Mikhail has interpreted my point correctly. Another way to look at it
is to say there are two kinds of SLS (external facing and internal facing),
and clearly the customer or peer ISP is only interested in the external
facing SLS. But it's the internal facing SLS that defines PDB parameters,
which is the diffserv interest.
Except, as I said, when the customer is marking and shaping the traffic
before it reaches the ISP - in which case the PHBID/DSCP question definitely
arises on the customer side of the SLS. [You can argue that the RSVP DCLASS
object is a concrete realisation of this.]
Brian
Mikhail Smirnov wrote:
>
> Jaroslaw,
>
> I think the list is discussing a particular case for SLS over
> DiffServ networks (SLS_DS), hence (PHBID,DSCP) and their mapping.
> If you want SLS_generic be defined for internal parameters [Brian],
> then, for DiffServ you'll need to define (SLS_generic, SLS_DS) mapping.
> Am I missing your point?
> cheers
>
> Michael
>
> On Thu, 8 Feb 2001, Jaroslaw Sydir wrote:
>
> > I think that the SLS should be specified in a way that is independent of
> > the underlying technology used to implement the service. So I don't
> > think
> > that the SLS should contain PHBIDs or DSCPs. The network should map the
> > service
> > described in an SLS into a PDB, which in turn is mapped to a PHB at each
> > hop.
> > It seems contrary to the diffserv architecture to have end users specify
> > the PHB
> > that should be applied to their flow.
> >
> > Jerry Sydir
> >
> >
> > Brian E Carpenter wrote:
> > >
> > > As I implied in another note, I think the SLS needs to describe both
> > > the external parameters (the ones that the SLA will require to be met)
> > > and the internal parameters (the ones the QOS management system will
> > > use to configure diffserv, intserv, traffic engineering etc.)
> > > An SLS is a technical device, not a contract. Of course, the internal
> > > parameters don't all have to be exposed to the other party at a border.
> > > But the PHBID and DSCP mapping may need to be, if the upstream traffic
> > > source is doing its own shaping and marking.
> > >
> > > Brian
> > >
> > > Yves T'Joens wrote:
> > > >
> > > > Brian E Carpenter wrote:
> > > > >
> > > > > A detail - what should be in an SLS is not the DSCP value,
> > > > > but the PHBID value (RFC 2836) and the PHBID to DSCP mapping.
> > > > > (this mapping may be different on the two sides of the border).
> > > > > DSCP values are not invariants.
> > > >
> > > > one can ask for a certain 'network' forwarding behaviour (aka a SLS) for
> > > > a specific 'flow', and get a DSCP to apply to the stream back from the
> > > > provider. As such, the DSCP has only value on the access link itself,
> > > > and of course the provider is free to change this to another value at
> > > > its border. I assume we have a different idea about the procedure of
> > > > negotiation. Whether this DSCP is then used as indicator for PHB or not
> > > > in the network is dependent on the QoS architecture used in the network,
> > > > but should not be limited to diffserv.
> > > >
> > > > Yves
> > > >
> > > > >
> > > > > Brian
> > > > >
> > > > > Yves T'Joens wrote:
> > > > > >
> > > > > > Brian E Carpenter wrote:
> > > > > > >
> > > > > > > Actually John, isn't that list rather focussed on one particular
> > > > > > > approach to SLSs, wherease Demir's question is more general?
> > > > > > >
> > > > > > > I've been ruminating for some time whether, having figured out what
> > > > > > > a PDB is, we could now figure out the parameters to be defined
> > > > > > > in a generic SLS - and I don't think the tequila list is quite
> > > > > > > in that space. Correct me if I'm wrong.
> > > > > >
> > > > > > imho, a PDB is a piece to the puzzle, when relying explicitly on per hop
> > > > > > behaviour adherence in each hop along a 'known path' in the domain. one
> > > > > > can define an SLS that exactly maps on the parameters of e.g., the
> > > > > > virtual wire PDB. However, it doesn't stop there.
> > > > > >
> > > > > > The SLS discussed on the sls@ist-tequila.org list is somehow more
> > > > > > generic applicable then to diffserv networks. One could assume the DSCP
> > > > > > to be part of a flow descriptor on the access link to an IP domain (of
> > > > > > which a diffserv domain is an example) if it is available.
> > > > > >
> > > > > > as to demir's questions :
> > > > > >
> > > > > > >
> > > > > > > Brian
> > > > > > >
> > > > > > > John Schnizlein wrote:
> > > > > > > >
> > > > > > > > There is a mailing list for Service Level Specification.
> > > > > > > > Please take these questions there.
> > > > > > > >
> > > > > > > > To subscribe send mail to <Majordomo@ist-tequila.org> with
> > > > > > > > the following command in the body of your email message:
> > > > > > > > subscribe sls
> > > > > > > >
> > > > > > > > At 11:32 AM 02/01/2001 -0800, demir wrote:
> > > > > > > > >Is there any assumption(s)/constrain(s) on the relationship between SLA
> > > > > > > > >and service class; such as an SLAs are only for per service class (mapped
> > > > > > > > >into a DSCP)- meaning there can be only one SLA per service class between
> > > > > > > > >two domains/there can be at most N SLAs where a DS domain is supporting N
> > > > > > > > >service class?
> > > > > >
> > > > > > if packet flows at an ingress link are also characterized by e.g. a
> > > > > > destination prefix, then the combination of DSCP and destination prefix
> > > > > > may identify different streams, with as such different QoS. e.g., one
> > > > > > has a maximum delay of 20 ms (because it is on the same ISP network),
> > > > > > the other has service class (low delay, not quantified, since it is
> > > > > > transfered to another ISP network.
> > > > > >
> > > > > > > > >if the answer is "NO", is it realistic to assume that applications
> > > > > > > > >(/microflows) might ask for a "dynamic SLA"? If so, I assume a source
> > > > > > > > >domain might have more than one SLA per service class (for the sake of
> > > > > > > > >simplicity, I assume an application has an SLA interface).
> > > > > > > > >
> > > > > > > > >Any idea/insight/comment? Thank you very much.
> > > > > >
> > > > > > dynamic SLSses are theoretically feasible. in practice, much will depend
> > > > > > on the implementation that is adhered for negotiation of the SLSs.
> > > > > >
> > > > > > see also http://www.ist-tequila.org
> > > > > >
> > > > > > > > >
> > > > > > > > >Alper K. Demir
> > > > > >
> > > > > > cheers
> > > > > > Yves
> > > >
> > > > --
> > > > +------------------------------------------------------------------+
> > > > | Yves T'Joens |
> > > > | Project Manager Internet Access and Edge |
> > > > | Alcatel Network Strategy Group |
> > > > | Francis Wellesplein, 1 phone : +32 (0)3 240 7890 |
> > > > | 2018 Antwerp fax : +32 (0)3 240 9932 |
> > > > | Belgium email: yves.tjoens@alcatel.be |
> > > > +------------------------------------------------------------------+
This archive was generated by hypermail 2b29 : Thu Feb 08 2001 - 23:47:45 CET