Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

RE: [EFM] Necessity of DBA mechanisms ...




Francois,

I went back to Ajay's original e-mail and understand now. The comparison is
not between DBA and Stat Mux but rather Stat Muxing of traffic *behind* the
ONU (as in the case of SBA or static bandwidth allocation) and using DBA to
stat mux the traffic *amongst* ONUs. 

I do question the 20% number for effective bandwidth improvement in DOCSIS
DBA in the case of a typical DOCSIS user. Here is why:

If DOCSIS used SBA to divide up the 40Mbps downstream bandwidth equally per
user the result of a 1000:1 sharing of the CMTS (which is not at all
unusual) would provide a per modem bandwidth of 40kbps. Is anyone really
going to "upgrade" from a 56k dial-up modem to a 40k cable modem???

Obviously the effective bandwidth provided by DBA is not 20% but closer to
an order of magnitude or more. This makes perfect sense if I look at my own
home usage of bandwidth. The computer is probably only used 10% of the day,
(even including my kids). Even when it is used it often is not downloading
anything but instead a web page or e-mail is being read, an on-line ordering
form filled out, etc. However sometimes we download software or a Napster
song or a movie trailer. At those times the bandwidth is at least 10 times
what we experienced with a 56k v.90 dial-up modem. This is more like the
typical improvement that DBA can provide.

In the future, when high definition streaming video is the 'killer ap' then
the DBA bandwidth multiplier will not be 10+ times but something smaller,
maybe 3-5 times. If all that can be achieved is a paltry 20% then we will
all be spending too much of our time continuously consuming info-tainment.

The whole question of what DBA will buy is dependent on too many assumptions
to be boiled down to a number like 20%. I believe Ajay's 20-30% number is
based on the fact that a lot of stat muxing is going on behind the modem, as
if it is the gateway to a business. However in a residential model there is
very little stat muxing going on, even with a home LAN.

Hal

-----Original Message-----
From: Menard, Francois [mailto:f.menard@xxxxxxxxxxxxxx]
Sent: Monday, July 16, 2001 1:27 PM
To: Roberts, Hal
Cc: stds-802-3-efm@ieee.org
Subject: RE: [EFM] Necessity of DBA mechanisms ...


> I confess I am a little lost by the 20% bandwidth loss number. 

You seem to have missed the beginning of the thread ;)
I'm not stating this number, Ajay at Broadcom is.  He is saying that DBA
confers a 20 to 30% bandwidh advantage over Stat Mux.  DBA is equivalent to
request/grant mechanisms of DOCSIS.

> I also don't understand why a 32-way PON is being married to DBA but the
16-way PON is not. 

This was a mere example of my question surrounding the costs associated with
implementing support for a reques/grant scheme over using a default
statistical multiplexing mechanism.

-=Francois=-