Hi Curtis, 

I think the judgment you made was based on these slides: 
http://www.nanog.org/meetings/nanog54/presentations/Wednesday/Wobker.pdf. 
But these slides do not say the card temperature is determined by the traffic 
load. Even a card does not carry any traffic, it has to be provisioned with a 
high level of idle power (more than 80% as we measured). The difference between 
the card temperatures under the idle state and under the full-speed state is 
far less than you implied. 

Moreover, the "temperature" the slides mentioned is the ambient temp rather 
than the card temp. The increase of load on a chip means more heat, but 
meanwhile it will be removed by the heatsink and airflow. So the contribution 
it makes to the ambient temp is slim. Based on our measurement, I'd tell you 
that the change of the traffic load on a card almost has no effect on the fan 
speed at all.

Having said above, I must say that we never aggregate traffic to the extreme 
case. You can find a threshold was set in our model for the QoS consideration. 
A low 50% was used.

The slides said " line-cards are dominant consumers" while cooling power is 
only a small amount of the total power. The power consumes by one or two 
line-cards may outweigh the total cooling power. If we can brings several 
line-cards down, we surely save power even if the cooling system runs at its 
full speed and consumes more power than its normal speed.

In summary, we can get a net gain of power saving by traffic aggregation and 
power down line-cards regardless of the change of the cooling power.

Thanks,
Mingui

>-----Original Message-----
>From: [email protected] [mailto:[email protected]] On Behalf Of
>Curtis Villamizar
>Sent: Sunday, April 07, 2013 11:38 PM
>To: Alvaro Retana (aretana)
>Cc: [email protected]
>Subject: Re: IETF 86 Minutes Posted
>
>
>In message
><[email protected]>
>"Alvaro Retana (aretana)" writes:
>
>> Hi!
>>
>> The minutes have been posted here:
>http://www.ietf.org/proceedings/86/minu=
>> tes/minutes-86-rtgwg
>>
>> Many thanks to Sue Hares for being our note taker.
>>
>> Please send any corrections/additions.
>>
>> Thanks!
>>
>> Alvaro.
>
>
>WGLC is on CL requirements, not CL framework.
>
>The comment I made regarding air flow is that if you concentrate
>traffic onto one card that card runs hotter and the fan speed is
>determined by the amount of airflow needed to cool the hottest card.
>So air flow has to increase, not stay the same and therefore fan power
>will increase.
>
>Curtis
>_______________________________________________
>rtgwg mailing list
>[email protected]
>https://www.ietf.org/mailman/listinfo/rtgwg
_______________________________________________
rtgwg mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/rtgwg

Reply via email to