Clever Geek Handbook

Selectel Networking Academy Conference





On October 24th, a multi-brand conference on network technologies - Selectel Networking Academy - was held in our conference room. Representatives of the largest manufacturers of network equipment - Extreme Networks , Juniper Networks , Huawei and Arista Networks made presentations at the event.



In this article, we will discuss both the topics discussed at the conference and the speeds at which the network equipment serving the data centers operates. In addition, we will share with you the current tasks for the further development of the Selectel network infrastructure.



Want to know how much Tbit / s will pump the new Juniper Trio chipsets in 2020 or what kind of "rake" you may encounter when upgrading to major versions of network equipment software? Then welcome to kat!



The conference began with a welcome speech by Kirill Malevanov, Selectel Technical Director:

“Selectel is a center of competence in various areas of IT: in the server part, in the software part, in the clouds. But one of the most unforgettable parts of the data center is, in general, the network that lies at the head of all these services ... Today we are going to talk about the network and have gathered here reputable vendors of various network equipment ... "
After completing the presentations, the participants were asked to organize a panel discussion on one pressing topic: the life cycle of software and hardware.



Extreme networks



The first to submit their report to Extreme Networks. Over the past 5 years, Extreme Networks has been actively developing through the acquisition and takeover of various network business companies. The main goal of the acquisition is not so much the acquisition of a business as the development and integration of advanced network technologies in their own products. Recent landmark acquisitions include the purchase of a wireless business from Zebra, a network business from Avaya and Brocade.









The topic of the report of Extreme Networks' representative, Pavel Denisov, was about border routers, the so-called Border Routers. These are devices that connect trunk and peripheral networks and are characterized by the following parameters:





Extreme Networks offers two models in the SLX line as edge routers:



  1. The flagship Extreme Networks SLX 9640 model with 24 10 GbE ports and 12 100/40 GbE ports. The device supports up to 5.7 million IPv4 routes with Optiscale proprietary compression technology.
  2. The “younger brother” of Extreme Networks SLX 9540 with 48 10 GbE ports and 6 100/40 GbE ports. The number of routes in it is lower (1.5 million), but this number is quite enough for the role of the border router.


An interesting feature is the ability to run a guest virtual machine inside the router. It's no secret that all modern routers are, in fact, just virtualization servers, inside of which there is a virtual machine programming ASIC chips to process traffic.



Extreme Networks engineers went even further and made it possible to upgrade their separate guest virtual machine with a dedicated 10 GbE interface. This makes it possible not only to easily analyze traffic, but also to automate many processes, thereby integrating network equipment into the company's ecosystem.



Selectel Network Infrastructure Development



We also did not stand aside, and talked about the development of our infrastructure. Previously, the Selectel network of data centers consisted of two large routing domains - in Moscow and St. Petersburg. Each domain was served by a separate router. In the event of any problems, this could affect the region as a whole. With the growth and emergence of a large number of critical services, we thought about the fact that it is necessary to rethink the approach to building a network.



For the several years that we have been working on the network, we have been solving a set of tasks, one of which is dedicated servers (Internet and local area network). The network of dedicated servers is the simplest and is built according to the classical architecture: root routers - aggregation level switches - access switches. But there is an important feature - the servers of one client can easily be in different racks. As a result, we get a large number of VLANs, “stretched” to many racks.



Another task is to ensure network operation for virtual servers in the cloud. We use the Spine Leaf topology to ensure minimal packet transit times from one server to another. The active use of the IP-storage concept imposes its own characteristics, in particular the presence of a large number of high-speed links. At the same time, the network should adequately perceive the situation when part of these links ceases to function.



Well, perhaps the most important task is the network between the data centers ( Data Center Interconnect ). Here we use a separate network with its own rules and redundancy. The use of such a multitude of networks is due to the needs of our customers, who need to not only gain access to all services, but also connect them together. For example, for cloud servers of one customer to have access to a network of dedicated servers, and even physically located in different data centers.









Currently, we have the following priority areas for network development:





During operation, we often encounter quite non-trivial problems. For example, some vendors do not consider the stack as a backup tool, but rather as expanding the number of ports on a single device. When updating the software of such a stack, a complete reboot of the components of the stack is required, which we simply cannot afford.



Other vendors consider the stack from the point of view of Fabric , that is, in theory, the switches are independent enough to update the software separately from each other. This only works well when updates to minor versions occur. Service interruption does not occur, despite the need to reboot the equipment to complete the update. But as soon as the transition to the major version of the software is required, difficulties are possible. No one guarantees that the stack will function smoothly during the upgrade process.



The solution to this problem is the creation of the so-called Update Groups to update the switches in batches. But the problem is that by updating such a group of switches, we won’t get control over them until the remaining groups are updated. Whether this will work correctly or not after the update - the question remains open. To assemble a full-fledged test bench, for example, from 16 switches, costing more than $ 10k for each, is quite problematic.



Juniper Networks



Then Juniper Networks took the floor. Since 1998, Juniper Networks has been releasing innovative solutions for service providers, but over the past 10 years more attention has been paid to the segment of corporate customers and data centers. In addition to the development of devices and microcircuits, the company’s priority is the development of its own JunOS operating system, which has become one for the entire range of manufactured equipment.



First of all, attention is paid to the uniformity of functionality and control of various devices. The management plan uses one set of code so that when adding any functionality it appears immediately on all platforms - from hardware devices to virtual solutions.



To automate routine operations, JunOS supports a wide range of tools - from programming languages ​​to various APIs. The Junos Continuity concept allows the introduction of new modules into devices without the need to upgrade to major software versions.



Oleg Prokofiev, a systems engineer at Juniper Network, spoke about the MX series of routers, as well as developments in the field of creating their own chipsets called Juniper Trio , which are actively used in the architecture of hardware solutions. For 2018, these chips are capable of processing traffic of more than 0.5 Tbit / s on a single chip. In 2020, it is planned to release a new version that can do this three times faster, reaching 1.5 Tbit / s on a single chip.



Inband Telemetry Technology



A presentation by Alexander Bespalov, Arista’s systems engineer, was dedicated to the new Inband Telemetry / Inband Flow Analyzer (INT / IFA) technology developed by the IETF with the participation of Arista. At the moment, the technology is still in the draft stage (currently the second version already exists).



The meaning of the technology is that each transit device adds its own headers to the transit packets with a sufficiently large amount of metadata, including:





Devices that support INT / IFA operate within their own INT / IFA domain. The device that starts adding metadata is called the Initiator Node . The transit nodes of the domain are designated as the Transit Node and update the headers, adding their portion of information to the transit packets. The device on which metadata from packets are removed is the Terminating Node .



When using the Spine Leaf topology, Spine nodes will serve as transit nodes, and Leaf nodes will be initiators and terminators, respectively. There is another option, in which Spine and Leaf will play the role of transit, and the terminal servers will be able to both initiate the process of collecting metadata and accept them for further processing.



The process of adding headers is very simple: on the initiator side, an IFA / INT header is added once, which does not change further when the packet passes through the network. After this header, the collected metadata will be added when passing through each transit node within the domain. The end node exports the collected data to the analytics server and completely removes the header and all metadata from the IFA / INT package. This allows you to understand with the highest accuracy exactly how packets pass through the network and identify problems long before they affect the operability of the network infrastructure.



Huawei



The presentation by Alexander Dorofeev, Huawei Key Account Manager, was very full of new products and solutions for high-speed data transfer. The constant growth of traffic between servers (the so-called East-West traffic) began to necessitate the transition to higher speed standards. In particular, Selectel has already started working with equipment supporting 100 GbE links.



Special attention should be paid to the development of semiconductor technologies from Huawei, which is actively developing its own network processors for switches, gradually abandoning third-party products such as Broadcom.









During the event, Huawei introduced one of its latest products to its guests - the F1A-14H24Q ​​high-density router with a bandwidth of 2 Tbit / s , powered by Huawei Solar 5.0 chips. Half of the router ports support transceivers up to 1/10/25 GbE, and the second half supports 10/40/100 GbE.



Instead of a conclusion



After the session of reports, the guests went on to discussions with representatives of the vendors. First of all, a very important question was raised for many regarding how many products will be supported. It is no secret that the equipment itself continues to function properly for a long time even after the end of the EOL (End-of-life) period.



From the customer’s point of view, any bugs and vulnerabilities found can cause serious damage to the network, and if the vendor does not support the software part of old devices, then it will become simply unsafe to use them. So you should clearly understand what to count on after the onset of the EOLA (End-of-life announcement).



If you look from the position of the vendor, then the same situation appears on the other hand - the equipment may work properly, but become obsolete, as a result of which the customer will have to update it one way or another, so the EOL terms seem quite logical and correct.



In addition, the guests were interested in what kind of virtualization systems are used in the devices; some expressed their views on the quality of software for network devices.









The discussion turned out to be quite useful and informative. Representatives of the vendors were able to get feedback, and the guests received clearer information regarding the use of network equipment and technologies.



We thank all participants and guests of the event and will be glad to see you again!



Reports from previous 2019 events :





Articles on our blog on networks :







More articles:


All Articles