Hybrid Disaggregation Algorithm for Network Traffic
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing network traffic management systems face challenges in efficiently distributing network traffic across multiple server resources without causing disruption, especially when the number of server resources changes, as static and stateful algorithms either reallocate existing flows inefficiently or require excessive memory and processing time.
Innovation Solution
A hybrid Disaggregation Algorithm (DAG) that categorizes network traffic into subsets based on static packet fields, learns new flows during changes in server resources, and efficiently routes packets between primary and secondary resources using a consistent hash table and Bloom filters to minimize disruption and resource usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a stateful algorithm is used to distribute network traffic, then traffic disruption during server resource changes is reduced, but memory usage and processing time increase excessively
Solution Approach 1:
The patent segments network flows into two distinct subsets: long-lived flows and short-lived flows. This segmentation allows different distribution strategies to be applied to each subset. Long-lived flows use a stateful approach with flow tables to minimize disruption, while short-lived flows use a stateless hash-based approach to reduce memory usage and processing overhead.
Solution Approach 2:
The patent implements dynamic flow distribution by learning the characteristics of network flows over time. The system dynamically adjusts which flows are classified as long-lived versus short-lived based on observed behavior, allowing the distribution strategy to adapt to changing traffic patterns and server resource availability.
2Device complexity
If a static algorithm is used to distribute network traffic, then memory usage is reduced, but traffic disruption during server resource changes increases
Solution Approach 1:
The patent segments network flows into two distinct subsets: long-lived flows and short-lived flows. This segmentation allows different distribution strategies to be applied to each subset. Long-lived flows use a stateful approach with flow tables to minimize disruption, while short-lived flows use a stateless hash-based approach to reduce memory usage and processing overhead.
Solution Approach 2:
The patent changes the operational parameters of the load balancing algorithm based on flow characteristics. For long-lived flows, it uses connection-tracking parameters to maintain state information. For short-lived flows, it uses hash-based parameters that require no state storage. This parameter adaptation resolves the contradiction between memory efficiency and disruption reduction.
3Productivity
If the number of server resources is increased to handle more network flows, then system capacity and availability improve, but traffic disruption during resource changes increases
Solution Approach 1:
The patent performs preliminary classification of network flows into long-lived and short-lived subsets before server resource changes occur. This pre-classification allows the system to prepare appropriate distribution strategies in advance, minimizing disruption when servers are added or removed. The learning component continuously updates flow classifications to anticipate future traffic patterns.
Solution Approach 2:
The patent introduces an intermediary classification layer between the network traffic and server resources. This intermediary component learns flow characteristics and determines the appropriate subset classification, acting as a buffer that smooths out disruptions during server resource changes while maintaining high system capacity and availability.
Data Source
AI summary
Technology related to disaggregating network traffic is disclosed. In one example, a method can include determining whether individual network flows are members within a first subset of the network flows. A second subset of the first subset of network flows can be learned in response to determining a change in a number of servers available to service the network flows. A first network packet can be forwarded to a first server in response to the first network packet being a member of the first subset of network flows and a member of the learned second subset of the first subset of network flows. A second network packet can be forwarded to a second server in response to the second network packet being a member of the first subset of network flows but not a member of the learned second subset of the first subset of network flows.