Often, when designing small data centers, designers are forced to choose network equipment and platforms with disproportionately high performance. Having a reserve for the future is, of course, great, but sometimes such a reserve is economically inexpedient. And then it is not at all involved.
Therefore, we made a network "factory" specifically for small data centers. Including those where the standard three-tier network infrastructure model is redundant.
A general look at CloudFabric
How does CloudFabric differ from CloudFabric Easy DC? First of all, the scale. But before we get into details, let's go over what CloudFabric is.
In short, this is a "factory" from Huawei . It was introduced to the market several years ago in response to market trends: CLI (command line interface) is gradually giving way to API (application programming interface); controllers appear that solve the problem of changing the configuration of active network equipment; the number of virtual entities (virtual machines, switches, routers, ITU) is multiplying; the number of streams increases, their width increases. As a result, traditional monitoring and control tools are no longer enough.
As the network facilities improved, the question of a full-fledged flow analysis (collection and evaluation of telemetry data) arose. This can be done using various network protocols, accumulating statistics on the operation of network equipment online. The captured information is packed, and this telemetry data is sent for processing to the collectors of the streaming telemetry analysis system. As a result, a universal platform emerges that allows us to solve three main tasks:
- fast automated troubleshooting;
- the ability to predict the state of services, including with certain configuration changes;
- ensuring information security due to the fact that the system is able to independently identify a number of threats that are not detected by modern intrusion detection tools.
Another notable trend is the proliferation of "factories" consisting of underlay networks (standard protocols and fixed configurations) and overlay (client service network). Finally, a modern network system must be open and integrated through the "north" and "south" interfaces.
At CloudFabric, we have implemented all of the above ideas and tools. Its functionality includes the deployment of underlay and overlay networks, resource orchestration, and organization of service chains. Equally important tasks of network diagnostics and service disaster recovery in minimal time (TTR) are solved using our Fabric Insight telemetry analysis system.
CloudFabric includes the iMaster NCE Fabric network control and management system, the configuration of which allows you to select the functions necessary for a specific project: SDN controller (AC-DCN), analytics tools (Fabric Insight), firewall policy management and configuration synchronization tools (SecoManager) , standard monitoring tools (eSight).
Now let's get back to the CloudFabric Easy DC platform. This is a somewhat lightweight version of CloudFabric, which, however, includes additional tools to quickly assemble the required specification.
Positioning CloudFabric Easy DC
As mentioned, the CloudFabric Easy DC platform can work with both underlay and overlay networks, as well as integrate with higher platforms. It supports micro-segmentation and Service Function Chaining, is equipped with logical network visualization and has the ability to add additional service devices (VAS). The Fabric Insight functionality is optional.
CloudFabric Easy DC has limitations on the number of active network devices and computing platforms. For example, Fabric Insight uses multiple collector systems, multiple analytic compute nodes, and a storage platform as standard. This is a minimum of five to six servers. With CloudFabric Easy DC, the Fabric Insight platform is represented by a single server. Hence the other limits:
- maximum 600 servers;
- virtual machines or containers no more than 24 thousand;
- no more than two spines (spine switch);
- no more than two border leaf switches, etc. (see picture above).
All of the above makes CloudFabric Easy DC the optimal choice for the rapid design of HLD solutions and the operation of small data centers (under a hundred servers).
Since in the solution under consideration, Fabric Insight has only one server, without separate collectors, the full functionality of flow analysis is not available, however, all other telemetry tools (CPU and memory usage, MAC / ARP / FIB tables loading, interface and queue statistics, statistics on packet loss and optical transceiver information) will be at your disposal.
Sample HLD Solution Based on CloudFabric Easy DC
Let's take a typical solution for a small data center that has:
- computing load (physical and virtual machines);
- network infrastructure;
- Internet and WAN connections.
CloudFabric Easy DC allows you to span multiple such datacenters - with an eye on the limitations mentioned in the previous section.
The above diagram shows the usual three-tier architecture: a server leaf associated with the computational load, a spine network core and a border leaf connected simultaneously to both external and service networks, and to neighboring data centers. The hardware component of the core is represented by the CloudEngine CE8850-64CQ-EI switch, at the access level - CloudEngine 6865-48S8CQ, but if an electrical interface is needed, it is advisable to use CloudEngine 6856-48T6Q-HI. Huawei positions such an infrastructure solution as a direct competitor to the Cisco mini APIC.
When 3-Tier Architecture Is Not Needed
With all the prevalence of the three-tier architecture, in some cases it is not optimal. You can start building a data center with just two switches that have all the necessary functionality, and then grow horizontally, implement spine and border leaf.
iMaster NCE as a standalone DCN analyzer
iMaster NCE bundled with Fabric Insight can serve as a standalone DCN analyzer and collect data across a switch-based network. Such a solution perfectly matches the concept of an easy solution, while it is based on our proprietary principle of dealing with network problems - 1-3-5: detecting a problem in a minute, finding out its cause in three minutes, restoring a service in five minutes.
Simple calculation of data center design
Gone are the days when you had to work in Excel, calculating transceivers and ports by the number of uplinks. Now all these and many other network building actions can be performed on the automated Huawei Smart configuration tool web platform.
At the first stage (Requirement design), a designer, filling out an online form, chooses a suitable design, determines the presence of optical or electrical ports, indicates the number of data centers.
Then you can select the iMaster NCE operating mode and the required x86-based server types. Among other things, servers can be provided by the customer or generally deployed in a virtualization environment. Both our Huawei FusionSphere and Microsoft and VMware solutions are available as management software.
At the Product Sourcing stage, the designer can select the types of power supplies, determine the number of leaf switches and transceivers in uplinks, indicate the need to deploy microsegmentation services and the Fabric Insight platform. Also, communication channels with neighboring data centers are separately configured. At the same time, the Smart configuration tool constantly checks the design for errors and, if any, reports them.
Upon completion of the configuration, the Smart configuration tool calculates the design and provides a set of HLD documentation available for download, ready to be sent to the client. The report includes the specifications of each device, information about the network topology, traffic patterns, configuration recommendations, costing and much more.
You can get a closer look at the capabilities of the Smart configuration tool and see the configurator in operation by watching our webinar or by directly accessing this tool on the Huawei website (if you have partner access rights).
CloudFabric Easy DC benefits
For comparison, let's take the previously mentioned Cisco mini APIC. In a standard architecture (two spines and two leaves), NCE functionality (without Fabric Insight) can be implemented in virtual machines. This, in comparison with the competitor's solution, gives a gain in rack space.
As the data center grows, it makes sense to use extended design and physical servers. In particular, to be able to organize the Multi-PoD design, when applied, the "factory" is able to operate on several physical data centers. Clustering physical servers allows you not to depend on the failure of one of the components, and also to use clusters "stretched" between two data centers (if latency allows).
High availability solutions based on CloudFabric Easy DC allow you to ensure system uptime in the event of failure of any of the three servers in the cluster.Finally, due to the technical characteristics of the switches included in the CloudFabric Easy DC, it is possible to assemble solutions with higher performance than those closest to them in the industry.
Learn more about switch models
The 2U core switch CE8850-64CQ-EI is built on the Broadcom chipset and supports telemetry capture via gRPC and ERSPAN, which can then be transmitted to Fabric Insight. By collecting TCP packet headers, the administrator can fully analyze any session. As for UDP traffic, it requires a dedicated AI Fabric chip to analyze it, which is also available in some of our solutions.
CloudEngine 6857 platform is equipped with 10 Gbps interfaces. The telemetry capabilities of the 6857 are exactly like those of the previous switch, all other main characteristics are marked on the diagram.
Finally, the CloudEngine 6863 switch available for ordering is presented with electrical interfaces.
We add that in the line of Huawei network equipment, the transition from third-party chipsets to its own, expanding its capabilities, continues. We are waiting for the arrival of switches that are built on our element base and completely cover all functional niches in the data center segment.
***
You can find more information about Huawei solutions and options for their application on our website - and of course, we will be glad to have your questions in the comments.