Data storage acceleration-Gigaom


Flash memory is now the standard for storing active data in data centers. NVMe and fiber-based NVMe (NVMe-oF) are on the rise and ensure that storage is no longer a bottleneck in the infrastructure. On paper, the CPU can now access data as quickly as possible, and can make full use of the network, which opens the door to better performance and faster results. In fact, everything is fast, but the efficiency is also very low. Suppliers cannot solve larger problems and solve their problems, which leads to increased complexity and cost.

In the past few months, I have written articles about hardware acceleration and resources, such as GPU, FPGA, smart NICS, DPU (Network Data Processing Unit) and so on. The focus of these solutions is to improve overall system efficiency by reducing the workload of required tasks that consume CPU cycles and hinder application performance. This dedicated hardware can severely affect several operations.

Two ways to speed up storage

The goal is always to achieve greater success and move forward faster. It is related to scale and is becoming more and more important because many applications need to access large amounts of data. Faster is usually related to latency, and less is better.

Optimizing access to data can improve the efficiency of the entire stack, so that you can do more with less computing resources. These advantages will affect all the main indicators of modern infrastructure design: power efficiency, computing density, physical footprint, etc. In fact, for most users, this is not about optimal performance, but about a balanced architecture and a better total cost of ownership.

In my previous article, I did not mention several interesting and innovative storage acceleration methods: computational storage and storage processors.

Computational storage is a simple but powerful concept. Instead of bringing data into the CPU, bring the CPU into the data. In fact, this involves a flash memory device-SSD-onboard RAM and CPU. The device can be programmed to act as a standard SSD, but the data dropped in the device can be processed on-site. This allows applications to transfer many simple operations directly to the storage hardware-no need to move data for analysis, and each new flash memory device added to the server adds more computing resources that can work in parallel.

There are several applications that can benefit from this approach, such as image recognition or video analysis. You only need to write the data to the flash memory device once, and it will automatically index and sort it based on the content without touching the server’s CPU. To learn about the technology I’m talking about, visit the websites of companies such as NGD Systems.

Another method is to use a specially designed accelerator to reduce the burden on the CPU. In this case, storage accelerator vendors provide APIs or device drivers to replace operations that are usually done using standard libraries or other software components. The speed of performing storage operations on dedicated hardware is much faster than performing storage operations on a general-purpose CPU.

The advantage of this approach is that you don’t need to adjust the application to support the model, but only need to make minor changes to the application stack. A common example is to replace part of the database structure to speed up indexing, querying, data storage and information retrieval. A commercial example of this product is Pliops. Here is a video of a recent storage site day event that describes the potential of this solution:

Closed circle

Data center acceleration is a hot topic, and for good reason, because it can improve infrastructure efficiency and total cost of ownership. Depending on the use case and the size of the infrastructure, there are several methods available. The two examples I introduced in this article are just the first examples I think of. What is really interesting is that no matter the size of the organization, everyone faces many of the same challenges, including rapid data growth and the need to support new data-intensive applications.

Accelerators are rapidly entering every data center, even smaller data centers, and the solution has been validated by major vendors such as VMware. Every IT organization should seriously consider this new area and plan accordingly to keep its infrastructure as efficient and responsive as possible.

Related Articles

Back to top button