Conference netcad::hub_mgnt

  Either direction will do. Flushing air velocity in the
  cabinet is typically small compared to the velocities
  in the modules, so the choice can be made according
  to convenience of installation. As you point out there 
  is no consistency in module level airflow direction. 
  Each cabinet with 3 fully loaded 900 hubs requires
  a minimum of 0.2 m�/s (425 cfm) of flushing airflow.
  More would be better.

  But how did we reach this sorry state of affairs?
  It's a long story...

  The Hub90 began with pure natural convection, which of
  course would be upwards airflow.
 
  At the time the Hub900 was on the drawing boards, the
  infamous Tall Tower (KO's last hurrah) was in full swing.
  The Tall Tower was to be a special wiz-bang cabinet,
  accommodating any and all of DEC's hardware.
  The Tall Tower had downwards flushing airflow so that 
  the exhaust would come out by people's feet rather
  than in their face - a reasonable decision. The Hub900
  was to be tightly integrated into the Tall Tower, so
  it too would have downwards airflow. The Tower's
  flushing airflow would more than compensate for the 
  counterflow of the Hub90 natural convection. The
  Tall Tower got canned, but the Hub900 still lives
  with that initial decision for downwards airflow.

  After several 900 modules had been fully designed, 
  the power supply was discovered to have significant
  thermal problems. One problematic component was
  at the bottom of the power supply and benefitted
  greatly by reversing airflow direction in the power 
  supply. So it was decided to have airflow in opposite
  directions between the 900 modules and the power
  supplies. This presented the opportunity for 
  recirculation between the power supplies and 
  adjacent modules. Analysis and measurements showed 
  this recirculation to be a minor factor.
 
  Then came the Brut, the first 90 module which 
  required a fan. The small 25mm fan used in the
  90 modules is so weak that it must work in conjunction
  with bouyancy forces, not against them. So the
  90 module fans blow upwards.

  Thus we reached the confusing situation where the
  90 modules and the power supplies want upwards
  airflow, but the 900 modules want downwards;
  an illogical conclusion reached by a series
  of steps, each logical in its own small way.


			John Thayer
	Thermal Engineer for Networking Products

 thanks John,

 Now i know the real (and poor) story .....

 jean-yves

    re. .1
    
    Murphy's Law:  Logic... its the way of coming up with the wrong answer with
    confidence.
    
    Since the fans are modular, it would be relatively easy to get them
    reverse mounted in manufacturing.  Although it may be better to wait
    until the supply issue is solved.  No sense in confusing things and 
    slowing down things any more.  8^(
    
    dave 

	
	Mechanically it would be a simple matter to turn the
	fans around and reverse airflow.

	However the temperature sensor circuit used for fan speed
	control and over-temperature protection was calibrated
	assuming airflow in the original downwards direction.
	After the power supply fan direction switch, the original
	set of modules would have to be re-calibrated to reverse their
	direction. Not insurmountable, but....

	Furthermore, and this is one of the fine points of thermal
	design, the airflow in the enclosure is more uniform if
	the fans blow out rather than blow in to the box. If you 
	have 1 hot chip and can put it directly in front of the
	fan, blowing in works great. But in a "1 box fits all
	modules" design such as the hub, it is preferable to have
	uniform airflow.