Modern lifestyles are increasingly reliant on cloud computing and storage infrastructure. Whether at home, at work, or with the smartphones and other mobile computing devices we carry with us, the need for cloud computing and storage is everywhere. Especially with the rapid development of big data and the Internet of Things, the demand for this type of infrastructure is growing at an alarming rate. As the number of apps and users increases, the annual growth rate is about 30 times per year, and in some cases as high as 100 times per year. Such high growth rates make it difficult for Moore’s Law and new chip development to meet the demands of computing and networking infrastructure. To this end, many data and communications network vendors have solved this problem by investing in a variety of parallel computing and storage methods. In this way, many cloud data centers today generally form hundreds or even thousands of servers connected together.
As more and more switches are needed, the interconnection of all these computing and storage devices is becoming a real challenge. In the data center, the traditional approach to networking is layered, with individual racks using leaf switches (also known as rack switches, or ToR switches) for intra-rack connectivity, while spine switches are used to connect a series of racks and core switches for the entire data center. And, like the servers and storage themselves, these switches need to be managed. In the past few years, there have typically been one or two vendors that provide network switches and related management control software for data centers, but things are changing quickly. Many leading cloud service providers, with their strong purchasing power and technical capabilities, recognize that they can save a lot by designing and building their own network equipment. In the data center industry, many see this as the first step in disrupting legacy networking hardware and its control management software. Cloud providers have their own software engineers, so they develop and manage the software in-house and outsource the hardware design. While this satisfies the business needs of data center operators to some extent, from a technical and operational management perspective, there is no simplification, and a large number of switches need to be managed.
In 2009, with the introduction of what we now know as Port Extender, the first breakthrough was achieved in simplifying network complexity. This concept is based on the idea that many nodes in the network do not need the management capabilities that most switches have. This is similar to the mother-child relationship, where the control switch is the “mother”, the managed switch is the “child”, and the port extender provides the connection. This port extender method was included in the networking standard 802.1BR in 2012, and every network switch currently developed complies with this standard. The technology of port extenders is not complicated, and the most obvious advantage over full-bridge switches, in addition to lower power consumption, is the reduced unit cost.
Controlling bridging and port extenders helps simplify the management of network switches, but it doesn’t stop there. Dig deeper into the port extender and you’ll see that the same switch chip is used as the “mother” bridge switch. Instead of chips specifically designed for port extenders, switch vendors continue to use their standard chipsets, thus not adequately reducing costs and power consumption. However, with the introduction of Marvell’s 802.1BR-compatible port extender IC, known as PIPE – Passive Intelligent Port Extender, a true modular approach to network switching has taken a leap forward, enabling interoperability with all industry-leading OEM control bridge switches. It also offers high cost and power advantages, which is why port extender technology is immediately noticeable. In the second step of network decomposition, this approach effectively separates port connections from the processing power of the parent switch, establishing a more modular approach to networking. The parent switch no longer needs to know what type of device it is connected to, so all the logic and processing functions can be concentrated on the parent switch, with the port extender taking on the port I/O functions.
Marvell’s Prestera®PIPE series targets data centers running at 10GbE and 25GbE speeds that face the challenge of further reducing CAPEX and OPEX costs as bandwidth requirements increase. The Prestera PIPE series enables top-of-rack switch deployments at half the cost and power of traditional Ethernet switches. The PIPE approach also includes fast failover and resilient recovery capabilities, which are critical for continuity and high availability of critical infrastructure.
Written by George Hervey, Chief Architect of Marvell