Moving to Data Centric Data Centers and DPUs

By Hubert Yoshida posted 10-06-2020 20:00

Like

The Compute Centric data center is running out of gas. The CPU was the center of the data center from the very beginning. Besides running the operating system and the applications, the CPU managed the memory, the storage, and the network. Thanks to Moore’s Law and the advances in technology by Intel, AMD, and others the CPU was able to support this load for a long time. However, the advances in Moore’s law has slowed and the CPUs are pushed to the limit with virtualization and containers. On top of that the storage and network is getting much faster and more demanding with NVMe and 5G, and the demand for security, compliance and AI have changed the way programs are developed and processed. Also managing the sheer volume of data and new data sources is consuming more of the CPU’s cycles. As the data center becomes more data centric, the management of storage, network, and data security needs to be offloaded from the CPUs, so that they can concentrate on managing virtualization, containers, AI and application processing.

The move to offload data management functions has already begun with the introduction of Smart NICs and FPGAs. Hitachi’s recent announcement of new enhancements and capabilities to its hyperconverged infrastructure (HCI) portfolio with updates to Hitachi Unified Compute Platform (UCP) HC and Hitachi UCP RS provide added benefits to customers including faster provisioning with new Hitachi UCP Advisor, certified support for SAP HANA workloads, new Intel Cascade Lake Xenon Refresh processors that increase performance, and enhanced lifecycle management capabilities that deliver non-disruptive upgrades. In the cloud, AWS Nitro is an example of how EC2 instances can break apart storage, networking and management functions and offload them to dedicated hardware and software to enable AWS to innovate faster with increased security and new instance types.

NVIDIA has given a name to this new processing function. Just as they have developed the concept of the GPU (Graphics Processing Unit) for Increased parallel performance, they are now promoting the concept of the DPU or Data Processing Unit. According to NVIDIA, “The CPU is for general purpose computing, the GPU is for accelerated computing and the DPU, which moves data around the data center, does data processing.”

Here is how a DPU would fit into a data centric data center.

I like the concept of a DPU and would encourage the use of that term. NVDIA would like to tie DPU to their product, the BlueField-2 DPU.

I would apply the concept of DPU to our Lumada platform since it off loads the integration and management of IoT data and makes the processing of IoT more efficient. IoT is data centric.

Expect to hear more about DPUs as this concept develops and vendors build out the ecosysytem.

#Blog
#Hu'sPlace

0 comments

1 view

Blogs