them. The networking vendors competed primarily in terms of hardware speeds and feeds. The networking software did not evolve during this period – monolithic kernels, processes corrupting each other with shared memory spaces, legacy architectures that doesn’t scale well, proprietary features that didn’t interoperate with other vendor devices, and lack of manageability, programmability, and debuggability features.
The web2.0 companies did not like this trend in the networking industry. They saw this as a threat to their profit making, in some cases and to their survival in many cases because most of the web2.0 companies were either not making any revenue or were at the mercy of ad revenue. They wanted to change this trend in the networking industry but their voice was not heard. Cisco which owned almost 80% of the datacenter market was very adamant and unwilling to change the status quo. The web2.0 companies decided to build the ammunition for the battle in their own backyard, silently and diligently. Google formed a small team in 2007 to build their own switches and routers using merchant silicon. They used Openflow based SDN to run and manage their production network. They eventually opened up their architecture and solution to the world to prove the point that one doesn’t have to buy premium priced products available in the market to build and manage a very complex and scaled network similar to that of the Google. This specific event forced the CIO’s of various companies to rethink about their decision to investment in legacy but premium priced networking technologies available in the market. It also raised questions in the minds of various network operators on how they install, run, manage, and upgrade their production network.
It was not just Google that formed a small team in 2007 to compete against Cisco. A bunch of early investors in Google, who were also ex-Cisco veterans, talked to Google about their challenges in building and managing their networks using the legacy products available in the market. The team quickly put together a business plan and formed a company known as Arista Networks. Arista also built their products using cost-effective merchant silicons and more importantly they architected their networking software from scratch. They truly built a next generation networking OS that scaled well and had plenty of open standards based programmability, debuggability and manageability options. Moreover, the datacenter requirements were very simple compared to the enterprise and service provider segment requirements. So Arista was able to easily break the entry barrier into the datacenter segment.
Cisco felt, for the first time, that they were being challenged. Cisco initially resisted this force – the force that was collectively being aligned behind the SDN movement, which asked for more open standards based architectures, more programmability, debuggability and manageability options. Cisco was unwilling to give up its premium pricing model and the huge monopoly it had over the networking industry. However, the industry forces pushed Cisco to a corner and forced it to start embracing the SDN movement or walk out of the market segment. Cisco, which didn’t have a choice, eventually began opening up some pieces of its software. It provided open API’s to program its software. It implemented Openflow protocol in its boxes. It was willing to sit down, listen and work with the cloud titans – Google, Facebook, and Microsoft. It even built products using merchant silicons. One interesting thing to note about Cisco is that, once it has decided to make a move in a particular direction it does so with full commitment. Only Cisco could do that because of the huge war-chest that Cisco has. Cisco committed additional money, resources and time to correct some of the mis-steps it took in the past. It didn’t fix just one product it fixed its entire productline to become more programmable, debuggable and manageable through a centralized controller. Cisco is also making notable strides internally to change the culture – to become an agile organization with more focus on automation. It even revamped its entire top management team, including the CEO.
Unfortunately, Cisco realized this change in trend too late. This delay provided Arista to became a formidable force in the cloud datacenter networking industry. What made Arista a unique beast among the other failed competitors of Cisco is that they spent almost 4 years to build the OS and development tools from scratch that enabled them to produce a high quality software. Arista’s EOS was architected for scalability, modularity and handling complexity in the cloud. The development and testing tools used at Arista simplify the development and testing efforts, significantly improving the employee productivity, satisfaction and code quality.
I have worked for four different networking companies in the valley – Foundry Networks, Brocade Communication Systems, Arista Networks and Cisco Systems. I can confidently say that Arista’s EOS software is miles and miles ahead in terms of software quality and functionality. Everyone else in the industry focussed heavily on the hardware while Arista focussed heavily on the software. I would attribute Arista’s success to their unique culture: quality is in their DNA, quality of the product is more important than the additional revenue generation opportunity, employees feel welcomed and respected, open cubes for managers and no office rooms, everyone is empowered to fix anything that is broken or to do the right thing for the company, everyone is valued for their contribution and not for their years of experience, and a 360 degree based performance appraisal system.
No wonder Arista is so successful in the networking industry. They were able to foresee the forces that changed the trend in the networking industry, initially though only in the datacenter segment. They built products around this trend. Now they are taking this trend and products to the Service Provider and Enterprise segment.
Cisco for its part has diversified its investments away from routing and switching into areas such as network security, network analytics, IoT, Cloud Application monitoring, online collaboration tools such as Webex, Cisco Spark, Telepresence and many more.
As with any industry that gets matured and gets consolidated, the networking industry also matured and got consolidated. Foundry Networks got acquired by Brocade Communication Systems which inturn was acquired by Broadcom and Extreme Networks. HP acquired Avaya networks. HP formed a partnership with Arista Networks and stopped making networking gears for the datacenter segment. Dell acquired Force10 Networks and closed many of their productlines. Ericsson closed down its operation of the RedBack Networks division that it acquired in the midst of the housing bubble.
The last decade saw many interesting changes happen in the networking industry, for good. It will be interesting to see how Cisco, Arista and other remaining ones play out in the market. Moreover, the merchant silicons are becoming more and more programmable and cost-effective. It will be interesting to see how Broadcom, Barefoot Networks, Marvell(Cavium/Xpliant) and the MPLS startup Pensando Systems play out in this market. Evolution doesn’t seem to stop here in the networking world.