IETF-115 alto agenda Agenda Agenda for the ALTO 115 WG Session (Hybrid) ------------------------------------- https://datatracker.ietf.org/meeting/115/materials/agenda-115-alto Session: Friday, Nov 11,2022 Session I (09:30-11:30 UTC,04:30-06:30 EST, 01:30-03:30 PDT,17:30-19:30 Beijing Time) WG Chairs: Mohamed Boucadair (mohamed.boucadair@orange.com) Qin Wu (bill.wu@huawei.com) Available During Session: ICS: https://datatracker.ietf.org/meeting/115/sessions/alto.ics MeetEcho: https://meetings.conf.meetecho.com/ietf115/?group=alto Audio Only: http://mp3.conf.meetecho.com/ietf115/alto/1.m3u Jabber: xmpp:alto@jabber.ietf.org?join Available During and After Session: CodiMD: https://codimd.ietf.org/notes-ietf-115-alto Slides: https://datatracker.ietf.org/meeting/115/session/alto Drafts (PDF): https://datatracker.ietf.org/meeting/115/agenda/alto-drafts.pdf Drafts (TGZ): https://datatracker.ietf.org/meeting/115/agenda/alto-drafts.tgz Available After Session: Recording: http://www.meetecho.com/ietf115/recordings#ALTO Jabber Logs: https://www.ietf.org/jabber/logs/alto Introduction Chairs (10 minutes) Session Intro & WG Status Chartered items: ALTO over New Transport (30 minutes) https://tools.ietf.org/html/draft-ietf-alto-new-transport-03 https://tools.ietf.org/html/draft-schott-alto-new-transport-push-00 Discussion Leader: Roland Scott/Richard YANG ALTO OAM Support (15 minutes) https://tools.ietf.org/html/draft-ietf-alto-oam-yang-02 Discussion Leader: Jensen Zhang/Dhruv Dhody Deployment experience Update: CERN / LHCONE / ESnet ALTO Deployment Update (20 minutes) https://github.com/ietf-wg-alto/wg-materials Discussion Leader: Jordi Ros Giralt/Richard YANG TCDN Deployment Update (10 minutes) https://tools.ietf.org/html/draft-contreras-alto-service-edge-06/ https://tools.ietf.org/html/draft-lcsr-alto-service-functions-01/ Discussion Leader: Luis Contreras Non-Chartered items: for Network Aware Application (10 minutes) https://tools.ietf.org/html/draft-huang-alto-mowie-for-network-aware-app-05/ Discussion Leader: Yuhang Jia Bottleneck Structure Graphs in Multidomain Networks (10 minutes) https://tools.ietf.org/html/draft-giraltyellamraju-alto-bsg-multidomain-00 Discussion Leader: Jordi Ros Giralt Database-based Open Resource Service Framework, Usecase and Requirements (10 minutes) https://tools.ietf.org/html/draft-zhou-alto-dbors-framework-00/ https://tools.ietf.org/html/draft-zhou-alto-dbors-requirement-usecase-00/ Discussion Leader: Dongyu Yuan/Fenlin Zhou Wrapup (5 minutes) Chairs ----------- Introduction (Qin Wu) Socializing ALTO Collective effort ALTO tutorial Presented in POPSAREAD, MOPS, CERN, LHCONE, ESnet Paper to IPJ Hackathon Telefonica deployment ALTO New Transport (Roland Schott) Feedback from 4 HTTP expoert / AD Four operational modes to transfer updates from ALTO server to client Client pull Client lon pull Server push server put Major changes from 114: Split doc into multiple docs Decompose HTTP version independent components 3 docs: Doc 1 common model Doc 2: client pull Doc 3: server push (leave server put as future work, because of complexities like NAT) Doc 1: HTTP version independent Sequence of incrementl updates ALTO server is master Doc 2: Client -> Server simple design, only GET method Server -> client Long poll support Transfer scheduling Doc 3: Server push updates to clients Indicate acceptance by setting up state The latest doc in WG git Will upload the latest docs to data tracker by the weekend Seek WG reviews [MD] Are doc 2 and doc 3 doing same thing or independent? [RY] 2 and 3 are independent [MD] Clearly discourage two different solutions for the same thing Is one better than the other? [RY] SSE [MD] Push not much beloved Would be inclined towards doc 2 approach rather than doc 3 I would do 1 and 2 or 3, and in a perfect world SSE [RY] I agree, doc 2 is way to go [MD] Whether doc 1 is merge with doc 2 or 3, is editorial decision. If we pick 3, we need to keep an SSE solutio for HTTP 1 solution ALTO OAM Support (Jensen Zhang) Goal define a YANG model Progress: T1: how to handle data types defined by IANA T2: whether how to supply server-to-server communication for multi-domain T3: how to build connection between data sources and algorithm data Overall update to the doc: Align with all the 7 basic requirements Implementation specific data model T1: Decision is to use identify to define data types More modular way Future docs can define new data types Different opinions on IANA maintained module Can generate compatibility If not expect to have frequent change, is overdesign T2: Server to server communication Configure how the server to be discovered by another server/client (ready) configure how the server to disconver another server (in progress) Configure how the server to communicate to a discovered server Mechanism aligned iwth alto-server-discovery-grouping C/S can be the simplest approach to leverage existing ALTO T3: Connection between data sources and algorithms 7285 => data sources and algorithms two major components to be configured ALTO info resource = Algorithm(data sources) Real deployment in National Research Platform (NRP) (1) Common information resource, (2) implementation specific config paramgerers, (3) Conffg parameters for algorithms to access data sources Main lesson: diferent data sources may have conflicts More in the next deployments section [MD] I thought we were a bit beyond things that are not ALTO. Data collection is not part of ALTO. T3 seems not ALTO. With T2 there is no server-to-server comm in ALTO, correct? [JZ] Yes [MD] If that is correct, you are welcome to put in a different wrap, but would like us to focus on configuring the ALTO protocol and not especulate on other stuff. ALTO deployment in Telefonica (Luis Contreras) Goal of TCDN is to provide an efficient delivery of contents within the network Also presented in MOPS Network map and cost map Association of prefixes per PID <=> BGP Cost map hopount mong PIDs Pilot run on Oct 27th ALTO as an element of the production network and full intregration with TCDN Security aspects to connect with the network Initial known restriction: Topology hierarchy Restriction between hierachical leves 2 and 3. We cannot the proper info in OSPF. We have it solved for a smaller region but not the whole country yet. Bright news: Total of 16383 summarized IP address ranges IP addressses are internal and also external to Telefonica Cost map correctly reflects the defined IGP metric Load of the server is feasible Example: Luis shows metwork map and cost map snapshots (anonymized for sensitive info) Things not to bright to be : No info of IP ranges in 4% of the PoPs Some IP ranges missing Only 27 PIDs are in both network map and cost map PIDs for CDN nodes not yet captured Next steps: Understand hot to consume the ALTO info, how often? Continue analyzing data Debug mentioned issues Wait for OS issue resolution for building a complete picture of the network Productification Topology load automation Document the pilot Identify gaps/learning lessons Provide another update in IETF 116 [Fujitsu] Slide on hierarchy, security concerns? [LC] Interested in trust related areas, let’s catch up later on CERN / LHCONE / NRP ALTO Deployment Update (Jordi Ros / Kai Gao / Jensen Zhang) Deployment in Science/ academic networks across different regions (Asia, Australia, Eurpe, Americas) Leveraging on openALTO for visualization of routing state. Integration with FTS and Rucio for consumption of ALTO information. OpenALTO architecture: control palne, data plane (control FIB), data plane data. ALTO as frontend of all of this. Different updates on deployment in CERN/LHCONE (server), NRP (server) and Rucio/FTS integration (client) G2 used for determining bottleneck structures Integration examples provided Future deployment plans on these three updates targeted for IETF 116 Reporting of Experiences, Challenges and Lessons from the deployments above [Qin Wu] How this relates to ALTO OAM [Kai Gao] Conflicts detected during integration. Possible solutions: prioritization of data sources, specification of atomic query spaces. Protocol extensions could result some of the issues found (e.g., constraints on links) [Jensen Zhang] Different formats in OAM for handling that. Use of ALTO for Determining Service Edge (Luis Contreras) Motivation: Multiple computing resources (CPU, memory, bandwidth) Info from both compute and communicaiton Updates to draft: Refinement of text Clarification of statement about ALTO ntoday does not allow an edge server to be seen as an entity in both domains Getting info of edge server as an entity in the ipv4 and ane domains Luis shows snapshot examples of query request What info we can get e.g. from kubernetes Filtered property map query example Luis shows snapsho of query response example showing both compute and communication Next steps: Improve description of the interaction with cloud managers Collect comments from WG Consider the draft for recharter ** Bottleneck structures (Jordi Ros) Problem of partial information faced in cases of multi-domain (i.e., different ASs), where full knowledge of external domains is not available (full topology not shared) Proposed distributed border protocol, sharing only one metric per path (being the path a set of flows) Path gradiant graph as scalable way of sharing information between domains Description of algortihm for converging on a similar representation of the multi-domain topology in all the domains [Qin Wu] how the distributed border protocol is related to BGP? [Jordi Ros] Similar to BGP, interchanging messages betwen neighbourgs, sharing traffic ingeneering information. It can run as part of ALTO or separately. Database-based Open Resource Service Framework, Usecase and Requirements (Dongyu Yuan/Fenlin Zhou) Challenges: Fine-granularity service provisioning Capcabilities of the network remain invisible, thus differentiated service are not achieved Apps with different requirements cannot be distinguished and custome served Example: Topology in China showing how without visibility we can have bottlenecks and underutilized network regions In addition to bandwidth, network provides other services (e.g., network slicing) Several eamples of use cases. Differences between DB-ORS and ALTO Next steps Solve mentioned problems More coments and drafts Promote standardization [Liwei] How it compares with CAN [Dongyu] Same problem different approach [Luis] Delay is in microseconds, probably not relevant? [Dongyu] Discussion taken offline MOWIE (Yuhang Jia) 3GPP 3GPP R17 5G_AIS Release 18 5GAA/Predictive QoS ETSI Updates baed on exerts comments Further clarification about predicting the network Added related 3GPP work Discusses 3GPP 5G / IETF ALTO architecture Convergence of ATO and MOWIE framework is proposed in r05 [Qin Wu] Do you have implementation [Yuhang Jia] Yes. [Qin Wu] Take it to the list