Modernize with DPU-enabled switches
Traditionally, CPUs alone provided the processing power for hyperscale and
enterprise data centers. More recently, graphics processing units (GPUs), have
taken a significant role. Originally used to deliver rich, real-time graphics, their
parallel processing capabilities make them ideal for accelerated computing
tasks, including artificial intelligence, deep learning, and Big Data analytics
applications.
A new type of processor has emerged on the scene called a data processing
unit (DPU), quickly becoming an important part of the data–centric accelerated
computing family. DPUs are purpose-built silicon used to offload data traffic so
compute-intensive tasks can be optimized on CPU and GPU resources.
Using its own hardware-based processor, DPUs are often deployed in
hyperscale data center servers and perform a large set of accelerated
compute, cloud, network, security, and storage functions including encryption,
firewall, load balancing, NAT, telemetry and more. These capabilities enable
the isolated, bare-metal, cloud-native computing platforms that define the
next generation of cloud-scale computing.
DPU technology has evolved from being server only to now being available
in top-of-rack (ToR) switches. This new category of distributed services
switch marries standards-based Ethernet/IP based switching with embedded,
hardware accelerated, fully programmable DPU technology in a unified, high
performance secure networking solution for the data center and cloud.
This allows operators to extend industry-standard leaf-spine networking with stateful distributed microsegmentation,
east–west firewalling, NAT, encryption, and telemetry services—delivered closer to where critical compute and storage
workloads are being processed at the compute-network edge.
Unlike dedicated SmartNICs, which are installed in servers, a distributed services switch can be deployed at the server
ToR and provides distributed services to all servers and hosts in the rack.
Distributed services switches do not require changes to server hardware or software or make assumptions on server
operating systems, require any driver or agent to be installed on servers to provide distributed services at scale and
wire-rate, and can be integrated into new or existing brownfield enterprise data centers and private clouds.
HPE Aruba Networking and AMD Pensando™ have enabled the industry’s
first distributed services architecture that allows enterprises to create
and operate network infrastructures that perform and scale just like the
hyperscale infrastructure giants.”