Storage speed after the fix
Slow storage makes every query slow.
Sometimes the storage is slow for a reason that has nothing to do with the disks.
Problem
Data files were taking 100 to 565 ms to respond while the server moved only 7 MB per second. Queries that used to be fast suddenly took longer, and nothing in the code had changed.
The storage team said everything was fine. But those numbers said otherwise.
You do not expect high latency under almost no load from tier 1 storage. Anything above 20 ms is already a problem.
Solution
First we measured the storage with CrystalDiskMark, a free disk benchmark tool (64 GiB test file, 70% read and 30% write mix) to get a baseline. Then we looked for the cause.
The cause was the Windows power plan. The server was running on the default Balanced plan, which slows the CPU down to save power. That slows I/O processing too.
This is what we did to fix the issue: we changed the power plan to High performance and ran the same test again.
Before vs. After
Random 4K throughput went from 33 MB/s to 336 MB/s on the same disks.
| Random 4K, queue depth 32, 16 threads | Before (Balanced) | After (High performance) | Improvement |
|---|---|---|---|
| Mixed 70/30 (MB/s) | 33.07 | 336.38 | 10.2x |
| Read (MB/s) | 25.94 | 329.10 | 12.7x |
| Write (MB/s) | 6.07 | 342.69 | 56x |
| Read IOPS | 6,333 | 80,346 | 12.7x |
| Read latency (ms) | 79.4 | 6.4 | 12.4x |
Sequential speed improved too, from 510 MB/s to 1,532 MB/s for reads.
The server and the storage were the same in both tests. Only the power plan changed.
Final thoughts
Check the power plan on every database server. Balanced is the Windows default, and it is the wrong setting for SQL Server.
It takes a minute to check and a minute to fix, and we still find it on production servers during health checks.
Before you blame the SAN or buy faster disks, look at your operating system settings.



