1U Server Combo - airflow

[Alvin Oga]

On Fri, 25 Apr 2003, Roy S. Rapoport wrote:

> On Tue, Apr 22, 2003 at 03:46:28PM -0700, Heather Stern wrote:
> > Beware taking your cooling for granted just because the system works
> > with the lid open - or the heat for granted just because it runs warm
> > that way.  Only the cooling with it closed and/or sandwiched between
> > its fellows on the rack counts.  
> 
> Amen.  I've generally found cooling works much better with the lid on, as it
> forces air to go along the components you're trying to cool down.

fun stuff... :-)

CPU lifespan degrades at 1/2 for every 10C increase in operating temp
	CPU MTBF starts at say 30,000 hrs at 25C
	http://www.linux-1u.net/CPU/cpu.gwif.html

typical spacing above the cpu + heatsink + fan + airgap
	http://www.linux-1u.net/HeatSinkFans/heatsink.gwif.html

ideally, i like to see 20mm high heatsinks, and use cpu cooling fans on
the side of the heatsink ( side-to-side air flow ) instead of top-down

cpu temp w/ and without cover depends on the fan design and the
motherboard
	- some of the better motherboards has a temperature sensors that
	increases the voltage to spin the cpu fan faster ...  

	- if the fan does NOT have air above the cpu, than the fan
	blades will NOT be able to spin efficiently, and you will not 
	get the rated airflow and desired cooling of the cpu

- it also makes a difference on which "fan blade" design is used
  on the fan that sits on top of the cpu heatsink
	-- so many variables

	- there is a big differencs between fan manufacturers 
	and fan blades within the same models

- when you go to a colo, you will see lots of dead 40mm fans ...
  	- it died because it's cheap fans ( sleeve bearings vs hard ball
	bearing fans )

- the manufacturer's heatsink design and model also drastically affects
  your cpu temp 
	- some are inherently worthless
	- some heatsinks perform better than others

- to do a fair/proper cpu cooling comparason across different systems,
  i typically leave the cpu in "idle bios" mode and let its own hardware
  bios monitor tell me the cpu temp 
	- this method is presumably a good rough operating temp ballpark
	across all motherboards and all cpus and all cooling fans 
	( with lots of statistical/technical errorneous assumptions
	( when trying to compare differences of various mb, heatsinks,
	( fans, etc

	P4-2G should be sitting at 30-35C  at normal air conditioned room
	temp ( with or without covers ) 
		( better heatsink/fans will be cooler )

	AMD XP2000 runs about 30-38C at normal air temp...

	if you did infinite kernel compiles for say an hour and than
	quickly view the bios temp, the cpu typically is 5-10C hotter
		- little or no cpu change if you have a good heatsink 
		and very good airflow

- lm_sensors is good, but sometimes it is not supported on the motherboard
  and/or requires additional calibration 

- ideally, there should be ZERO difference in cpu bios temperature with
  and without the cover ...
	- typically, cpu temp goes up 1C- 3C higher when its covered 
	due to lack of air flow intake into the fanblades

	- if the cpu fans is blowing top-down onto the cpu, you're s.o.l.
	and your difference w/ and w/o covers can be drastically different
		- worst case, crashes w/ covers and stays up w/o covers

- ambient operating temperature also affects cpu cooling
	- office environment vs  computer/server rooms

	- airflow in the office obviously also affect airflow inside the
	chassis

- airflow is critical inside a 1U chassis...
	- air must only flow in one direction and free of obstructions
	like power supply, pci cards, memory, capacitors, connectors, etc

	- air does NOT bend around (90degree) corners as easily

	- so you see lots of the creative ways to guide/bend the air :-)

- you should have as much air intake holes as you have holes for air
  to flow out .. otherwise, you will be building up static air pressure
  and air will tend to stop flowing ( cooling the components ) as fast
  as it could otherwise be cooling the system
	- if you have the wrong size/shape holes for airflow, 
	than you will have a noisy system ...

- cpu cooling is an ideal for "smoke tests"
	- light a cigarette or incense and see where the smoke flows
	thru the chassis

	- if you have any turbulence, you're NOT getting the best cooling
	that you could otherwise get by adjusting the air flow

- copper heatsinks are good "heat retainers" making it feel hotter
  to the touch
	- aluminum is NOT as good as a heat conductor, so it feels
	cooler to the touch

	- a cooling fan would be able to keep an aluminum heatsink cooler
	since aluminum heatsinks wont try to retain its heat
	
	- thermal conductivity vs thermal resistivity issues between
	copper and aluminum heatsinks
		and mroe fun stuff to deal with when you have a copper
		base plating or copper core inside of an aluminum heatsink

- whomever decided to blow air down directly onto the cpu probably 
  didnt care to get the ideal cpu cooling and just live with minimal
  costs and ease to install
	- some of the new xeon cooling is done properly where air flow
	is blowing side to side .... not top down onto the cpu

	- the air flow from the cpu cooling from top-down onto the
	cpu assumes that air flows sideways upon hitting the bottom
	in a 1U chassis ( too much turbulence does NOT cool the heatink )

- whole new ballgame with liquid cooling and peltier coolers
	- start all the cpu/heatsink/airflow testing all over again
	
- somebody needs to write a finite element analysis app for running
  weather report ( air flow/temp ) simulations inside a 1U chassis
	- should be a fun project ...
	- raw data would come from autocad's 3D models of the components

	( sw at bottom of the page )
	http://www.linux-1u.net/HeatSinkFans/misc.gwif.html


and yes... have fanless 1U systems ... ( well only the power supply has
the fan, but no cpu fan )

c ya
alvin