If we look at traditional data center architecture, data needs to move from storage devices to compute resources for processing and then transfer back. This constant shuttling of data in data centers takes a lot of energy. This not only goes into the data transfer itself but also into giving power to the networking equipment and cooling systems. These are required to support the movement of data transfers. 

This movement in a data center is said to consume over 40% of the data consumption. Experts expect this energy demand to rise with the increase in data volumes. In this article, we will learn about how computational storage in data centers helps you avoid this issue. So, let’s dive in.

The Computational Storage Approach

Computational storage systems incorporate processing capabilities directly inside the storage device. As a result, it allows data to be processed in situ, as opposed to sending it over the network for processing. So, this helps in addressing the issue of high energy consumption.

This method lessens the amount of data that must be moved as it only transfers the pertinent subset of data. This is needed for additional processing or analysis. Furthermore, computational storage in data centers reduces the amount of energy used by network equipment. It also reduces the energy consumption for storage interconnects, and related cooling systems. This is by minimizing the length and frequency of data transfers.

Key Components of Computational Storage in Data Centers

Now that we know what is the purpose of computational storage in data centers, let us see the Key elements of a typical computational storage system:

• Computational Storage Drives (CSDs)

These are storage devices, including hard disc drives (HDDs) or solid-state drives (SSDs). It incorporates computing power, frequently through the use of specialized processing units (SPUs). It is also through system-on-chips (SoCs), or field-programmable gate arrays (FPGAs).

• Computational Storage Processors (CSPs):

The processing components of CSDs that carry out computational operations on the data kept in the device are called CSPs. These might be GPUs, general-purpose CPUs, or workload-specific accelerators.

• Computational Storage Software Stack:

This comprises the apps and programming frameworks. These allow programmers to create and run programs that take advantage of the CSDs’ and CSPs’ computing power.

• Computational Storage Interconnects:

These are the fast connections that enable effective data transmission. This is between traditional storage devices, host systems, and networks, as well as computational storage devices.

Combining these elements allows computational storage solutions to transfer data-intensive tasks directly to the storage devices. As a result, it minimizes the need for data movement and lowers the data center’s overall energy consumption. Examples of these tasks include data filtering, compression, encryption, and even some AI and analytics workloads.

Energy Savings Potential of Computational Storage in Data Centers

Computational storage in data centers has a large potential for energy savings. The estimates range up to 90% for specific workloads. This is in terms of data center energy consumption reduction. Moreover, there are several ways by which this reduction is done:

Reduced Data Movement

Computational storage minimizes the energy used for data transfers across networks and interconnects. This is by processing data closer to its storage location.

Offloaded Compute Workloads

Using computational storage devices can decrease the energy consumption of CPUs and GPUs. These devices handle some of the computing tasks that these high-performance computing resources typically handle.

Reduced Cooling Requirements

Computational storage can lessen the energy demands of data center cooling systems by transferring compute workloads to storage devices. It in turn reduces the heat created by traditional computing resources.

Improved Storage Efficiency

Data deduplication, compression, and indexing may all be done directly on the storage devices using computational storage. It lowers the total storage footprint and related energy usage. It also enables more effective data storage and retrieval.

Optimized Data Pipelines

Computational storage can improve overall energy efficiency by streamlining data pipelines. Additionally, it can lower the number of intermediary data transfers and processing steps. This is by moving computation closer to the data source.

Computational Storage in Data Centers – Real-World Implementations and Use Cases

Although computational storage in data centers is still in its infancy as a technology, several businesses and academic organizations are actively investigating its possibilities. They are also putting trial programs into place to assess its advantages. So, here are a few noteworthy use cases and real-world implementations:

Samsung SmartSSD Computational Storage Drive

The SmartSSD is a computational storage drive designed by Samsung. It combines an FPGA and an ARM-based CPU in an SSD form factor. Moreover, tasks like data compression, encryption, and database operations may be offloaded to the SmartSSD. It minimizes data movement and lightens the stress on the host system.

NGD Systems Computational Storage Platform

NGD Systems provides a platform for computational storage. It combines software and APIs with its Newport computational storage devices. This is to offload operations that need a lot of data. It includes transcoding videos, big data analytics, and artificial intelligence. Moreover, their technology is made to work with cloud environments and sustainable data center infrastructures that are already in place.

Seagate Kinetic Open Storage Platform

Seagate’s Kinetic Open Storage Platform is a computational storage solution. It gets rid of traditional file systems and storage servers. This is by allowing direct access to and processing of data within storage devices. This strategy may drastically cut down on energy use and data transportation for applications. it includes object storage, data analytics, and also media processing.

Computational Storage in Data Centers – Challenges and Considerations

Even though computational storage in data centers has a lot of potential to increase data center energy efficiency, there are a few issues and things to keep in mind:

Hardware and Software Ecosystem Development

Establishing a strong network of hardware and software solutions for computational storage necessitates substantial investment. it also requires industry cooperation. Widespread acceptance will depend on standardization initiatives and compatibility across systems from various providers.

Programmability and Developer Adoption

Developing programming models, APIs, and toolchains that are easy to use is necessary. It enables developers to take full advantage of computational storage capabilities. So, to successfully implement computational storage, it will be crucial to guarantee a seamless developer experience and ecosystem support.

Data Management and Security Considerations

The distribution of processing among storage devices poses novel difficulties. These are related to privacy, security, and data management. So, it will be necessary to put in place strong procedures for safe execution environments, access control, and data protection.

Integration with Existing Data Center Infrastructure

For sustainable data center operators to move smoothly and with the least amount of disturbance possible, computational storage solutions must be seamlessly integrated with the current data center infrastructures. It includes networking, storage, and management frameworks.

Cost and Performance Trade-offs

Although computational storage in data centers has the potential to save a large amount of energy, its implementation may come with additional expenses and performance trade-offs. Careful assessment and optimization will be necessary to guarantee that the increases in energy efficiency outweigh any possible negative effects on performance or additional expenses.

Computational Storage in Data Centers – Future Outlook and Opportunities

With the increasing need for data-intensive applications, computational storage in data centers is expected to become increasingly important. This is in facilitating data center operations that are more sustainable and energy-efficient. The following are some significant prospects and upcoming changes to be aware of:

Advancements in Computational Storage Hardware

Researchers anticipate that they will further develop computational storage hardware. This is by utilizing advancements in fields such as FPGA technology, specialized accelerators, and low-power system-on-chip designs. It will result in more potent and energy-efficient hardware.

Intelligent Data Management and Analytics

By outsourcing data pretreatment, filtering, and transformation duties straight to storage devices, computational storage systems can allow intelligent data management and analytics capabilities. It also minimizes the need for expensive data transfers and centralized processing.

Edge Computing and IoT Integration

Reducing the requirement for data transmission to centralized data centers and increasing overall energy efficiency may be achieved by integrating computational storage with edge computing devices and Internet of Things systems. It enables more effective data processing and analysis at the edge.

Sustainable Data Center Designs

Anticipators expect that the development of computational storage will be crucial for creating and using greener technology. It is also crucial for more energy-efficient data center designs. Furthermore, it opens up new possibilities for power distribution, cooling, and data management.

Collaboration and Ecosystem Development

The broad adoption and standardization of computational storage systems will need sustained cooperation. This is between hardware manufacturers, software developers, data center operators, and academic institutions. Furthermore, it will create an ecosystem of compatible and interoperable solutions.

To Sum Up

Sustainable data center design is being revolutionized by computational storage. It gives a viable answer to the rising energy costs related to data mobility. Computational storage in data centers has the potential to dramatically lower energy usage by putting computing closer to the data storage location. As a result, it allows for more sustainable and effective data center operations.

The data center sector can fully realize the promise of computational storage through ongoing research, cooperation, and practical applications. The premier event devoted to investigating the newest technologies, approaches, and best practices for optimizing operational efficiencies and sustainability while lowering energy costs and emissions in data centers is the Energy Efficiency for Data Centers Summit, which will be held May 16–17, 2024, in Dallas TX, USA.  Experts in the field, scholars, and IT executives who are at the forefront of sustainability or energy efficiency in data centers will be giving sessions, case studies, etc to make the attendees one step ahead in the revolution. Register today!