Load Testing Storage Performance for SQL Server
summary
Shows concrete tests and outcomes: uses DiskSpd, HammerDB, Workload Tools and SQL Server DMVs with PowerShell extract performance metrics from SQL Server DMVs and PerfMon to prove where latency and queuing occur. Single-thread read hits ~3,277 IOPS (~400 MB/s); 10 threads gives ~2.3 GB/s, 40 threads pushes latency to ~1.59 ms and exposes a network/HBA lane bottleneck that moving the HBA fixes. Demo of parallel query shows 48-core plan queues heavily; limiting to two cores restores sub‑millisecond latency.
AI-written summary · about
chapters (8)
- 0:00Storage load testing intro
- 5:23SQL Server hardware limits
- 10:21Physical vs VM vs cloud
- 15:37IO flow and queueing
- 21:35Synthetic vs real workloads
- 27:57Designing storage tests
- 38:10Key metrics and where to measure
- 45:41Interpreting results
related talks
42mSQL Server on Azure VM - Configuring for price-performance
53mOptimizing your Azure SQL Database workloads
49mSQL Server on Azure VM - Configuring for Price-Performance
58mTroubleshooting CPU Performance Issue for SQL Developers
45mBig Data and SSDs
44mBenchmarking in the cloud
47mArchitecting for High Performance SQL Server on Virtual Machines
49mHidden Gems in SQL Server 2022 Database Engine