Advances, Systems and Applications
From: CLQLMRS: improving cache locality in MapReduce job scheduling using Q-learning
Scheduling method | Locality level | Technique | Advantages | Disadvantages |
---|---|---|---|---|
HybSMRP | DL (Map tasks) | Localization marker, Job priority | Avoids wastage of resources | Does not consider some environmental features for job priority |
LARD | CL | Predict cached data location | It is suitable for small files | Poor performance for large dataset |
Improved CL, and DL for map tasks | DL, CL (Map tasks) | Weighted bipartite graph, Maximum matching algorithm | Improves data locality rate for Map tasks | Only considers Map tasks |
ACL | DL, CL | Delay tasks to launch them on local nodes | Increase tasks locality rate | Latency to schedule tasks |
CATS | CL | Buffer cache probe and find tasks with the greatest amount of cached data | Increases cache locality rate | Only considers cache locality |
CAVA | VMM, PMML, RMML, VMDL | Memory locality-aware and application’s cache affinity | It is suitable for virtualized clusters | Only focuses on Map tasks |