Understand Processing Gravity To Optimize Your AI Architecture Design
The era of AI is coming, and AI applications are sprouting everywhere — from AI applications that are already widely embedded in enterprise business scenarios to massive models containing trillions of parameters that are ready to solve complex problems. They are also crossing into the physical world, from data centers to factory floors, customer touchpoints, and wild environment.
The increasingly pervasive nature of AI applications poses a challenge for technology leaders when it comes to designing their infrastructure architecture. How should they choose the proper processing location among public clouds, on-premises data centers, edges, or endpoint devices? How should they maximize output for a given budget? To simplify design decisions, Forrester proposes the concept of processing gravity. In the physical world, the impact of gravity on an object depends on two factors: mass and distance. Similarly, when measuring processing gravity, the two factors are the power of the processing source and the latency between the application and processing source:
- Processing sources take a variety of forms. Heterogeneous processing sources are constructing the computing equivalent of a celestial constellation: Public clouds and large-scale data centers are like stars with tremendous mass, while on-premises data centers and edge computing are like planets and satellites. The mass of each will have an impact on your AI architecture choices.
- Connection, execution, and privacy create a compound processing latency. Transferring data/results to/from the processing sources and executing the task take some time. In addition, privacy regulations can exclude some processing sources for specific applications, which means the latency is infinite for these use cases.
Guided by processing gravity, you can then choose the most suitable processing source for a specific application. Moreover, to maximize the aggregated gravity, you can modernize your application architecture to leverage more processing sources, and you can also reduce the compound processing gravity with an advanced enterprise network, specialized compute infrastructure, and Zero Trust model. To learn more about processing gravity, read Optimize Your Artificial Intelligence Infrastructure With Processing Gravity, or set up some time for a discussion with me.