OPTIMUSCLOUD: Heterogeneous Configuration Optimization for Distributed Databases
summary
System builds an online reconfiguration engine that jointly tunes NoSQL parameters and EC2 VM types, partitions clusters into complete sets to limit heterogeneity, and decides the minimal subset of nodes to change to avoid long downtime. Evaluated on Cassandra and Redis with real workloads, it reduces the search space from ~1.3×10^9 to 680 configurations for a 20-node cluster, and reports 40% higher throughput/$ and 4.5× lower 99‑percentile latency versus CherryPick, Selecta, and SOPHIA.
AI-written summary · about
chapters (11)
- 0:00Intro
- 2:54Cassandra's Performance on different VM types/sizes
- 4:02OPTIMUSCLOUD'S OVERVIEW
- 5:57Dynamic workloads and online reconfiguration Dynamic workloads
- 8:51OPTIMUSCLOUD's Solution Heterogeneous configurations: Reduce reconfiguration downtime
- 10:10Complete Sets This concept of Complete Set relies on selecting the fastest replica fora given request
- 11:15How partitioning the cluster into Complete-Sets reduces the search space?
- 12:28Using data-placement info to identify Complete Sets
- 14:33Performance Prediction Accuracy HPC
- 15:35Baselines 1. Homogeneous Static: the single best configuration to use for the entire duration of the predicted workload. Impractical because assumes perfect knowledge of future workload
- 20:18Conclusion For cost-optimal performance of a distributed Key-Value store in the cloud, it is critical to jointly tune Key-Value store and cloud configurations
related talks
18mSOPHIA: Online Reconfiguration of Clustered NoSQL Databases for Time-Varying Workloads
11mDatabase Evolution: Partitioning and Indexing
38mDeploying Database Clusters in the Cloud
13mOptimizing Oracle Database for Multicloud
01h 05mRoundtable: Cloud Databases as a Service
17mToo Many Knobs to Tune? Towards Faster Database Tuning by Pre-selecting Important
44mRunning Non-Cloud-Native Databases in Cloud-Native Environments
36mMigrating Applications to the Cloud in a Cost Effective Manner