Document matching Dell on CPU rather than out-bidding it

With the service stopped under sustained load, Dell held 5880-6120 RPM
with CPU2 at 70-74c. The CPU ramp calibrated from an idle reading asked
for 36-50% over that same range - up to 9720 RPM - while the drives ran
2c warmer. The extra airflow bought nothing.

Dell has the chassis thermal model and does CPU cooling well; what it
cannot see is drive temperature. Keeping the CPU ramp out of the way
until 75c and reaching full at 77c (iDRAC's own warning point) matches
stock RPM for 50 of 60 samples under load.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
This commit is contained in:
Zeb Hering
2026-08-28 23:05:58 -07:00
parent 4dffbb070c
commit 1f7db6d4b0

View File

@@ -189,6 +189,25 @@ The rule: find the drives' normal peak, put `DISK_RAMP_LOW_OFFSET` a couple of d
above it, and let the alarm (`DISK_ALARM_OFFSET`, 5c below the limit) catch anything real.
Cooling a healthy drive from 57c to 53c is not worth 27% more fan.
### Let Dell do the CPU; the disks are what it cannot see
Dell's profile has the chassis thermal model and is good at CPU cooling. What it does not
do is look at drive temperature *at all*. So the CPU curve should approximate Dell's
rather than out-bid it, and the disk curve is where this script earns its keep.
Measured on `iz-pve0` under a sustained migration, with the service stopped so Dell had
control: Dell held **5880-6120 RPM with CPU2 at 70-74c** — 3c below its own 77c warning
point. An earlier CPU ramp of 62-73c asked for 36-50% across that same range, i.e. up to
**9720 RPM against Dell's 5880**, while the drives ran 2c *warmer*. The extra 3800 RPM
bought nothing.
The fix was to keep the CPU ramp out of the way until 75c and reach full speed exactly at
77c, where iDRAC itself starts warning. Result: 26% / 5880 RPM for 50 of 60 samples,
matching stock, with brief excursions to 35% when CPU2 touched 76c.
Calibrating the CPU ramp against an *idle* reading is what caused this. Idle tells you
where the floor should be; only a loaded CPU tells you where the ramp should start.
### Expect to run above stock on some hosts, legitimately
Dell's profile does not look at drive temperature at all — only CPU, inlet and exhaust.