Ian - motes today have between 2kB and 10kB of RAM typically, and 50kB to 128kB 
of flash.  Many have a separate flash with many hundreds of kB (slower, more 
energy intensive to access).  Cheap processors have come a long way since the 
first motes in 2000/2001 where we had 8kB of flash and 512B of RAM (but still 
did some pretty neat things, e.g. 
http://robotics.eecs.berkeley.edu/~pister/29Palms0103).
People are spoiled now, and it's only going to get worse.  I mean better.
Several companies have demonstrateded and/or announced motes with 32 bit 
processors and ten times more RAM and flash.  These 32 bit cores often burn 
less power per MHz than the cheesy 8 bit processors with which they compete.  
You can assume that others will follow suit.  In 0.18um CMOS, you get about 
10kB of SRAM per square millimeter, and 100kB of flash per square millimeter.  
32 bit cores are around a half a square millimeter.  8 bit cores are only about 
half that size.  Radios and all of the analog peripherals are very very roughly 
5mm2.  In volume, tested functional chips cost very very roughly $0.10/mm2 .  
So today a 32bit core and 100kB of flash and 10kB of RAM is comparable in size 
to the radio+peripherals.  Radios and most of the peripherals don't scale much 
as you go to 0.13um or 90nm, but the digital does, so in a few years the cost 
of the core(s) and memory will have shrunk by 2..,4..,8..X, or more likely the 
relative die area will be about the same, but you'll just have a lot more 
memory.
 
Transmitted packets cost you a few milliseconds at roughly 20mA today, or 
roughly 100uC (microCoulombs).  The trick is receiving the packet - how do you 
know when it's coming?  There are several approaches:
* If you're time-synchronized, then the receiver can turn on for a couple of 
milliseconds and cost you maybe 50 uC if there's no packet, and 100uC if there 
is one.  If you do this once a second, that's 50uA to get 1/2 second average 
latency when idle, or 100uA to receive one packet per second.  That's TSMP 
(time synchronized mesh protocol).
* If you don't want to time synchronize, you can do preamble sampling.  You 
still turn your receiver on regularly, and listen to hear if anyone is 
transmitting a long preamble.  If they are, you stay awake until the packet 
shows up.  This is LPL (low power listening) if you're a tinyOS person.  You 
burn less charge on the listens - maybe only 20uC if there's nothing to hear - 
but you probably want to do them more often.  If you sample once per second, 
then maybe you only burn 20uA to get the same 1/2 second latency, but...now the 
transmitter of a packet has to send a 1 second long preamble before each packet 
so that you stay awake, meaning a TX now costs 20,000uC, not 100uC.  The 
receiver on average is going to have to listen to half of the preamble, which 
is going to cost 10,000uC if you only sample once a second.  So you probably 
sample a lot more often, and burn a lot more power on idle listening.  Not very 
efficient, but hey, at least it's easy.
* If you're lucky enough to have a powered infrastructure, then use WiFi.  Hmm. 
 Try again: if you're lucky enough to have a powered infrastructure, and for 
some reason are still interested in using 802.15.4, then transmit whenever you 
want, and listen immediately after your transmitions to see if the local 
powered router has anything to say to you in return.
 
ksjp
--
Prof. EECS, UC Berkeley
Founder & CTO, Dust Networks

________________________________

From: Ian Chakeres [mailto:[EMAIL PROTECTED]
Sent: Wed 5/9/2007 6:34 PM
To: 6lowpan; [EMAIL PROTECTED]
Cc: Charles E. Perkins (work)
Subject: [6lowpan] Re: 6lowpan devices constraints



Also, I'm adding RSN.

It appears that my question might have been misinterpreted or more
likely I  presented the two choices inappropriately. Let me try pose
the question in another way.

How much short term memory can we expect these devices to have? And
how much can we use for routing functions? Can a routing function make
use of 100's or 1000's of bytes of short term memory?

My understanding was that memory is scarce and very hard limited.
Alternatively, energy though scarce does not have as hard a limit.

Ian Chakeres

_______________________________________________
6lowpan mailing list
[email protected]
https://www1.ietf.org/mailman/listinfo/6lowpan



_______________________________________________
6lowpan mailing list
[email protected]
https://www1.ietf.org/mailman/listinfo/6lowpan

Reply via email to