Networking 101: NVMe over TCP

The volume of data the world processes these days is staggering. It is estimated that both humans and machines, from gaming PCs to weather balloons and Mars rovers, produce about 2.5 quintillion bytes daily, and growing.  

In direct correlation, data centers are also expanding as the demand for computing power increases. According to Research and Markets, the market for these spaces will achieve a compound annual growth rate (CAGR) of 2 percent from 2019 to 2025. In the United States alone, several colocation service providers are investing around $200 million in building and upgrading data centers.

To ensure that data centers can process huge amounts of data, they need the right combination of a nimble storage infrastructure and a hyperscale-capable protocol, such as NVMe over TCP.

What Is NVMe?

 Some people often confuse NVMe with M.2 drives since these terms are closely linked, but they are completely different. An M.2 SSD refers to the form factor, while NVMe is the protocol, which means it contains the instructions on how the drive can store and deliver information.

 NVMe stands for “non-volatile memory express”, and it offers the following features:

• As a non-volatile memory, it can store and retain data even when the power is removed. Thus, it works the same way as many standard storage devices like flash and read-only memory (ROM).
• It is also optimized for non-uniform memory access (NUMA). It is a type of server architecture wherein multiple microprocessors create a single system, and they can share data locally. This way, the cost-to-performance ratio increases, and the system itself can be expanded by simply adding more microprocessors into the cluster.

One can also integrate NVMe over networks, giving rise to extension protocols like NVMe-oF (over fabric). This suggests that servers can already share storage by implementing the concept of NVMe over InfiniBand, Ethernet, and fiber channels.

 The issue with this setup, however, is that these options still need custom host drivers and bus adapters. It can take so much money to implement and still does not maximize the power of scalability.

 One of the viable solutions is NVMe over TCP, where TCP stands for transmission control protocol. It helps remove storage bottlenecks of the standard NVMe-oF by now allowing the same storage to be shared among data centers through the Internet protocol. It does not require any physical modifications to these servers or to the storage devices.

Also read: New Momentum for IPv6 Deployment

Benefits of NVMe over TCP

NVME over TCPs are beginning  is growing in popularity among PC builders, gamers, and data centers for the following reasons:

• When compared to SATA, they win on many fronts. NVMes perform way better in throughput, latency, queues, and commands per queue. The typical throughput of SATA is only 6 Gbps, but it climbs to between 16 and 32, depending on the actual NVMe used. Latency significantly drops, while it can process over 65,000 queues at any given time. Lastly, it can process over 60,000 commands per queue.
• Lower latency means data is available fast. The ability to access any data is just as important to a server’s capability to store it. This makes NVMe high-performing.
• Prices are going down. Today, you can buy an NVMe drive for $150 or less.
• They can make the microprocessor system more scalable and easy to deploy. This way, data centers are less likely to spend more money on investing in more storage devices if they need to process larger volumes of information in the future.

The future of computing is big data. Users are only going to shop more, launch more types of tech that gather information, and implement the Internet of things (IoT) devices. However, along with their growth, is the need for proper support for data storage. Data centers require one that is easy to implement, high-performing, and scalable without breaking the bank. So far, NVMe, particularly NVMe-TCP, fits the bill.

Read next: Also read: Networking 101: What is NVMe over Fabric?