Hi Sabine, all who are working on filtering, Here is a use case that I use, personally, to evaluate a filtering service:
- this paper, which won IMC'12 Best Paper, measured the structures of several applications, including Skype, Google Hangout, and iChat: http://eeweb.poly.edu/faculty/yongliu/docs/imc12tech.pdf - Suppose ALTO is ready and can be used for the Skype case (see Figure 2 page 4). The setting then is: Input: - a set S of candidate relay servers (say VIP represented by pid or real IP); - a set C of participants; Output: - determine who should be the initiator (we often try different people as initiator to try o improve performance): init(C); - determine a relay server for each c in C: relay (c) One can see that there can be multiple constraints: - audio paths: c1 -> init(C) -> c2 latency requirement, for any c1, c2; Download and upload bw of init(C) >= all video streams; - video paths: latency and bw on the servers. I am hoping that any filtering service, if we design, is evaluated by the preceding case. Cheers! Richard On Friday, October 18, 2013, RANDRIAMASY, SABINE (SABINE) wrote: > Hi Richard, **** > > ** ** > > Please see inline**** > > Sabine**** > > ** ** > > *De :* [email protected] <javascript:_e({}, 'cvml', > '[email protected]');> [mailto:[email protected]<javascript:_e({}, > 'cvml', '[email protected]');>] > *De la part de* Y. Richard Yang > *Envoyé :* vendredi 18 octobre 2013 08:14 > *À :* RANDRIAMASY, SABINE (SABINE) > *Cc :* IETF ALTO; ROOME, Wendy D (Wendy); > [email protected]<javascript:_e({}, 'cvml', '[email protected]');>; > Qin Wu; Young Lee; Greg Bernstein; [email protected] <javascript:_e({}, > 'cvml', '[email protected]');>; dhruv.dhody > *Objet :* ALTO Extension: A document defining multi-metrics filtering?**** > > ** ** > > Hi Sabine,**** > > ** ** > > Good comments. Please see below.**** > > ** ** > > On Thu, Oct 17, 2013 at 1:40 PM, RANDRIAMASY, SABINE (SABINE) < > [email protected]> wrote:**** > > A few questions and comments inline to make sure I understood. If we are > to document proposed filtering services, should we put separate sections, > one on filtering services for the basic protocol and the other one, more > prospective for protocol extensions on topologies?**** > > Following the guideline of modular design, how about using different > documents whenever possible?**** > > *[ ] agree* > > * * > > Here is some thinking on organization. A main design feature of the > ALTO base protocol is its single-switch abstraction. Hence, how about we > first focus on filtering on this base abstraction, and not beyond?**** > > *[ ] Ok then I will start on my side and document the set of > filtering services for the base protocol, based on the Multi-Cost 07* > *proposal, > the proposals of Young et al. and Qin et al. , list what I see as > possible harmonization items to have a discussion basis. ***** > > ** ** > > Specifically, in this base abstraction model, a network map and a cost map > based on the network map defines an abstract, single switch. Each PID in > the network map defines a PoP (i.e., the set of endhosts, which can be > either servers or clients in an application context, attached to this PoP), > and the cost map of each cost metric defines the pair-wise values of the > given cost metric. Let V be the set of n PIDs (nodes), and E the set of n^2 > edges. Filtering is to help an application to narrow down its choices when > building an application overlay based on this single-switch model. Hence, > we need to define the following components:**** > > ** ** > > - how to filter the set of nodes?**** > > - how to filter the set of candidate overlay paths (one hop or multiple > hops in extreme e.g., origin -> edge server -> consumer)?**** > > *[ ] Ok***** > > ** ** > > Richard**** > > **** > > Thanks**** > > Sabine**** > > **** > > **** > > *De :* [email protected] [mailto:[email protected]] *De la part de > * Y. Richard Yang > *Envoyé :* lundi 14 octobre 2013 04:42 > *À :* IETF ALTO > *Cc :* RANDRIAMASY, SABINE (SABINE); ROOME, Wendy D (Wendy); > [email protected]; Qin Wu; Young Lee; Greg Bernstein; [email protected]; > dhruv.dhody > *Objet :* Re: ALTO Extension: A document defining multi-metrics filtering? > **** > > **** > > Let me add on: although a filtering service can be a very useful service, > it can also be quite involved, and hence the WG may need to think through > the issues when designing this service. For example, there are two types of > use cases:**** > > **** > > - end-to-end: given src set {s1, s2, s3, ..., sn} and dst set {d1, d2, d3, > ..., dm}, return all pairs (si, dj), where si in {s1, ..., sn} and dj in > {d1, ..., dm} such that (si, dj) satisfies the constraints;**** > > *[ ] This use case corresponds to the base protocol. The requested > service is the ALTO Cost Map Filtering Service w.r.t metric values. It > seems easy to extend this service by extending the set of constraints with > additional constraints metrics and additional logical operators. ***** > > **** > > - relay: given src s, dst d, and a relay candidate set {r1, r2, ..., rk}, > return all of the ri such that s -> ri -> d satisfies the constraints. *** > * > > *[ ] This use case relates to a proposed ALTO extension on topologies > providing cost maps where there may be several paths between an (S,D) pair. > Is the ALTO Service here the ALTO Cost Map Filtering Service w.r.t metric > values adapted to a multi-path topology? The cost map here would need to > include some index on the cost value w.r.t. the relay ID and the service > would only return cost values on the indexed (S,D, idx) triples satisfying > the constraints. If this is correct, then such a use case would well fit > into filtering services for topology based ALTO extensions. ***** > > **** > > Note that with relay, we will then need to worry about the "composition" > semantics of metrics. For example, delay might be additive, loss rate > (unless small and independent) may not be.**** > > *[ ] I think the composition and optimization operators should be > documented in any case (I think in the “use and application” field of the > RFC6390 template). Indeed, the OPTIMUM-COMPOSITION attributes may be > MIN-SUM, MAX-MIN, MIN-PROD etc… ***** > > ** > >
_______________________________________________ alto mailing list [email protected] https://www.ietf.org/mailman/listinfo/alto
