4 Annual Report 2022
The Napatech Opportunity
THE NAPATECH OPPORTUNITY
Modern Datacenters Supercharged by Programmable SmartNICs
cally integrated systems. A new method of software-defined net-
A TIDAL WAVE OF DATA HAS ARRIVED
working exists where network functions are being disaggregated
from the hardware, virtualized, and delivered as software in-
stances on open, standard, low-cost computing platforms – serv-
ers. This results in a world where best-in-breed solutions can in-
terchangeably come together across servers, processors, accel-
erators, applications, and operating systems, in a low-cost man-
ner.
We have entered an era of massive, distributed, digital, data that
has changed nearly every aspect of communications and deci-
sion making. At the forefront of this surge is an enormous set of
newly connected devices, that includes hundreds of millions of
personal computers, billions of smartphones and mobile de-
vices, and trillions of new connected devices such as home auto-
mation, autonomous vehicles, smart cities, factory automation
and so much more.
The new datacenter is cloud-native, enabling increased levels of
flexibility and scalability. It is more intelligent, powered by ma-
chine learning and artificial intelligence. It has advanced security
by distributing microservices into every network element and lo-
cation that needs them. Cloud, edge, mobile, and enterprise
multi-cloud networks work in harmony to create, process,
transport, analyze, secure, store, archive, and share a massive
amount of data.
Along with these connected devices comes an even larger explo-
sion in applications, services, and use-cases for these items.
These devices, applications, and services combined to create an
estimated 75 zettabytes of data in 2021. The increase from those
items is forecasted to more than double to 175 zettabytes in
2025, and exceed 600 zettabytes by 2030.
SMARTNICS PLAY A CENTRAL ROLE
To begin to comprehend the enormity of this scale, one zetta-
byte has twenty-one zeros after the one. A single zettabyte con-
tains enough high-definition video to play continuously for
36,000 years. To realize how fast things are moving, more than
90 percent of the world's digital data has been created in the
past two years alone. The information created from sources such
as video, audio, sensors, and other sources, will be compounded
further as the raw data is further processed and analyzed with
emerging machine learning and artificial intelligence algorithms.
A central part of the modern datacenter architecture is the de-
sign principle of heterogeneous computing. Legacy architectures
viewed the central processing unit (CPU) as the primary device
in datacenter computing. Now it is just one of many processing
elements that make up the modern datacenter, which includes
devices that are purpose-built for the specific workloads they are
optimized to handle. This results in a new view that is no longer
accelerating the CPU, but a model of coprocessors spread
throughout the network.
LEGACY DATACENTERS ARE IN DECLINE
It is here where the programmable SmartNIC plays a central role
in the datacenter. These programmable, data processing units
(DPUs) or infrastructure processing units (IPUs) now become the
fabric or spine of the network, distributing workloads across vir-
tual machines, containers, network slices, storage elements, ma-
chine learning, artificial intelligence engines, and applications
and services running on CPUs. This architecture allows the CPUs
to deliver the applications and services they were intended to,
and leave the networking, storage, security, and analytics to de-
vices better suited for those tasks.
Consequently, we are generating more information, from more
sources, than the existing IT resources can create, process, de-
liver, secure, analyze and store in the timeframes needed. When
faced with shifts in scale in the past, organizations defaulted to
methods such as cyclical server CPU and network infrastructure
upgrades. Today several factors have conspired to diminish the
impact provided by those actions. Most significant is the rapid
decline of Moore’s Law, which has reduced the performance and
cost benefits that are realized in each upgrade cycle. Further, the
workloads required of those CPUs are increasing exponentially,
due to additional algorithms being applied actively in-flight
across the network including analytics powered by machine
learning and artificial intelligence; signature and pattern match-
ing in response to a constantly evolving threat landscape with
sophisticated actors changing their posture frequently, and net-
work infrastructure I/O increasing to keep pace – all faster than
the networks and the hardware that underpins them can evolve
or be upgraded.
DEPLOYED IN NETWORKS OF EVERY TYPE
This approach to datacenter design originated in the hyperscale
cloud network operators and proved to deliver enormous bene-
fits in terms of cost, power, performance, utilization, and sus-
tainability. Although in its infancy today, these same designs are
in high demand by network builders everywhere. Engineers that
build appliances and servers for original equipment manufactur-
ers (OEMs) are using programmable SmartNICs to power their
next generations designs. End users who build their own systems
are following suit. Tier-2 cloud operators are following in the
footprints of their larger tier-1 peers to host applications and
centrally store data. Communications service providers (CSPs) in
mobile, telco, and cable networks are providing secure, high-
bandwidth access to millions of simultaneous users. Enterprises
MODERN ARCHITECTURES HAVE EMERGED
To overcome these challenges a new network architecture for
modern datacenters has emerged. The networks are transition-
ing away from expensive, large, proprietary, monolithic, verti-