Jitter-less FaaS via Flavor Clustering and SD-WAN
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Edge data centers face challenges in providing deterministic distributed execution of Functions as a Service (FaaS) due to dynamic composition of compute nodes with varying profiles and fluctuating network conditions, leading to inconsistent performance and jitter issues.
Innovation Solution
Implementing jitter-less distributed FaaS through flavor clustering, where FaaS functions are mapped to specific hardware resources with guaranteed Quality of Service (QoS) requirements, using a Jitter controller and Jitter-less Software Defined Wide Area Network (SD-WAN) to ensure predictable latency and throughput across multiple edge nodes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If FaaS functions are executed on dynamically composed edge data centers with multiple compute nodes having different profiles, then adaptability and resource utilization are improved, but execution consistency and jitter performance deteriorate
Solution Approach 1:
The patent segments compute nodes into distinct hardware flavor profiles (e.g., FPGA-based, x86-based, GPU-accelerated) with standardized performance characteristics. Each flavor type is further divided into virtualization instances with guaranteed resource allocations, creating homogeneous execution environments that maintain consistency while allowing diverse hardware types to coexist in the edge data center.
Solution Approach 2:
The patent applies local quality by assigning specific QoS parameters and performance guarantees to each hardware flavor type based on its inherent characteristics. Each flavor profile receives tailored resource allocation policies, scheduling parameters, and performance guarantees that match its capabilities, ensuring that execution consistency is maintained within each flavor while accommodating hardware diversity across the system.
2Adaptability or versatility
If edge data centers use networks with different latency, throughput properties and saturation levels, then network flexibility and coverage are improved, but deterministic execution and jitter performance deteriorate
Solution Approach 1:
The patent dynamically adjusts network transmission parameters such as packet size, transmission rate, and protocol selection based on the detected network conditions and the performance requirements of the FaaS function being executed. This allows the system to adapt to different network environments while maintaining deterministic execution by optimizing parameters for each specific network path and workload type.
Solution Approach 2:
The patent introduces a network abstraction layer that acts as an intermediary between FaaS functions and the underlying heterogeneous networks. This layer provides standardized network interfaces and implements traffic shaping, quality of service enforcement, and path selection mechanisms that hide network variability from applications, ensuring predictable latency and throughput regardless of the underlying network conditions.
3Productivity
If compute nodes have variable utilization characteristics over time, then resource efficiency is improved, but execution reliability and QoS guarantees deteriorate
Solution Approach 1:
The patent implements dynamic resource allocation and scheduling mechanisms that continuously monitor compute node utilization and adjust resource assignments in real-time. The system dynamically migrates FaaS functions between different hardware flavors and virtualization instances based on current load conditions, performance metrics, and QoS requirements, maintaining both high resource efficiency and reliable service guarantees through adaptive orchestration.
Solution Approach 2:
The patent employs feedback mechanisms where performance metrics, utilization data, and QoS measurements from compute nodes are continuously collected and fed back to the orchestration system. This feedback enables the system to identify underperforming or overloaded nodes and automatically rebalance workloads, scale resources, or migrate functions to maintain QoS guarantees while maximizing overall resource utilization across the edge data center.
Data Source
AI summary
Methods and apparatus for jitter-less distributed Function as a Service (FaaS) using flavor clustering. A set of FaaS functions clustered by flavor chaining is implemented to deploy one or more FaaS flavor clusters on one or more edge nodes, wherein each flavor is defined by a set of resource requirements mapped into a jitter Quality of Service (QoS) and is executed on at least one hardware computing component on the one or more edge nodes. One or more jitter controllers are implemented to control and monitor execution of FaaS functions in the one or more FaaS flavor clusters such that the functions are executed to meet jitter-less QoS requirements. Jitter controllers include platform jitter-less function controllers in edge nodes and a data center FaaS jitter-less controller. A jitter-less Software Defined Wide Area Network (SD-WAN) network controller is also provided to provide network resources used by FaaS flavor clusters and satisfy connectivity requirements between the edge nodes.