Hi, Jensen: 发件人: Jensen Zhang [mailto:[email protected]] 发送时间: 2021年10月26日 9:54 收件人: Qin Wu <[email protected]> 抄送: [email protected]; alto-chairs <[email protected]>; Dhruv Dhody <[email protected]> 主题: Re: Kick off discussion on ALTO OAM work
Hi Qin and all, The updated version of the ALTO OAM draft is available here: https://openalto.github.io/draft-alto-oam-yang/draft-zhang-alto-oam-yang.html [Qin Wu] Thank for the update, I believe many of us want to see the diff, https://tools.ietf.org/rfcdiff?url1=https://tools.ietf.org/id/draft-zhang-alto-oam-yang.txt&url2=https://openalto.github.io/draft-alto-oam-yang/draft-zhang-alto- oam-yang.txt<https://tools.ietf.org/rfcdiff?url1=https://tools.ietf.org/id/draft-zhang-alto-oam-yang.txt&url2=https://openalto.github.io/draft-alto-oam-yang/draft-zhang-alto-%20%20oam-yang.txt> We missed the early submission deadline of the datatracker. But we will upload the document once the submission page is reopened. Before the IETF 112, we still want to share this updated work and get some early feedback from WG. We will also have some early discussions in the coming ALTO weekly meeting. If you are interested in this work, please feel free to join. Also, thanks for the comments from Qin. Please see my feedback inline. On Sun, Oct 17, 2021 at 12:11 PM Qin Wu <[email protected]<mailto:[email protected]>> wrote: Hi, All: I want to kick off discussion on ALTO OAM work. One relevant draft is https://datatracker.ietf.org/doc/html/draft-zhang-alto-oam-yang-00.txt Thank authors for this proposed work. I have gone through this draft and have several comments and suggestions: 1. I agree management consideration provide a set of requirements for ALTO data model design and is a good input to this document. I am wondering whether we have other reference work as input such as server discovery, server to server communication, I assume ALTO deployment document can be one of them, related to sever to server communication, what about server discovery? Do we need to configure the ALTO client for server discovery? Do we need to configure ALTO server for server discovery, suppose we use DNS mechanism to discover ALTO server, I think we actually need to configure DNS server? What am I missing? I encourage to take a close look at server discovery aspect, what is needed for ALTO data model? I totally agree with you. Although the current version does not define any data model for ALTO server discovery, it is in the plan. [Qin Wu] We need to decide what is in the scope, what is not in the scope? For ALTO server discovery, I am thinking this is more related to ALTO client configuration, At current stage, we didn’t cover ALTO client configuration. ALTO client may use DHCP mechanism to discover the ALTO server or DNS mechanism for ALTO server discovery or Neighbor discovery, I am not sure anycast can leveraged, this needs to be investigated, I think. 2. I agree we need to better manage ALTO information resource and data source, Do we need to monitor ALTO information resource lifecycle management, what is missing part is performance measurement aspect, I think we should reference section 16.2.5 to see how to provide ALTO information resource monitoring? That is a good point. The new version also has an initial proposal for statistics suggested by Sec 16.2.5 of RFC7285. But we add the statistics very carefully. Also, I think we should make one principle clear: if a feature can have already been provided by an existing OAM tool, we shouldn't define it in the ALTO OAM data model repeatedly. In other words, this document should only focus on ALTO-specific features. [Qin]:Sure, we MUST reuse existing OAM tools, avoid inventing new wheels, the focus of this draft, in my thinking is how to leverage existing OAM tools to measure ALTO service performance. Therefore defining some performance evaluation method or performance index, metrics are the key that can be covered by this models. Also consider how to integrate generic measurement framework into this data model, one relevant work is draft-xie-alto-lmap-00? This work is quite interesting. But for my understanding, it leverages the ALTO base protocol, not the OAM data model. So I guess it is more related to your third item? [Qin Wu] I think this is related to data source and data collection mechanism modeling, we can see performance data as another type of data source. you may also need to collect performance data using some OAM tools, LMAP provides generic measurement framework for High speed internet service or broadband network service. ALTO OAM model may need to consider how to generalize their measurement framework and integrate with ALTO. The relevant work in LMAP WG is RFC8193, RFC8194. 3.For data source aspect, I am wondering whether we should also consider not only where to collect data, but also how to collect data or what kind of data we can collect? e.g., we can use pub sub mechanism to collect the data, suppose we collect the routing data, topology data, performance related data, how these data are translated into network map or cost map? I know we support reactive update and proactive update, but it looks both are poll based which is slow. We also realized the limitations of the current proposed model. And yes, you are right. For the data source aspect, we should support more widely used southbound. So, the URI-based configuration may not be enough. [Qin Wu] This issue has been brought up last night meeting, I think we use internal and external to distinguish different data source type are confusing, we need to support integrating various different data sources, e.g, whether we can configure ALTO server to support BGP to collect BGP data from BGP data source, or configure ALTO server to support IGP to collect IGP data. In the current document when you say internal, I think you are referred to use NETCONF YANG to retrieve data. When you say external, I think you mean to use HTTP to retrieve the data. About how these data are translated into ALTO information resources, it is not the job of the data source part. We propose to use different algorithms to handle this. [Qin]: Good point, can we provide some references for these algorithms. Integrate different data source with different schema, translation algorithm seems important. 4. For Access Control, I feel it is confusing, I don't think access control is about a list of permissions associated with a system resource (object),e.g, data flow with the specific 5 tuples, I think access control is related to security policy such as HTTP authentication, TLS client and server authentication, TLS encryption parameters, this can be used not only in client server communication but also in server to server communication. I am wondering how this can be modelled in the ALTO data model? Yes, the access control is related to security policy. But not just server-level authentication. As Sec 16.2.4 of RFC 7285 suggests: Security policies mapping potential clients to the information that they have privilege to access. Therefore, it should be at the information resource level. But you are right, the current data model is too simple. More authentication approaches should be added. [Qin Wu] Good, I see Roland has provided some suggestions for security part. Thanks, Jensen -Qin (as individual)
_______________________________________________ alto mailing list [email protected] https://www.ietf.org/mailman/listinfo/alto
