[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Consensus ?



I completely agree that we should focus on doing FUNCTIONAL
blocks simillary to what diffserv WG has done for example in
"A Conceptual Model for Diffserv Routers"
(draft-ietf-diffserv-model-01.txt).

We should have a similary model but for a Policy based concept.

// Patrik Lindergren

----- Original Message -----
From: Francis Reichmeyer -NJ <FranR@iphighway.com>
To: Shai Herzog - NJ <herzog@iphighway.com>; <Polterm@ops.ietf.org>;
<mumble@ops.ietf.org>; <policy@raleigh.ibm.com>; rap <rap@iphighway.com>
Sent: den 14 november 1999 00:13
Subject: RE: Consensus ?


> Overall, I've got no problem with the terminology
> propsed here. I see it as expanding on the "functions"
> that were discussed in the presentation in the polterm
> BOF by removing the arrows from the individual functions
> given in the BOF (this is a good thing since more than a
> few people noted that the functions themselves may actually
> be carried out in more than one place in the policy system).
>
> I would like to see emphasized in the discussion below that
> the boxes shown are FUNCTIONAL blocks and not system components.
> These functional blocks may reside in different components in
> different systems, especially the PA and ED functional blocks,
> depending on the implementation.
>
> -Fran
>
> > -----Original Message-----
> > From: Shai Herzog [mailto:herzog@iphighway.com]
> > Sent: Friday, November 12, 1999 6:37 PM
> > To: Polterm@ops.ietf.org; mumble@ops.ietf.org; policy@raleigh.ibm.com;
> > rap@iphighway.com
> > Subject: Consensus ?
> >
> >
> > Hi,
> >
> > During the meeting on Friday (policy terminology) some of us had
> > seem to converge on terminology. I'd like to lay it down
> > and test whether we actually do agree on it (and if not, consider
> > this a continuation to the discussion)
> >
> > This family naming scheme came from a discussion we had after the WG,
> > on Friday. (Ed Ellesson "volunteered" to play the "Kid" in a family
> > where the PA stays at home with the kids, and MA is out at work ;-)
> > Putting aside the humor in this naming scheme, it may actually be
> > quite effective in explaining the role of each component.
> >
> > 1. The bottom of the food chain is an enforcement device (formally
> >     known as PEP), which has the ability to actually performs the
> >     task (service packets, encrypt IPSec, etc.)
> >
> >     Lets (temporarily) call it "Enforcement Device" or ED.
> >
> > 2. The top of the food chain is a non-volatile repository (formally
> >     known as LDAP directory). Which is entrusted with storing
> > the standard
> >     high level policy in an interoperable format. This storage isn't
> >     necessarily centralized, and may comprise of a set of distributed
> >     repositories.
> >
> >     Lets (temp) call it "MA" (Master Archive).
> >
> >     [Other options: "ESP" (External Storage Point) or
> >     GOD (Goals & Objective Directory) have been rejected
> >     at this time ;-)]
> >
> > 3. In some cases (Outsourcing or Provisioning), the enforcement
> >     device needs assistance from an external policy advisor
> >     (if it is incapable or deficient in policy processing)
> >     on it's own. (Formally known as "Policy Server").
> >
> >     Lets (temp) call it "PA" for Policy Advisor.
> >
> >     Unlike ED, PA is totally incapable of actually enforcing
> >     anything. It is a mere advisor, providing ED with
> >     instructions, hoping that the "ED" will implement them.
> >
> > 4. As in most application there is a GUI/UI component
> >     which receives administrator input.
> >     The Form of this input, or where the GUI is hooked up to
> >     is totally a vendor dependent issue: for example, some vendors may
> >     have a GUI that is a smart "LDAP Directory Editor", which
> > allows the
> >     administrator to edit the contents of the "MA", while others may
> >     have a GUI that works directly with the "PA", while the
> > PA indirectly
> >     populates the MA repository.
> >
> > This 4 components can be viewed architecturally as:
> >
> >                 +----------+
> >    Master       |    MA    |<----+
> >    Archive      +-----+----+      \
> >                      /|\           \     +-----------+
> >                       |  LDAP       >----+    UI     |
> >                      \|/           /     +-----------+
> >    Policy       +-----+----+      /
> >    Advisor      |    PA    |<----+
> >                 +-----+----+
> >                      /|\
> >                       |  Standard Policy Protocols (COPS etc.)
> >                      \|/
> >    Enforcement  +-----+----+
> >    Device       |    ED    |
> >                 +----------+
> >
> >
> > In the outsourcing case: ED has a policy question and it goes to ask
> > PA's advice. PA looks at the policy set by MA, and after careful
> > consideration provides ED with a reply.
> >
> > In the provisioning case: MA sets policy instructions. PA examines
> > them and gives ED his specific instructions (at his level of
> > understanding).
> >
> > --------------------------
> >
> > Now that we defined the basic functional blocks, the next question is
> > what kind of functions are needed to bring the policy set by MA to
> > the form understood by ED. In the meeting we discussed
> > several functions.
> > (Note that this list may not be conclusive; there may be more such
> > functions to be discovered later.)
> >
> > a. Translation
> >
> >     "Compilation" of High-Level policy into lower level one.
> >
> >     An example could be translation of DNS names or UserIDs into
> >     IP Addresses. "Gold Service" into AF1, etc.
> >     The main idea here is that it is pure translation that
> > always produces
> >     the same results (w/o taking into account dynamic network
> > information).
> >
> > b. Decision
> >
> >     "considering multiple independent sources of information
> > and policy
> >      to form an "Ad Hoc" lower level policy.
> >
> >     An example could be combining congestion information with
> > type of device
> >     to decide on the class of service a certain flow should
> > be mapped to.
> >     Naturally a decision may change over time even with the policy in
> >     effect is NOT changed (hence, it is NOT a translation).
> >
> > c. Formatting
> >
> >     Once either Decision or Translation has been made into lower level
> >     policy it may be the case that the device needs proprietary (non
> >     standard) set of instructions (for instance, a non-COPS
> > legacy device).
> >     It may then be needed to reformat the policy instructions into the
> >     proprietary format for this legacy device. This feature was
> >     previously known as "COPS Proxy").
> >
> > So, the overall components may look like:
> >
> >
> >                         +----------+
> >                         |    MA    |<----+
> >                         +-----+----+      \
> >                              /|\           \     +-----------+
> >                               |  LDAP       >----+    UI     |
> >                              \|/           /     +-----------+
> >                +--------------+----+      /
> >                |             PA    |<----+
> >                | +------+  +-----+ |
> >                | |Trans.|  |Dec. | |
> >                | +------+  +-----+ |
> >                +--------------+----+
> >                  /|\         /|\
> >                   |           |  Standard Policy Protocol (COPS etc.)
> >                  \|/         \|/
> >                +--+----+   +-------+
> >                | Format|   |  ED   |
> >                +-------+   +-------+
> >                  /|\
> >                   |              Proprietary Management Protocol
> >                  \|/
> >                +--+----+
> >                | Legacy|
> >                | ED    |
> >                +-------+
> >
> > I do believe that translation and Decision are used in
> > typical policies
> > in any combination and order. I don't see how we can decide
> > one is done
> > prior to the other. In fact, this is where different vendors
> > will choose
> > different ways.
> >
> > There may be other policy related funtions that should reside in
> > PA: contention resolution, error resolution (where a decision must
> > change as a result of an error from the device), etc. But we haven't
> > discussed those in the BOF meeting.
> >
> > Comments? Tomatoes?
> >
> > Shai
> >
> > __________________________________________________________________
> > Shai Herzog, Founder & CTO         http://www.iphighway.com/herzog
> > Cell: (917) 449-7889                      Parker Plaza, 16th Floor
> > Tel : (201) 585-0800                                 400 Kelby St.
> > Fax : (212) 656-1006                             Fort-Lee NJ 07024
> >
>