Centralized Scheduling for Non-Uniform Data Center Traffic
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Data center networks face challenges in efficiently scheduling non-uniform traffic, leading to traffic congestion and significant delay differences between hotspot and non-hotspot areas due to the limitations of existing scheduling algorithms like iSLIP and DRRM, which are vulnerable to hotspot conditions and do not effectively consider the bursty and data-intensive traffic characteristics.
Innovation Solution
A centralized scheduling method and apparatus that uses a cyclic scheduling template with a scaling factor to improve performance under hotspot conditions, employing a round-robin-pointer (RRP) sequence matrix and transportation-request (TR) Boolean matrix to uniformly distribute pointers for input and output arbiters, ensuring efficient matching and minimizing delay differences through real-time and off-line algorithms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a typical maximum matching scheduling algorithm (iSLIP or DRRM) is used, then 100% performance is guaranteed for uniform traffic, but the algorithm becomes vulnerable to hotspot conditions and cannot effectively handle non-uniform traffic distribution
Solution Approach 1:
The patent applies local quality by differentiating the scheduling approach based on traffic characteristics. For uniform traffic, the standard iSLIP/DRRM algorithm is used, while for non-uniform traffic with hotspot conditions, a modified algorithm that considers traffic load and uses scaled cyclic periods is applied. This allows the system to optimize for local traffic patterns rather than using a one-size-fits-all approach.
Solution Approach 2:
The patent introduces dynamics by making the scheduling algorithm adaptive to changing traffic conditions. The centralized scheduler monitors traffic distribution and dynamically adjusts the cyclic period scaling factor based on detected hotspot conditions. This transforms the static scheduling algorithm into a dynamic one that can respond to and adapt to non-uniform traffic patterns.
2Device complexity
If a centralized scheduler performs N-to-N input-output matching within each scheduling time using standard algorithms, then scheduling simplicity is maintained, but traffic congestion occurs even when link efficiency is about 25% due to non-uniform traffic characteristics
Solution Approach 1:
The patent applies preliminary action by having the centralized scheduler perform traffic monitoring and analysis before executing the scheduling decision. The scheduler detects traffic distribution patterns and determines appropriate cyclic period scaling factors in advance, allowing it to proactively prevent congestion rather than reactively responding to it. This preliminary assessment enables more efficient scheduling decisions.
3Device complexity
If standard scheduling algorithms are used without considering traffic load, then algorithm simplicity is maintained, but average delay difference between hotspot traffic and non-hotspot traffic becomes large
Solution Approach 1:
The patent applies parameter changes by modifying the cyclic period parameter based on traffic conditions. When hotspot conditions are detected, the scheduler scales the cyclic period for affected input-output pairs, effectively changing the scheduling timing parameter to reduce delay. This allows the system to maintain algorithmic simplicity while adjusting critical parameters to address delay issues.
Data Source
AI summary
The present disclosure relates to a centralized scheduling method and apparatus that considers non-uniform traffic and, more particularly, to a centralized scheduling method and apparatus for performing effective scheduling based on a characteristic of non-uniform traffic in consideration of a traffic distribution in a data center network.