The development of data centers: technological trends

The development of data centers: technological trends
Rate this post
facebook twitter pinterest linkedin

In today’s material, we will talk about new technologies in the field of virtualization, software-defined networks and machine learning.

GPU Virtualization

Manufacturers of data center equipment offer all new solutions in this area. At a recent VMware conference, they talked about the modified ESXi hypervisor. It improves the performance of virtual GPUs to the level of bare metal graphics processors (the difference is only 3-4%).

The technology will find application in high-performance computing (HPC)  —the hypervisor will be used to train neural networks, render 3D graphics, and build complex mathematical models.

Accelerations were achieved thanks to two changes: they removed vMotion and added DirectPath I / O. The first technology balanced the load between the GPU. Now each virtual machine has its own graphics processor. This approach has reduced data sharing costs.

The second technology allowed the CUDA parallel computing driver to “communicate” with virtual machines directly, bypassing the hypervisor. As a result, the overhead of data transfer has been reduced.

VMware is not the only one who develops virtualization systems for graphics accelerators. Companies such as AMD and Intel are working in this area. The first is developing a technology that will allow dividing the graphics processor between sixteen users. In this case, the performance for all will be the same.

See also  Mastering the Science of Thread Likes: Top 10 Techniques

As for the second company, their technology combines the work of a standard GPU driver and a virtual machine that can display 3D applications and desktops on devices of hundreds of users.

Software Defined Networks

Today, more and more companies offer new solutions for software-defined infrastructure. This technology was presented by VMware.

The company has developed a firewall that monitors the operation of applications on the network. It consists of two components: the AppDefense threat detection system and the NSX platform. The first is responsible for building a behavioral model of running services (based on the telemetry provided by the company’s customers), and the second is managing cloud security policies.

Software-defined networks are being developed by other companies: Cisco and Juniper have SDN solutions. Their platforms simplify the monitoring and management of distributed IT infrastructure. For example, with their help, you can change the access policies for virtual machines in remote data centers.

Another organization working in this area is Nokia. At the end of last year, she presented a software-defined platform based on open technology OpenStack Heat. It will help telecommunications companies and mobile operators to deploy infrastructure for 5G networks.

Machine learning technology

In three years, half of the data centers will use smart algorithms to manage the infrastructure. For this reason, increased activity is observed more and more in this segment.

See also  7 Common Photo Editing Mistakes and How to Avoid Them

In March of this year, Juniper acquired the startup Mist. He is developing a platform for managing Wi-Fi networks and analyzing the status of devices in a wireless network. A special feature of the platform is Marvis chatbot with speech recognition technology. For example, he is able to respond to a voice query about how many routers are currently running and which ones are heavily loaded.

Similar solutions are created in other IT companies. Cisco is developing intent-based networking. It will help set the performance of Wi-Fi (for example, connection speed) and automatically configure the device.

Similar technologies work in other areas as well. For example, in the field of energy management and cooling systems.

So, at the beginning of the year, start-up Carbon Relay launched a  platform for monitoring the PUE ratio (determines energy efficiency) in data centers. The solution analyzes the temperature and the movement of air flow in the engine room using a data center model in telus store. Then, it predicts PUE and temperature indicators and gives recommendations on how to adjust the air conditioning system. A similar solution is developed on Google.


All these technologies will become more popular. And over time, they will begin to “intertwine” more strongly. For example, a new software-defined firewall from VMware already uses machine learning to look for abnormal activity on the network. Such tools will change the principles of data centers and improve the security of their IT infrastructure.

read also:


    Leave a Reply

    Your email address will not be published.