RE: [10GBASE-T] Power Down mode
Aww Brad, you don't want to ruin our fun do you?
Hugh,
The point of 10GBASE-T is to go faster than 1 Gig. To show decent efficiency at 1 Gig we had to add bursting - and then it didn't get used by products anyway. No one is making a 1 Gig repeater.
For 10 Gig at the interesting lengths for 10GBASE-T (meaning longer than what CX4 covers), CSMA/CD operation would be terribly inefficient. Remember that in addition to the link lengths in bits going up by a factor of 10, the MAC and PHY delays are much longer in bit times as much of the implementation is parallel. Unless there is a lot of bursting, it may deliver less than a full duplex 1000BASE-T link.
In the data center market and in the aggregation market, we want to know what the speed of the link is when we put it in. We do not want a link that performs between 10 and 90% efficiency because the suitability of the traffic pattern for bursting varies.
I also don't want to spend time on putting more pages into the standard that don't get used.
Not to mention the technical feasibility issues with half-duplex like staying trained.
Pat
-----Original Message-----
From: Booth, Bradley [mailto:bradley.booth@intel.com]
Sent: Wednesday, August 06, 2003 10:57 AM
To: stds-802-3-10gbt@ieee.org
Subject: RE: [10GBASE-T] Power Down mode
Hugh,
What is half duplex? :-) Seriously though, before all the other 802.3ae
and 10GbE experts jump on you, there is no half duplex in 10GbE. The
802.3ae MAC is full duplex only. Although EFM is using half duplex in
an interesting way to stall the MAC, we haven't seen anything like that
proposed for 10G.
Let's see if we can get the PAR, 5 Criteria and Objectives done first,
then we can toy with other concepts to reduce power or data rate.
Thanks,
Brad
-----Original Message-----
From: Hugh Barrass [mailto:hbarrass@cisco.com]
Sent: Wednesday, August 06, 2003 12:27 PM
To: Fakterman, Boris
Cc: stds-802-3-10gbt@ieee.org
Subject: Re: [10GBASE-T] Power Down mode
Boris,
This would only really be useful for battery powered equipment. Most
systems will be required to sustain maximum rate traffic and therefore
the idle periods will be less than 1% of the time.
Another possibility is half-duplex. If the pre-coding and pre-emphasis
is balanced well with the equalization & decoding then it may be
possible to make a half-duplex transceiver consuming half the power of a
full-duplex one. Of course there may be problems with collision radius &
bursting, but this could enable some early implementations to use HD
while the boffins are working on power reduction techniques.
Hugh.
Fakterman, Boris wrote:
> Hello all,
>
> Following the discussion regarding power, it looks like there is a
> consensus that the 10Gb Phy dissipated power will be very high at
> first silicon and relatively high at advanced future versions.
>
> The average power is important for most problematic topics, such as
> thermal conditions, power source availability and so on.
>
> The average power can be reduced by using the Power-Down mode. The
> transceiver does not transmit or receive data during significant
> periods of time. Instead of transmitting idle symbols while consuming
> full power, the system can enter the Power-Down mode. The transmitter
> power can be reduced by stopping the transmission, the receiver power
> can be reduced as only minimal receive functions will be active. The
> overall dissipated power during the Power-Done can be reduced
> significantly.
>
> Of course there are algorithmic issues to solve, such as how to
> maintain the synchronization during the Power-Down mode, but these are
> technical problems that can be discussed and solved.
>
> The average power with implemented Power Down mode depends on the
> length of idle periods.
>
> The desktop/laptop PC transmits idles most of the time ( > 90% ?). I
> don't know what happens in data centers.
>
> .If we can reduce even half of the dissipated power by the Power Down
> mode, it is worth to be considered.
>
> Regards,
>
> Boris Fakterman - Intel Communications Group, Israel
>
> Tel: 972-4-865-6470, Fax: 972-4-865-5999
>
> mailto:boris.fakterman@intel.com
>