Internal NSD representation

Owned by Stephen terrill
Last updated: Jan 29, 2019 by Gil Bullard
16 min read

1 Introduction

1.1 Problem Description.

There is an agreement to be able to:

  1. Onboard a Network Service Descriptor as defined by ETSI-NFV.

  2. To represent the Network service descriptor in the internal modelling.

The way to represent the network service descriptor is an ongoing discussion and there are two proposals.

Option A:  The network service descriptor is represented implicitily using the existing service descriptors.

Option B: The network service descriptor is represented explicity in the modelling.

1.2 Approach

Describe both options and cover the following information:

  • Show how it works for a particular use cases.  the vCPE use cases is a good approach.

  • Cover the following:

    • Model representation (Cover E2E service, NS, xNF): 

      • Onboard and defined model

      • Dynamic model in AAI 

    • Onboarding

    • Distribution from SDC to the rest

    • Service instantiation 

2 Option A: Implicit NSD representation

2.1 Assumptions:

  1. Management of the application-specific functionality of a VNF is envisioned as being outside of the scope of ETSI NFVO and VNFM functionality.  ETSI NFV is currently silent about how such application level management is to be performed for a given VNF, but the assumption seems to be that it will be performed by OSS/BSS (e.g., SM/NM) or EM.  This can be seen by the lack of specification of the EM to VNF interactions in Fig 5-1 below, as well as in the delegation of application configuration to the EM in the highlighted portions of the figure labeled B.3.2.2 below.  (Figures taken from ETSI GS NFV-MAN 001 V1.1.1, 2014-12). 

    1. Alternative Text:  ETSI NFV is currently silent about how application level data is managed for a given VNF, but provides elements in the VNF descriptor which are intended to hold this application level data.  The ETSI SOL003 Or-Vnfm  interface between the NFVO and the VNFM provides mechanisms to pass this data to the VNF but alternative mechanisms such as direct communication from the OSS/BSS (e.g., SM/NM) or EM.  The common alternative of using the vendor's EM is highlighted in the the figure labeled B.3.2.2 below.  (Figures taken from ETSI GS NFV-MAN 001 V1.1.1, 2014-12). 



  1. In other words, it seems that the ETSI NFV community took the perspective that there were already well-established systems, precedence and standards around the tools (e.g., BSS/OSS/SM/NM/EM) needed to manage traditional Network Functions (i.e., PNFs), and the need of ETSI NFV was to put standards around that which was new: how to instantiate and managed "Virtualized" Network Functions to the point that they can look to the established systems (SM/NM/EM) indistinguishable from "traditional Network Functions".

  1. Therefore, in the context of Figure B.3.2.2 above, "deployment specific parameters" should be understood as that subset of VNF configuration data that must be applied in the VNF in order to make it an endpoint capable of being managed by the non-ETSI NFV OSS/BSS/SM/NM/EM. 

  2. Thus, an ETSI Network Function Descriptor (VNF or PNF) would be expected to formally capture only those management requirements for the "deployment aspects" of that Network Function, and not the "application aspects".

    1. Alternative Text: An ETSI Network Function Descriptor (VNF or PNF) would be expected to formally capture those management requirements for the "deployment aspects" of that Network Function, and optionally also formally capture the management requirements for the "application aspects" of the VNF "application".

  3. Similarly, an ETSI Network Service Descriptor would be expected to formally capture only those management requirements for the "deployment aspects" of that Network Service, and not the "application management aspects".

    1. Alternative Text: Similarly, an ETSI Network Service Descriptor would be expected to formally capture the management requirements for the "deployment aspects" of that Network Service, and optionally also the "application management aspects".

  4. Certain Service Providers may already have the necessary BSS/OSS/SM/NM/EM systems in place to manage the "application aspects" of an ETSI NS and associated NFs, and look to ONAP only to manage the "deployment aspects" of such a NS and NFs.  Such Service Providers should have the option to have ONAP perform only the equivalent functionality that would be performed by an NFVO/VNFM given that same NSD/NFD collection.  Through the principle of "least astonishment", such a Service Provider would expect to elicit such behavior from ONAP by onboarding the corresponding NSD/NFD collection into SDC without any substantial extensions to that information content needing to be entered into SDC.

  5. However, other Service Providers may not have or want to leverage BSS/OSS/SM/NM/EM systems for this purpose, but rather look to ONAP to manage the "application aspects" of VNFs and Services.  These Service Providers should have this option as ONAP was envisioned to be able to do just this, such as perform application level configuration of VNFs within the context of their Services, and to perform control loop functionality on application level telemetry.  However, because the ETSI Descriptors as currently defined would not be expected to formally contain information as to the application management needs of the NS/NFs, a Service Provider that onboarded such descriptors into SDC would reasonably expect to have to provide substantial additional information content to SDC regarding the "application management aspects" of the "extended" Service and Network Functions.

    1. Alternative Text: However, other Service Providers may not have or want to leverage BSS/OSS/SM/NM/EM systems for this purpose, but rather look to ONAP to manage the "application aspects" of VNFs and Services.  These Service Providers should have this option as ONAP was envisioned to be able to do just this, such as perform application level configuration of VNFs within the context of their Services, and to perform control loop functionality on application level telemetry.  If the provided ETSI Descriptors do not formally contain information as to the application management needs of the NS/NFs, a Service Provider that onboarded such descriptors into SDC would reasonably expect to have to provide substantial additional information content to SDC regarding the "application management aspects" of the "extended" Service and Network Functions. If the the ETSI Descriptor does formally include information as to the application management needs of the NS/NFs, a Service Provider that onboarded such descriptors into SDC would reasonably expect to have SDC accept the information regarding the "application management aspects" of the NS/NFs.

  6. Even in the case that the Service Provider chooses to leverage ONAP's "application configuration" capabilities, that does not mean that no additional configuration of the corresponding NFs will take place "out of band", outside of the purview of ONAP.  For example, after ONAP has completed its application level configuration of the VNF, it may be the case that another system or person would access the VNF to perform additional configuration, e.g., via another system's direct configuration API to the VNF, or via a URL/GUI that the VNF exposes directly as a user portal.  In such a case, ONAP should remain unaware of and unconcerned (in a separation of concerns sense) with whether such configuration should or should not take place, and what specific information should be or has been configured into such VNF.

*** Summary of the following: ONAP "knows what it knows" and nothing more.*** 

  1. ONAP should support the ETSI construct of "nested Services", such that one ServiceA is composed of another "inner" ServiceB.  ONAP should support this construct irrespective of whether the Service Provider chose to have ONAP manage the "application aspects" of both the "inner" and "outer" Service, just one of the same, or neither.  This affords the Service Provider with maximum flexibility.

  2. In addition to this, ONAP should support the ability to support Service nesting patterns such that management of an "inner" Service is delegated to an external manager via a southbound interface.  ONAP should support various external manager types, including patterns where the external manager is another ONAP instance (e.g., from a partner Service Provider) or an ETSI NFVO.

  3. In order to maintain separation of concerns in such a delegation pattern, ONAP would view an "inner" nested Service as being opaque, delegating all aspects of internal management to the external manager.  For example, ONAP would understand how to interact with the peer external manager to request an instance of such a delegated "inner" Service, and ONAP would know what input parameters must be supplied to the external manager as part of this request.  However, ONAP should not understand how those input parameters are used by the external manager to configure that "inner" Service.  Certainly ONAP should remain unaware of and unconcerned with whether an additional configuration should or should not take place outside the purview of this external manager, whether by another system's direct configuration API or via a URL/GUI exposed as a user portal.  ONAP's responsibility for instantiation/configuration of such a delegated Service would end with successful handoff to the external manager.

  4. This separation of concerns approach should be such as to allow Service Provider to migrate management of the delegated "inner" nested Service from one external manager type to another, as long as the externally visible aspects of the "inner" nested Service (e.g., the input parameters required to request an instance of such Service) remain unchanged.  E.g., a Service Provider with an ETSI NFVO/VNFM installation may want to migrate management of a NS to ONAP, or vice versa.  To avoid complexity of such migration, our goal should be an architecture such that ONAP interactions across its southbound API with various external managers are adaptable to the various supported external managers. 

  5. In light of the separation of concerns delegation approach described above, and in order to preserve the flexibility to switch between external managers, the interactions between ONAP and such adaptors on its southbound interface should be fundamentally the same irrespective of whether the external manager is an ONAP that is fully responsible for all aspects of application configuration and management of the "inner" Service, an ONAP that is responsible for some aspects of application configuration and management with other aspects being handled "out of band", or an ONAP that is responsible only for managing the "deployment aspects" of the "inner" Service.   The last example being functionally equivalent to an ETSI NFVO/VNFM, it follows that ONAP interactions across its southbound API with an external manager should be fundamentally the same irrespective of whether the external manager is an ONAP or an NFVO.

2.2 Proposal

2.2.1 High Level Proposal

This option proposes to preserve the flexibility described in the Assumptions section by having ONAP SDC, when onboarding an ETSI VNF Descriptor, automatically capture the content thereof in an ONAP VNF model that assumes that ONAP will manage only the "deployment aspects" of that VNF.  Similarly, when onboarding an ETSI NS Descriptor, ONAP SDC would capture the content thereof in an ONAP Service model that assumes that ONAP will manage only the "deployment aspects" of that Service.  If the Service Provider intends that ONAP manage the application aspects of such a VNF/Service as well, the Service Provider would be expected to provide supplemental information into SDC that captures the "application management aspects" of the "extended" Service and Network Functions.  To limit the amount of manual work involved to supplement SDC, this proposal recommends that the ONAP community define a method through which the VNF vendor itself would capture the "application management aspects" of their VNF, and provide that to SDC as an attachment to the SOL004 package at onboarding time.

Alternative Text: This option proposes to preserve the flexibility described in the Assumptions section by having ONAP SDC, when onboarding an ETSI VNF Descriptor, automatically capture the content thereof in an ONAP VNF model that assumes that ONAP may manage both the "deployment aspects" and any included "application aspects" of that VNF.  Similarly, when onboarding an ETSI NS Descriptor, ONAP SDC would capture the content thereof in an ONAP Service model that assumes that ONAP may manage both the "deployment aspects" and the "application aspects" of that Service.  This proposal recommends that the SDC be able to extract the "application aspects" from the VNF Descriptor at onboarding time and that the ONAP community define a method through which the VNF vendor would provide the "application aspects" in a separate artifact in the SOL004 package that the SDC can ingest at onboarding time.

2.2.2 ONAP Runtime Processing for an Onboarded Network Service Whereby ONAP Manages Only the "Deployment Aspects" of that Service

2.2.2.1 Description 

One benefit of mapping an onboarded ETSI NS to the internal representation of an ONAP Service is that the ETSI NS can access the standard ONAP runtime functionality implemented or planned for support of ONAP Services.  This benefit will be presented in this section.

As described above in Assumption #4, this ONAP management pattern assumes that the Service Provider has onboarded the associated ETSI NSD/VNFDs into SDC without any substantial extensions to that information content being needed.   However, this scenario assumes that the Service Provider does want to leverage that ONAP functionality that is relevant to Network Services per se, but outside of the formally specified scope of ETSI NFV for both NFVO and VNFM.  I.e., the ability to perform "homing" on VNFs and to automatically make "network assignments". 

Specifically, for a "simple" (un-nested) Service, upon receipt of a service instantiation request:

  1. SO will decompose the Service (i.e., the translated NS) into its component VNFs which must be instantiated

  2. SO will call OOF, which will perform "homing" to determine cloud instance locations for each such VNF to be instantiated

  3. SO will perform the following steps for each VNF, performing cross-VNF orchestration in the proper sequence based on the relationships between the VNFs in that Service's/NS's topology template: 

    1. SO will call a Controller which will make any needed network assignments for that VNF

    2. SO can either:

      1. Call Multi-VIM to request instantiation of that VNF, passing in the corresponding "deployment data" values (i.e., either data values mapped from the input parameters received in the service instantiation request, or mapped from the output of the Controller network assignments interaction), or

      2. Call the SOL003 Adaptor, passing in the corresponding "deployment data" values, and requesting the subtending VNFM to instantiate that VNF.  (For details of this approach see ONAP – ETSI Alignment Proposals in Support of VNF Scaling and SO Plug-in Support for VNFM (SO VNFM Adapter).)

2.2.2.2 Model

There is no need for a separate object type in the ONAP internal model to capture the "Network Service", but rather the ONAP "Service" construct can be used to capture all ONAP Services.  In this scenario it would so happen that the metadata relating to the specific Service described above would contain no metadata pertaining to application management.  ONAP would interpret this lack of application management metadata as instructions that there is no application data management for ONAP to perform for this specific Service type.

2.2.2.3 Implications/Impacts

Having an ONAP Service type whose metadata contains no application management requirements for ONAP to perform causes no problems.  In fact, this is a common pattern in the current state of the ONAP art.

2.2.3 ONAP Runtime Processing for an Onboarded Network Service Whereby ONAP Also Manages the "Application Aspects" of that Service

2.2.3.1 Description

As described above in Assumption #5, this ONAP management pattern assumes that the Service Provider has onboarded the associated ETSI NSD/VNFDs into SDC, and then provided supplemental information into SDC that captures the "application management aspects" of the "extended" Service and VNFs, either manually via the SDC GUI or by uploading a document in the SOL004 package that captures this information in a formal manner.

Alternative Text: As described above in Assumption #5, this ONAP management pattern assumes that the Service Provider has onboarded the associated ETSI NSD/VNFDs into SDC, including the "application management aspects" of the Network Service and VNFs, either from the information in the VNF-D, manually via the SDC GUI or by uploading a document in the SOL004 package that captures this information in a formal manner.

This runtime pattern is the same as that specified in the above, with the following added step after SO having called either Multi-VIM or the SOL003 Adaptor:

c. SO will call the Application Controller which will configure that VNF as appropriate.

Alternative Text: SO will call the  SOL003 Adapter or Application Controller which will configure that VNF as appropriate.

2.2.3.2 Model

There is no need for a separate object type in the ONAP internal model to capture the "Network Service", but rather the ONAP "Service" construct can be used to capture all ONAP Services.  In this scenario it would so happen that the metadata relating to the specific Service described above does contain metadata pertaining to application management.  ONAP would perform the required application management functionality as described in that metadata.  Specifically:

  • SO would note the presence of application configuration data in the metadata, and interpret this as a requirement to call the appropriate Application Controller after the "Create" of the VNF.

  • Application Controller would use the application configuration metadata to:

    • Determine what application data parameters to be configured for this VNF type in the context of the Service in question

    • Determine the appropriate instance values to set for each such  application data parameter.   Application Controller would use the data sourcing and mapping rules (perhaps generic, or perhaps via custom microservices embedded in the distributed Service TOSCA itself) provided in the metadata to derive each such instance value.

    • Determine the southbound API to call to set each such value, and call said API.  Note that this southbound API could be a NetConf, Ansible, Chef, or other API, some of which may require special adaptors.

    • Alternative Text: Determine the southbound API to call to set each such value, and call said API.  Note that this southbound API could be a NetConf, Ansible, Chef, SOL003, or other API.  Some such APIs may require their own adaptors; for example, the Application Controller itself would not call the SOL003 API to the VNFM, but rather would call the VNFM Adaptor with the appropriate application level data parameter values.  The VNFM Adaptor would map these into the appropriate SOL003 API calls. 

2.2.3.3 Implications/Impacts

In all cases whereby ONAP SDC onboards a SOL001 NSD, SDC should preserve that original NSD as an attachment to the ONAP Service as a reference document.

Having an ONAP Service type whose metadata contains application management requirements for ONAP to perform causes no problems as this is the agreed upon target.

Alternative Text:  Having an ONAP Service type whose metadata contains application management requirements for ONAP to perform causes no problems as this is the agreed upon target.  If a VNFM is to be used in the application level configuration of the VNF, then the VNFM would have received the application level configuration data as part of the VNFD distribution to the VNFM.

2.2.4 ONAP Runtime Processing for an Onboarded and "Nested" Network Service the Management of Which is Delegated to an NFVO as an External Manager

2.2.4.1 Description

As described in Assumption #8 and other associated assumptions above, one can intuit that this ONAP management pattern would assume that the Service Provider has onboarded the NSD and associated VNFDs into SDC without providing any supplemental information into SDC.

Next the Service Provider has defined another ONAP Service in SDC (i.e., an "outer Service") and included the formerly described Service as a component therein as an "inner Service".  Presumably the "outer Service" would also include other components, either VNFs or other (also "inner") nested Services with which we are not interested.

If it is desired that ONAP provide "homing" and "assignments" functionality for the "inner Service" in the context of this "outer Service", then the Service Provider would capture the corresponding information and policies in SDC as well.

Of course, it will also be necessary to load the original NSD and associated VNFDs into the NFVO, with ONAP's involvement in so doing TBD.

Upon receipt of a service instantiation request for this "outer" Service:

  1. SO will decompose the "outer" Service into its component VNFs and nested Services (including our "inner Service") which must be instantiated

  2. SO will call OOF, which will perform "homing" to determine cloud instance locations for each such VNF to be instantiated.  For a nested Service that is managed by ONAP itself (i.e., not delegated to an external manager), OOF could be responsible performing homing on the VNFs comprising that nested service in a recursive manner (see Modular Orchestration and Homing of Complex Services and Allotted Resources for more information).  For a nested Service that is managed by an external manager (as is our "inner Service") and hence "opaque" to ONAP, OOF would likely be responsible only for determining the desired location of the "Service Access Point" for that delegated Service.

  3. SO will perform the following steps for each VNF and nested Service, performing cross-VNF/nested Service orchestration in the proper sequence based on the relationships between the VNFs/nested Services in that ("outer") Service's topology template:

  4.  

    1. For each component VNF now homed, the "outer" Service level flow will spawn a VNF level sub-flow to perform the "assign", "create" and "configure" of that VNF as described above.

    2. For each component "nested" Service that is managed by ONAP itself (as opposed to an external manager), the "outer" Service level flow will spawn a separate Service level sub-flow to orchestrate that "nested" Service, passing in the appropriate modular structure within the homing solution returned by OOF.  This is a recursive pattern in that processing of this Service level sub-flow is another thread clone of that of the "outer" Service, with the exception that decomposition and homing need not be repeated due to the presence of the homing solution provided.  See Modular Orchestration and Homing of Complex Services and Allotted Resources for more information.

    3. For each component "nested" (Network) Service that is managed by an external manager, specifically for our "inner" Service, the "outer" Service level flow will spawn a separate sub-flow that will do the following:

    4.  

      1. SO will call a Controller which will make any needed network assignments for that "inner" Service as an opaque structure.  I.e., the Controller will see only that which is externally visible from the inner Service, such as its external connection points.

      2. SO will call the SOL005 Adaptor, passing in the corresponding input data values (i.e., either data values mapped from the input parameters received in the "outer" Service's service instantiation request, or mapped from the output of the Controller network assignments interaction).

Note that for the "inner" (Network) Service SO orchestrated only the "assign" and "create" functions, and not the (application level) "configure" of the VNFs that comprise the Network Service.  This is because to ONAP the (externally managed) Network Service is an opaque structure, and ONAP is not aware of the individual VNFs that comprise it.  Only the external manager (in this case the NFVO) is aware of the internal structure of the Network Service, the VNFs that comprise it.  If, however, the application level configuration of the Network Service could be accomplished without ONAP needing to be aware of the internal structure of the Network Service (e.g., if the collection of VNFs which comprise the NS can all be collectively configured through a single configuration API endpoint), then it is perhaps possible that ONAP (the Application Controller) could also perform the application layer configuration function.  Of course this would require that the appropriate model information would have been input into SDC, maintaining the proper abstractions.

2.2.4.2 Model

There would need to be something in the ONAP internal model that captures the fact that a particular Service is "externally managed", and we may find it useful to define a new object type in the ONAP internal model to represent such an "externally managed" Service.  .  But because there will likely be numerous targets for such "externally managed" Services (e.g., another ONAP instance, an ETSI NFVO, an "as a Service" external provider, etc.), we should not consider having a different object type for each such target.  Rather, a single normalized "externally managed Service" object that relies on Adaptors per external manager target to adapt from the normalized ONAP internal model to the model exposed by the external API of that target manager.  Thus, from the ONAP internal model perspective, an externally managed Service managed by an NFVO via a SOL005 API would look like just another "externally managed Service", though in this specific instance it so happens that the target manager is hidden behind a "SOL005 Adaptor". 

2.2.4.3 Implications/Impacts 

In all cases whereby ONAP SDC onboards a SOL001 NSD, SDC should preserve that original NSD as an attachment to the ONAP Service as a reference document.  If that ONAP Service is considered an "externally managed Service" then the set of additional data to be populated by the Service Designer would be different than that of an ONAP Service that is managed by ONAP itself.  This fact could justify that such an externally managed Service be represented as a new "type" in ONAP. 

The Service designer would need to indicate in the externally managed Service metadata what "target manager type" will manage this Service.  This information would be used for ONAP to determine the appropriate Adaptor type that should be used for this externally managed Service.  ONAP would distribute the associated ONAP internal metadata model for that Service to all instance(s) of that Adaptor type.  By doing so the SOL005 Adaptor would gain access to the original SOL001 NSD.

The Service Designer would need to also ensure the presence in the model of the information that ONAP OOF requires for "homing" such an externally managed Service, such homing perhaps being used to determine which target manager instance ONAP should call and the data to be passed in the calling.  This would allow ONAP runtime to determine the specific Adaptor instance to which to communicate.

3 Option B: Explict NSD representation

3.1 Assumptions

Can refer to above is necessary 

3.2 Proposal

3.2.1 Main proposal

3.2.2 Model

3.3.3 Implications/Impacts 

4 Analysis and comparison

5 References

  1.  Presentation on implicit representation of the NSD at the dublin architecture F2F: https://wiki.onap.org/download/attachments/41422290/Proposed%20ONAP%20Runtime%20Support%20for%20Network%20Service%20Onboarding%20v4.pptx?version=1&modificationDate=1540903774000&api=v2