Serverless DAG Workflow Execution Plan Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Serverless workflows in cloud environments face significant latency and cost issues due to communication latency between in-series workers and computation skew among parallel workers, primarily caused by executing each function in a separate isolated container or VM, leading to inefficient resource utilization.
Innovation Solution
An automated system generates an optimized execution plan for serverless Directed Acyclic Graphs (DAGs) by combining in-series functions, bundling parallel invocations, and allocating optimal VM resources, using a distribution and correlation-aware performance model to mitigate communication latency and computation skew, thereby reducing end-to-end latency and cost.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If each function is executed in a separate isolated container or VM, then resource isolation and security are improved, but communication latency between in-series workers increases and resource utilization efficiency deteriorates
Solution Approach 1:
The patent merges in-series functions into the same container or VM, eliminating inter-container communication overhead while maintaining resource isolation through virtualization. This combining approach reduces communication latency between sequential functions without completely sacrificing security, as the merged functions still operate in an isolated virtualized environment.
2Reliability
If each function is executed in a separate isolated container or VM, then resource isolation is improved, but resource utilization efficiency deteriorates
Solution Approach 1:
The patent makes containers or VMs universal by allowing them to host multiple functions simultaneously. A single container can execute multiple in-series functions or multiple parallel invocations, making the resource allocation more flexible and efficient. This multi-functionality approach improves resource utilization while maintaining adequate isolation through proper resource management.
3Ease of operation
If traditional VM size allocation methods are used, then simplicity of resource management is improved, but end-to-end latency and cost increase
Solution Approach 1:
The patent implements dynamic VM sizing that automatically adjusts container or VM resource allocation based on the specific requirements of each function or function bundle. Instead of using fixed, traditional VM sizes, the system dynamically optimizes resource allocation to minimize latency while managing complexity through automated orchestration and performance modeling.
4Ease of operation
If traditional VM size allocation methods are used, then simplicity of resource management is improved, but cost increases
Solution Approach 1:
The patent changes the parameters of resource allocation by using performance models to determine optimal container or VM sizes based on actual workload characteristics. This data-driven approach adjusts resource parameters dynamically, reducing waste and lowering costs while maintaining simplicity through automated model-based decision-making rather than manual configuration.
Data Source
AI summary
A system may receive a first directed acyclic graphic (DAG) for an application. The system may model performance of each function in the DAG to generate a performance model. The system may generate a plurality of variant DAGs. For each of the variant DAGs, the system may obtain a configuration vector and forecast, based on the performance model and the configuration vector, a plurality of end-to-end latency distributions for the variant DAGS. The system may select the variant DAG and configuration vector based on a selection criteria. The system may cause an application to be executed according to the variant DAG and configuration vector.