Re: [alto] draft-yang-alto-path-vector-03 and cost-map service

[Hans Seidel <[e-mail]>]

Hi Jensen,

Thank you for your response.

see comments inline.

On 11.08.2016 19:16, Jensen Zhang wrote:
> Hi Hans,
>
> Is it possible to compress the cost map by using routing state 
> abstraction? I think it is a potential solution.
As far as I understand the routing state abstraction draft, it can be 
used to eliminate links e.g. from path-vectors regarding client defined 
conditions to reduce the size of the cost-map.
However, I see two potential limitations of routing state abstraction in 
full cost maps.

1. Assuming a full mesh cost map where costs from every PID to every PID 
are provided, so we have a flow from every PID to every PID. The more 
flows, the more fine grained is the routing state abstraction, since the 
number of shared links among different flows increases. If no 
information should be lost, I think the number of links that could be 
eliminated decreases with the number of flows. So, having many flows 
will result in little to no size reduction. Please correct me, if I got 
this wrong.
2. The computation effort for a large number of PIDs in large networks 
might exceed an acceptable limit.

> Another solution may be to add a proxy layer to compress and 
> decompress ALTO protocol messages. It will reduce the communication 
> time and not change the implementation of ALTO servers and clients. 
> But compression and decompression will increase extra processing time.

Yes, compression is a way to reduce the size at the cost of extra 
processing. It reduces the map size to 3.7MB without and 11MB with 
path-vector but keeps the size ratio at 1:3.

Hans

>
> We haven't test such large-scale networks, but it is indeed a problem 
> which need to be handled.
>
> Best,
> Jensen
>
> On Thu, Aug 11, 2016 at 9:08 PM, Hans Seidel <[e-mail] 
> <mailto:[e-mail]>> wrote:
>
>     Hi Richard, all,
>
>     after the ALTO session in Berlin, we shortly talked with Ingmar
>     about the impact of path-vector on the size of cost maps
>     especially in large-scale networks.
>
>     I carried out some tests with path-vector cost maps based on our
>     data. Our cost maps are already very large but path-vector maps
>     are about three times larger (~50 MB vs. ~150 MB in uncompressed
>     state). In average we have round about 4 hops between two PIDs
>     which leads to an average path-vector of the same length. ECMP was
>     not considered in the test but it will certainly further increase
>     the size of the map.
>
>     Our idea to reduce cost map size is to provide topology
>     information, e.g. with the property map presented in the
>     unified-props draft, and let the client carry out the path
>     determination. This means, the ALTO server provides the network,
>     cost and property map to enable clients to get their desired level
>     of detail for the path costs.
>
>     I also think this approach can coexist with path-vector cost maps.
>     An ALTO server can provide both cost maps with and without path
>     vector and a property map providing the topology. This way it is
>     up to the client whether it wants to save bandwidth and invests
>     some processing time to perform path determination by itself or it
>     fetches the full path-vector cost map.
>
>     Any thoughts on this?
>
>     Cheers
>     Hans
>
>
>     On 01.08.2016 23:37, Vijay K. Gurbani wrote:
>
>         Folks: As the (draft) minutes [1] of IETF 96 reflect, there
>         was general
>         consensus on adoption of path vector and routing state abstraction
>         documents towards the charter deliverable of graph representation
>         formats in ALTO.
>
>         The chairs will like to start a call for adoption of the two
>         documents
>         --- https://tools.ietf.org/html/draft-yang-alto-path-vector-03
>         <https://tools.ietf.org/html/draft-yang-alto-path-vector-03> and
>         https://tools.ietf.org/html/draft-gao-alto-routing-state-abstraction-03
>         <https://tools.ietf.org/html/draft-gao-alto-routing-state-abstraction-03>
>         --- as deliverables towards the charter item.
>
>         Note that there remains some ambiguity (in the chair's mind)
>         on whether,
>         once adopted, these will proceed as two drafts or whether they
>         will be
>         merged in one.  The authors of these drafts are urged to provide
>         clarity on the evolution of these documents.
>
>         The call for adoption runs for two weeks, from Mon Aug 1, 2016
>         to Mon
>         Aug 15, 2016.  This will be a good time to comment on the list and
>         inform the working group of any issues with adopting these
>         items, or
>         whether the working group was remiss in considering other
>         documents.
>         Please post to the list.  (Even a simple "I support these
>         documents
>         towards charter deliverable" is a good indication.)
>
>         Thanks,
>
>         [1]
>         https://www.ietf.org/proceedings/96/minutes/minutes-96-alto
>         <https://www.ietf.org/proceedings/96/minutes/minutes-96-alto>
>
>         - vijay
>
>
>     _______________________________________________
>     alto mailing list
>     [e-mail] <mailto:[e-mail]>
>     https://www.ietf.org/mailman/listinfo/alto
>     <https://www.ietf.org/mailman/listinfo/alto>
>
>