Re: [Teas-ns-dt] issue discussion: NF resource

[Jeff Tantsura <[e-mail]>]

Non networking related metrics for networking services are difficult/impossible to normalize, if I want a FW with a particular performance matrix, it wouldn't  a number of CPU’s/memory but rather PPS/number of sessions/tunnels, and obviously availability.
I’m not sure I understand what  “Minimum SLOs” are? A SLO has its boundaries , as long as the service delivered is within these boundaries - SLO is met, otherwise it isn't.

I really like to following simple definitions:

	  SLI						SLO						SLA
X should be true     	Y proportion of the time		      or else


Define SLIs and SLOs for specific capabilities at service boundaries.
Each logical instance of a service gets its own SLO.
Combine SLIs for a given capability into a single SLO for that capability.

Cheers,
Jeff
On Jun 1, 2020, 12:44 PM -0700, Greg Mirsky <[e-mail]>om>, wrote:
> Hi Kiran,
> I much appreciate your consideration of my comments.
> As for your example, I think that we can expect an operator to be knowledgeable and experienced in mapping customer intentions into a particular technology the operator is using. That is where I'd leave space for operators and vendors to innovate and differentiate their products. What we can provide, in my view, is performance measurement methodology(ies) that will help customers to have quantitative representation of the TS service.
>
> Regards,
> Greg
>
> > On Mon, Jun 1, 2020 at 12:30 PM Kiran Makhijani <[e-mail]> wrote:
> > > Hi Greg,
> > > In order to resolve the issue:
> > > Since the section is “Minimum SLOs”, I agree to “move it outside the minimal”.
> > >
> > > My actual reply would put us back in the loop: If an operator gave less cpu and firewall runs slow then the user is not happy. Shouldn’t they be able to provide some criteria to get their ideal performance? How do you meet that objective upfront?
> > > -Kiran
> > >
> > > From: Greg Mirsky <[e-mail]>
> > > Sent: Thursday, May 28, 2020 6:37 PM
> > > To: Kiran Makhijani <[e-mail]>
> > > Cc: [e-mail]
> > > Subject: Re: [Teas-ns-dt] issue discussion: NF resource
> > >
> > > Hi Kiran,
> > > thank you for your kind consideration of my comments.. If we agree that a customer of a TS may specify a certain TS behavior that is externally observable on data packets, then I don't find where elements like "memory", "computing power" fit. I think that these are important to an operator when choosing the most economical and efficient method to realize the TS according to the specific LSO.
> > > What do you think?
> > >
> > > Regards,
> > > Greg
> > >
> > > On Thu, May 28, 2020 at 4:56 PM Kiran Makhijani <[e-mail]> wrote:
> > > > Hi Greg,
> > > > You are right. I should not have used term ‘virtual’ – I was visualizing a NF container in NBI data model. A user of transport slice just asks for a NF and associated properties. Whether it’s V- or P- is a realization time decision.
> > > >
> > > > I see NFs adding a lot of value from higher-layer perspective, without getting bogged down about network details users can attach app specific firewalls, IDS, and caching type of functions on the endpoints. These are all good examples for use of NFs in slices. Secondly you will be able to express your requirement from a single place instead of dealing with different interfaces.
> > > >
> > > > <your quote>
> > > > the TS does not specify which NFs should compose the TS but only what is the functional behavior and performance provided by the TS.
> > > > ^^^
> > > > I hope you agree that there’s a need for some “functional behavior”. If user can leverage provider network for that behavior, then why not?
> > > > -Kiran
> > > >
> > > >
> > > > From: Greg Mirsky <[e-mail]>
> > > > Sent: Thursday, May 28, 2020 3:15 PM
> > > > To: Kiran Makhijani <[e-mail]>
> > > > Cc: [e-mail]
> > > > Subject: Re: [Teas-ns-dt] issue discussion: NF resource
> > > >
> > > > Hi Kiran,
> > > > I think that an NF can be realized as a physical device or a VM. Would you agree? If that is the case, then, in my opinion, we cannot consider only the VNF case but add the PNF case as well.
> > > > Also, I don't see the real benefit to have NF, in all their variety now and more in the future, as part of a TS. I think that NF is either part of the TS implementation and thus is selected by an operator based on the SLO. Or part of a service that uses the TS. In the model I have in mind, a client of the TS does not specify which NFs should compose the TS but only what is the functional behavior and performance provided by the TS.
> > > >
> > > > On Thu, May 28, 2020 at 12:06 PM Kiran Makhijani <kiranm@futurewei..com> wrote:
> > > > > Hi All,
> > > > > To kick off discussion on this topic.
> > > > >
> > > > > During earlier discussions several of us asked for network functions should be part of the slices. For me the value of doing this is to be able to express different nuances of a service through single set of objectives. Then in NBI and implementation there is one controller to monitor the performance.
> > > > >
> > > > > At the entry and exit of transport slices, these NF can be supplied to enforce access control, policies, etc. on transport slice traffic. It is also safe to assume that they will be virtual.
> > > > >
> > > > > The objective is to apply those function on  traffic flows.
> > > > > 2 examples are  (1) use of firewall at entry point with the rules described by the user of transport slice. (2) Caching or DPI function on exit endpoints.
> > > > >
> > > > >
> > > > > Regards
> > > > > Kiran
> > > > >
> > > > > --
> > > > > Teas-ns-dt mailing list
> > > > > [e-mail]
> > > > > https://www.ietf.org/mailman/listinfo/teas-ns-dt
> --
> Teas-ns-dt mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/teas-ns-dt