1658 - 从瓶颈到突破；Storage Protect 中备份调度的多线程方法.pdf

1、Orlando,FLOctober 69IBM TechXchange 2025A Multithreaded approach to backup scheduling with Storage ProtectIntroductions and AgendaSession GoalPresent the architecture,implementation,and performance impact of MDSMProject ContextDeveloped for a non-standard backup use case requiring improved throughpu

2、t and logging across diverse filesystemsPerformance HighlightAchieved significant performance improvement over traditional scheduler using controlled,multi-worker execution2https:/ was this needed?Constraints in high volume environmentsArchitectural OverviewWorker Strategy&MAXPROC designBackup Modes

3、 in MDSMProcess Management&Lifecycle ControlError Handling&LoggingBenchmarking MethodologiesTrade-offs and Lessons LearnedOperational OutcomesFuture ImprovementsFinal RemarksQ&AFilesystem Layout4High Performance+LargeFS mode5Why was this needed?6Identifying BottlenecksGrowing Data VolumeOvernight ba

4、ckups struggled under high data churn&multi-terabyte filesystems7Inefficient Resource UseBackup jobs were serialized,leaving CPU,I/O,and storage bandwidth underutilized8Complex Client RequirementsCustom scenarios demanded per-filesystem control,scheduling isolation,and workload-specific tuning9Const

5、raints in high volume environments.10Sequential Processing PatternStorage Protect Client process typically scans and processes filesystems in a sequence,which may lengthen completion times for very large datasets.11I/O CharacteristicsSequential scanning and transfer simplify resource usage,but in ce

6、rtain workloads this can lead to uneven utilization across storage and network paths.12Scalability ConsiderationsLarge,diverse filesystems may require extra orchestration for optimal concurrency and resource use.13Architectural Overview14A lightweight,modular,scheduler extension for high-throughput