Arista Networks is a little different

Arista networks is a company that provides cloud computing network equipment for data centers. Its core strength is its network operating system EOS. In addition, Arista has several Facebook cards worthy of playing. The company's founders, current CEOs and chief scientists are all big players in the industry. Especially the understanding and emotions of the network may not lag behind any network giant, and try to pass simple, open standards. And the unique EOS soft power creates magic.

Looking at Arista's management team, CEO Jayshree Ullal was responsible for the Cisco Data Center Exchange product line for 15 years; the other two founders' resumes will be scared, Andy Once a founder of Sun, Granite, founded with David Cheriton, was acquired by Cisco and laid a solid foundation for Cisco switch products.

Although it is obvious that Arista, a young company, can make a big impact on network giants such as Cisco and Juniper in a day. But Arista was born at the right time, catching up with the transformation of the data center network architecture, and the breakthrough application trend brought by virtualization technology and cloud computing. And Arista also seized the opportunity to create greater success and saw the huge growth potential of the Chinese market in the cloud computing market in the future. Some time ago, Arista Asia Pacific Business Development Director Fang Bin shared the advantages and concepts of technology and product solutions with Alibaba's Asia-Pacific Business Development Directors Forum. The rear bin accepted an exclusive interview with ZDNet.

Network needs new ideas

When virtualization and cloud computing come, the original three-tier network architecture is more and more incompatible with the data center architecture, when performance bottlenecks, management Beginning to become complex, hardware network devices how to adapt to the flow of things brought about by virtualization, so that the network becomes more open and integrated to break the traditional chimney barrier. So the network needs new ideas and changes.

Fang Bin said, “Arista believes that vendor-specific chip update iterations are slow, closed software architecture, product ideas designed for traditional E-mail services have become the bottleneck of the network, such network architecture and The closed idea has been unable to adapt to the design ideas of data centers and cloud computing applications. The network needs more optimized architecture design to bear more east-west traffic; and Arista believes that network equipment self-developed chips have brought bottlenecks to the development of the network industry. It is a good way to choose commercial chips; and closed software architecture, It is also becoming a stumbling block to network innovation. Enclosure means that upper-layer applications can't get more intelligence from the network, and traditional network equipment vendors firmly control the advantages of software. Such a closed idea, Arista thinks that it needs to change. ”

So Arista is particularly heterogeneous in the eyes of peers. What is the difference between Arista and the user? What can be brought about by changes in the data center network architecture? The use of mature commercial chips seems to be easy, but will it lose the reputation of technical leadership? To what extent is the open network operating system open, does Arista not consider economic value from EOS?

The data center network architecture concept is not necessary to fancy

Compared with the fabric path advocated by Cisco, Juniper's Qfabric, etc., Arista has not given a similar data structure to its own data center architecture. Afabric&rdquo Marketing logo. Fang Bin said frankly, "Secondary network architecture was first proposed by Arista, and Arista called him & lsquo; branch level and leaf level & rsquo;. If you ignore the three-tier architecture of data center interconnection, today's data center common features are two-level, that is, branches and leaves, leaves in the cabinet, branches in the data center core. So everyone's data center network architecture is similar, some are their own private standards, but Arista always adheres to open standards; and regardless of the future cloud, how to develop, Arista does not agree to use the privatization technology zone to build a cloud for all. At the same time, there is still a large degree of uncertainty in the development of cloud computing. If the cloud does not have a complete consensus, following the closed concept, is there a big risk? The reason why Arista's data center network architecture has no fancy name is because Arista's technology is based on standards and is inherently simple. In addition, Fang Bin also mentioned that “Arista does not want to bind users through proprietary technology and let users make painful choices; there is also the need for private technology to let everyone follow. Existing technologies cannot solve problems, scalability, manageability, etc. cannot be solved with standard technologies. In fact, Arista has given a solution using technology based on proven standards. Private technology may appear at any time, but it will not be defined forever as the ultimate solution, so & lsquo; first standard technology & rsquo; is just a transition. ”

Is the scalable operating system EOS a killer?

In Fang Bin’s eyes, I hope that users will not regard Arista as a pure network equipment supplier. In his keynote speeches and interviews, he emphasized the importance of Arista's scalable operating system EOS. The biggest difference in EOS systems based on open source Linux systems is to maximize the openness and maximize the release of intelligent network operating system software.

EOS is a product that runs on Arista, all modular, multi-process, multi-threaded, stateful restart; core sysdb can communicate with all session states and processes; Simplify operation and maintenance management and operate an operating system for third-party applications based on a scalable architecture. Fang Bin said frankly, “EOS is the most unique place of Arista, and it is also the core value of Arista. Especially the core sysdb is revolutionary. sysdb is composed of flat files. Users can log in to sysdb to see the session state and process. Communication, and allowing users to develop applications, only Arista currently allows third-party software to be placed at this level of the operating system. Fang Bin stressed that “EOS has two important characteristics, open and sysdb centralized state management, sysdb architecture is very optimized, on the one hand makes Arista internal development very efficient, on the other hand Arista Providing the programming structure externally makes internal and external communication easier, giving customers and partners the flexibility to customize EOS to meet specific environmental, application and management integration needs. Unlike a few subsets that only open API interfaces to third parties, only limited secondary development can be done. For Arista customers, EOS can even allow users to modify mature OSPF and BGP network protocols. In fact, EOS is open to a possibility. It is understood that the current EOS version is EOS4.9, Arista created a completely open and open EOS open community site EOS Central, in the community users can share code samples, develop binary files, based on VM Emulator and API access and development and more. Provide users and partners with a platform for value-added development and inspiration.

Arista will not be a big deal for chip development

Fang Bin said bluntly, "Arista will not go to the chip of network equipment by itself, and the chip itself is the bottleneck of the industry. Everything is a good way to be in the X86 and open source eras today. ”

Actually, we can see. The ability to develop network device chips has always been one of the indicators for assessing the technical leadership of a vendor. They are also the industry leader for a few vendors. At the same time, we have observed that many manufacturers have not given up on the pursuit of commercial chips, even walking on two legs. Such a game will continue in a short period of time, and the idea is also benevolent.

Fang Bin said, “Independent R&D chips were once a product of a specific historical stage and are inevitable because the commercial chip market was not mature at the time, that is, at that stage, manufacturers need to promote technology development, and they need Independent research and development of chips. But today, the speed of performance, the frequency of commercial chip replacement is catching up with the speed of independent chip development; at the same time, specialized chip manufacturers are more professional, the team size is huge, and the division of labor is meticulous, they are constantly optimizing the underlying circuit. , the underlying module, optimize the middle-level logic and architecture, and then judge the market demand from the application level and combine the market needs to invest in research and development. In this regard, network manufacturers independently develop chips, which is somewhat incompetent compared with professional commercial chip manufacturers. So we can see that the upgrade cycle of commercial chips is 12-18 months. Network equipment vendors often choose commercial chips from generation to generation. The cycle of proprietary chips is usually 3-5 years, and the development cycle is too long. There are huge challenges in forecasting market demand for functions and performance, and now the pace of R&D and upgrade of proprietary chips is getting slower and slower, and the gap is getting bigger and bigger. Therefore, the chip has such a divergence, on the one hand, the commercial chip fast product can not keep up, and the other side is the special chip development slow product. Arista's philosophy is to find the right place for you, to help customers find the greatest value, and not to go big. ”