ietf-openproxy
[Top] [All Lists]

Re: Efficacy of rule specification, processing

2001-06-13 07:10:37

Hi Lily,

I am fully aware that end-to-end rate adaptation and specific application level
content adaptation mechanisms
exist. RealPlayer's SureStream mechanism, for example, maintains media files in
multiple formats and switches across the formats based on end-to-end feedback.

I would like to restate that the example quoted was ONE illustration of a bigger
concept - that supporting dynamic
rule injection will probably be very important to support if you truly intend to
support long running media
rich sessions (which is what most RTP sessions are), in heterogenous
networking/device  environments.

Could someone give me an example of the sorts of services you thought OPES could
provide for streaming
media (RTP) sessions  ?

Thank you,
Jayanth


"Yang, Lily L" wrote:

I agree with Christian here. Remember that the rule engine is in the data
path and should not be overburdened with many service-specific controls.

Many of the dynamic behavior can be achieved with a set of carefully written
rules -- the rules are static but the run time matching will result in
dynamic behavior accordingly.
Some other dynamic behavior can be achieved by the service itself. As
someone working on video codec algorithms for 5+ years, I know for a fact
that today's codec itself can indeed adjust video quality to the network
condition dynamically without any help from a control device like OPES.
Maybe it does not go to the extreme of converting color video to
balck-and-white yet, but on the other hand, OPES can not claim to satisfy
all of your wildest dream either.

We need to consciously draw a line between what OPES infrastructure provides
and what the services themselves provide.

Lily

-----Original Message-----
From: Maciocco, Christian 
[mailto:christian(_dot_)maciocco(_at_)intel(_dot_)com]
Sent: Monday, June 11, 2001 5:49 PM
To: 'Markus Hofmann'
Cc: 'Gamze Seckin'; 'ietf-openproxy(_at_)imc(_dot_)org'
Subject: RE: Efficacy of rule specification, processing



This will probably be the default behavior of devices having
requests load
balancing capabilities without requiring a rule.

I'm not against requiring state variables. The issue I see are:
- What will be the scope, e.g. network conditions, internal
I/O conditions,
CPU load, ...
- Rule system will have to support arithmetic expression to
do so or very
limited otherwise.
- Rule module can be written by independant parties. Are we
requiring all
the OPES devices to provide network monitoring, I/O
monitoring capabilities,
and what will be the default behavior for the one which don't
provide the
feedback information.

Christian

-----Original Message-----
From: Markus Hofmann [mailto:hofmann(_at_)bell-labs(_dot_)com]
Sent: Monday, June 11, 2001 4:52 PM
To: Maciocco, Christian
Cc: 'Gamze Seckin'; 'ietf-openproxy(_at_)imc(_dot_)org'
Subject: Re: Efficacy of rule specification, processing



"Maciocco, Christian" wrote:

It looks like that by inserting state variables about
external environment
into the rules we're expecting the OPES device to control
services behavior.
I think that the OPES device should instead only decide
whether or not a
service needs to be launched and have the service itself
take care about
adapting to dynamic behavior, for example by communicating
the required
information to the service itself.

Hm, what about a scenario in which the same service is available
locally (i.e. on the OPES device) as well as remotely (i.e. on a
callout server). In this case, it would be nice to have a something
like "if my local system load is below a certain threshold,
I call the
local service, otherwise I use a remote callout". How would you
implement this if only the service itself takes care about
adapting to
dynamic behavior? And, no, I do NOT like the idea of
chaining multiple
services :)

-Markus