At 6/19/2010 06:43 PM, Rubens Kuhl wrote:
>On Fri, Jun 18, 2010 at 9:43 PM, Faisal Imtiaz <[email protected]> wrote:
> > "that's a few radio hops away from anywhere.  And that's one 
> reason why per-hop latency is all-critical"
> >
> > To put things in context... from what we have seen typical 
> latency between radios (for a single link) are between 1ms to 
> 2ms... The Moto Canopy are an exception they have much higher 
> latency....because of what they do and how they do it.... so even 
> if you are going thru 20 radios.. you are talking about 15-20 ms ....
>
>
>per-hop performance may be thougher than per-hop latency... it usually
>divides by 2, so n hops would be 1/2^n performance of the main node.
>Which could be fine if you can provide fairness to prevent a kind of
>capture effect of the nearest nodes.

This is one of the problems with any kind of "best efforts" routing 
or bridging.  Loss does accumulate.  Of course it's the 
single-frequency meshes where loss goes totally gaga.  One of the 
advantages of Carrier Ethernet with Q-in-Q is that CIRs can be 
assigned to different points along the way, with reserved capacity, 
so the near-in nodes don't hog everything.  I don't think HWMPplus 
does full CE, but it may have some tools to play with.  If anybody 
can suggest a better software load for a field-mountable multi-radio 
processor, notably one that does MEF CE, I'm not wedded to 
MicroTik.  This is interim, after all; we hope to have our own code 
at some point.

On the Layer 2 v 3 thing, the distinction is artificial.  Off the 
shelf, LAN-oriented L2 switching does dumb bridging, based on an 
assumption that it's all on-site with plenty of zero-cost orange hose 
bandwidth to play with.  So STP just avoids loops.  IP itself is 
really a layer 2 protocol too!  This is non-obvious, but an IP 
address names the interface, not the application or host, and thus it 
is also a layer 2 address.  TCP/IP doesn't even have a network layer, 
just this stub that assigns two-to-three-level second names (IP 
addresses to interfaces whose MAC address is totally flat.  If you 
assign node IDs in Layer 2, it becomes smarter than IP, and IP can 
thus be run as a dumb stub protocol.

(Suggested reading:  Patterns in Network Architecture: A Return to 
Fundamentals, by John Day.)

  --
  Fred Goldstein    k1io   fgoldstein "at" ionary.com
  ionary Consulting              http://www.ionary.com/
  +1 617 795 2701 



--------------------------------------------------------------------------------
WISPA Wants You! Join today!
http://signup.wispa.org/
--------------------------------------------------------------------------------
 
WISPA Wireless List: [email protected]

Subscribe/Unsubscribe:
http://lists.wispa.org/mailman/listinfo/wireless

Archives: http://lists.wispa.org/pipermail/wireless/

Reply via email to