North American Network Operators Group

Date Prev | Date Next | Date Index | Thread Index | Author Index | Historical

Re: [NANOG] Larger packets to save power, was: Re: would ip6 help us safeing energy ?

  • From: Iljitsch van Beijnum
  • Date: Mon May 05 12:24:27 2008

On 5 mei 2008, at 17:14, [email protected] wrote:

>> Obviously there is a lot to be gained at that end, but that doesn't
>> mean we should ignore power use in the network. One thing that could
>> help here is to increase the average packet size. Whenever I've
>> looked, this has always hovered around 500 bytes for internet  
>> traffic.
>> If we can get jumboframes widely deployed,

> You don't need jumboframes, you just need to have working Path MTU  
> Discovery.
> Or hand-nail your MSS to 1400 or something.  But if you don't do  
> either of
> those, you basically need to assume that the minimum MTU is 512 or so.

???

Very few people out there use an MTU significantly below 1500 bytes. A  
1500-byte MTU will give you an _average_ packet size of ~1000 on long- 
lived TCP flows because there is one tiny ACK for every two full size  
data segments. (In the other direction, but let's not make things too  
complicated right now.) The reason that the average is more like half  
that is that on short interactions the last packet is shorter, and of  
course there's stuff like gaming, VoIP, DNS that simply uses small  
packets.

>> Now obviously this only works in practice if routers and switches
>> actually use less power when there are fewer packets, which is not a
>> given. It helps even more if the maximum throughput isn't based on  
>> 64-
>> byte packets. Why do people demand that, anyway?

> Max throughput, or max packets/sec?  Max data throughput happens at  
> the
> *other* end, with 9K mobygrams...

Right, with 9k packets you only need to send around 13 kpps to fill up  
1 Gbps, with 1500 bytes it's some 83 kpps. Helps in overhead, TCP  
performance and (potentially) power use.

But someone who is sending a 200 byte packet today isn't going to send  
something larger when her MTU is increased from 1500 to 9000 so the  
_average_ won't increase by a factor 6.

>> PS. Am I the only one who is annoyed by the reduction in usable
>> subject space by the superfluous [NANOG]?

> Those of us who are *really* annoyed by stuff like that usually cook
> up a procmail recipe to strip it out.. :)

I got my procmail set up so it mostly does what I need right now,  
better not mess with it...
_______________________________________________
NANOG mailing list
[email protected]
http://mailman.nanog.org/mailman/listinfo/nanog