| Network Working Group | D. Trossen |
| Internet-Draft | InterDigital Europe, Ltd |
| Intended status: Informational | D. Purkayastha |
| Expires: November 27, 2019 | A. Rahman |
| InterDigital Communications, LLC | |
| May 26, 2019 |
Name-Based Service Function Forwarder (nSFF) component within SFC framework
draft-trossen-sfc-name-based-sff-07
Adoption of cloud and fog technology allows operators to deploy a single “Service Function” to multiple “Execution locations”. The decision to steer traffic to a specific location may change frequently based on load, proximity etc. Under the current SFC framework, steering traffic dynamically to the different execution end points require a specific ‘re-chaining’, i.e., a change in the service function path reflecting the different IP endpoints to be used for the new execution points. This procedure may be complex and take time. In order to simplify re-chaining and reduce the time to complete the procedure, we discuss separating the logical Service Function Path from the specific execution end points. This can be done by identifying the Service Functions using a name rather than a routable IP endpoint (or Layer 2 address). This document describes the necessary extensions, additional functions and protocol details in SFF (Service Function Forwarder) to handle name based relationships.
This document presents InterDigital's approach to name-based service function chaining. It does not represent IETF consensus and is presented here so that the SFC community may benefit from considering this mechanism and the possibility of its use in the edge data centers.
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."
This Internet-Draft will expire on November 27, 2019.
Copyright (c) 2019 IETF Trust and the persons identified as the document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.
The requirements on today’s networks are very diverse, enabling multiple use cases such as IoT, Content Distribution, Gaming and Network functions such as Cloud RAN and 5G control planes based on a service-based architecture. These services are deployed, provisioned and managed using Cloud based techniques as seen in the IT world. Virtualization of compute and storage resources is at the heart of providing (often web) services to end users with the ability to quickly provisioning such virtualized service endpoints through, e.g., container based techniques. This creates a dynamicity with the capability to dynamically compose new services from available services as well as move a service instance in response to user mobility or resource availability where desirable. When moving from a pure ‘distant cloud’ model to one of localized micro data centers with regional, metro or even street level, often called ‘edge’ data centers, such virtualized service instances can be instantiated in topologically different locations with the overall ‘distant’ data center now being transformed into a network of distributed ones. The reaction of content providers, like Facebook, Google, NetFlix and others, are not just relying on deploying content server at the ingress of the customer network. Instead the trend is towards deploying multiple POPs within the customer network, those POPs being connected through proprietary mechanisms [Schlinker2017] to push content.
The Service Function Chaining (SFC) framework [RFC7665] allows network operators as well as service providers to compose new services by chaining individual "Service Functions (SFs)". Such chains are expressed through explicit relationships of functional components (the service functions), realized through their direct Layer 2 (e.g., MAC address) or Layer 3 (e.g., IP address) relationship as defined through next hop information that is being defined by the network operator, see Section 4 for more background on SFC.
In a dynamic service environment of distributed data centers as the one outlined above, with the ability to create and recreate service endpoints frequently, the SFC framework requires to reconfigure the existing chain through information based on the new relationships, causing overhead in a number of components, specifically the orchestrator that initiates the initial service function chain and any possible reconfiguration.
This document describes how such changes can be handled without involving the initiation of new and reconfigured SFCs by lifting the chaining relationship from Layer 2 and 3 information to that of service function ‘names’, such as names for instance being expressed as URIs. In order to transparently support such named relationships, we propose to embed the necessary functionality directly into the Service Function Forwarder (SFF), as described in [RFC7665]). With that, the SFF described in this document allows for keeping an existing SFC intact, as described by its service function path (SFP), while enabling the selection of an appropriate service function endpoint(s) during the traversal of packets through the SFC. This document is an Independent Submission to the RFC Editor. It is not an output of the IETF SFC WG.
The key words “MUST”, “MUST NOT”, “REQUIRED”, “SHALL”, “SHALL NOT”, “SHOULD”, “SHOULD NOT”, “RECOMMENDED”, “NOT RECOMMENDED”, “MAY”, and “OPTIONAL” in this document are to be interpreted as described in BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here.
We exemplify the need for chaining service functions at the level of a service name through a use case stemming from the current 3GPP Rel 16 work on Service Based Architecture (SBA) [_3GPP_SBA], [_3GPP_SBA_ENHANCEMENT]. In this work, mobile network control planes are proposed to be realized by replacing the traditional network function interfaces with a fully service-based one. HTTP was chosen as the application layer protocol for exchanging suitable service requests [_3GPP_SBA]. With this in mind, the exchange between, say the 3GPP (Rel. 15) defined Session Management Function (SMF) and the Access and Mobility management Function (AMF) in a 5G control plane is being described as a set of web service like requests which are in turn embedded into HTTP requests. Hence, interactions in a 5G control plane can be modelled based on service function chains where the relationship is between the specific (IP-based) service function endpoints that implement the necessary service endpoints in the SMF and AMF. The service functions are exposed through URIs with work ongoing to define the used naming conventions for such URIs.
This move from a network function model (in pre-Rel 15 systems of 3GPP) to a service-based model is motivated through the proliferation of data center operations for mobile network control plane services. In other words, typical IT-based methods to service provisioning, in particular that of virtualization of entire compute resources, are envisioned to being used in future operations of mobile networks. Hence, operators of such future mobile networks desire to virtualize service function endpoints and direct (control plane) traffic to the most appropriate current service instance in the most appropriate (local) data centre, such data centre envisioned as being interconnected through a software-defined wide area network (SD-WAN). ‘Appropriate’ here can be defined by topological or geographical proximity of the service initiator to the service function endpoint. Alternatively, network or service instance compute load can be used to direct a request to a more appropriate (in this case less loaded) instance to reduce possible latency of the overall request. Such data center centric operation is extended with the trend towards regionalization of load through a ‘regional office’ approach, where micro data centers provide virtualizable resources that can be used in the service execution, creating a larger degree of freedom when choosing the ‘most appropriate’ service endpoint for a particular incoming service request.
While the move to a service-based model aligns well with the framework of SFC, choosing the most appropriate service instance at runtime requires so-called ‘re-chaining’ of the SFC since the relationships in said SFC are defined through Layer 2 or 3 identifiers, which in turn are likely to be different if the chosen service instances reside in different parts of the network (e.g., in a regional data center).
Hence, when a traffic flow is forwarded over a service chain expressed as an SFC-compliant Service Function Path (SFP), packets in the traffic flow are processed by the various service function instances, with each service function instance applying a service function prior to forwarding the packets to the next network node. It is a Service layer concept and can possibly work over any Virtual network layer and an Underlay network, possibly IP or any Layer 2 technology. At the service layer, Service Functions are identified using a path identifier and an index. Eventually this index is translated to an IP address (or MAC address) of the host where the service function is running. Because of this, any change of service function instance is likely to require a change of the path information since either IP address (in the case of changing the execution from one data centre to another) or MAC address will change due to the newly selected service function instance.
Returning to our 5G Control plane example, a user’s connection request to access an application server in the internet may start with signaling in the Control Plane to setup user plane bearers. The connection request may flow through service functions over a service chain in the Control plane, as deployed by network operator. Typical SFs in a 5G control plane may include “RAN termination / processing”, “Slice Selection Function”, “AMF” and “SMF”. A Network Slice is a complete logical network including Radio Access Network (RAN) and Core Network (CN). Distinct RAN and Core Network Slices may exist. A device may access multiple Network Slices simultaneously through a single RAN. The device may provide Network Slice Selection Assistance Information (NSSAI) parameters to the network to help it select a RAN and a Core network part of a slice instance. Part of the control plane, the Common Control Network Function (CCNF), the Network Slice Selection Function (NSSF) is in charge of selecting core Network Slice instances. The Classifier, as described in SFC architecture, may reside in the user terminal or at the eNB. These service functions can be configured to be part of a Service Function Chain. We can also say that some of the configurations of the Service Function Path may change at the execution time. For example, the SMF may be relocated as user moves and a new SMF may be included in the Service Function Path based on user location. The following diagram in Figure 1 shows the example Service Function Chain described here.
+------+ +---------+ +-----+ +-----+
| User | | Slice | | | | |
| App |-->| Control |->| AMF |-->| SMF |-->
| Fn | | Function| | | | |
+------+ +---------+ +-----+ +-----+
Figure 1: Mapping SFC onto Service Function Execution Points along a Service Function Path
[RFC7665] describes an architecture for the specification, creation and ongoing maintenance of Service Function Chains (SFCs). It includes architectural concepts, principles, and components used in the construction of composite services through deployment of SFCs. In the following, we outline the parts of this SFC architecture relevant for our proposed extension, followed by the challenges with this current framework in the light of our example use case.
SFC Architecture, as defined in [RFC7665], describes architectural components such as Service Function (SF), Classifier, and Service Function Forwarder (SFF). It describes the Service Function Path (SFP) as the logical path of an SFC. Forwarding traffic along such SFP is the responsibility of the SFF. For this, the SFFs in a network maintain the requisite SFP forwarding information. Such SFP forwarding information is associated with a service path identifier (SPI) that is used to uniquely identify an SFP. The service forwarding state is represented by the Service Index (SI) and enables an SFF to identify which SFs of a given SFP should be applied, and in what order. The SFF also has information that allows it to forward packets to the next SFF after applying local service functions.
The operational steps to forward traffic are then as follows: Traffic arrives at an SFF from the network. The SFF determines the appropriate SF the traffic should be forwarded to via information contained in the SFC encapsulation. After SF processing, the traffic is returned to the SFF, and, if needed, is forwarded to another SF associated with that SFF. If there is another non-local hop (i.e., to an SF with a different SFF) in the SFP, the SFF further encapsulates the traffic in the appropriate network transport protocol and delivers it to the network for delivery to the next SFF along the path. Related to this forwarding responsibility, an SFF should be able to interact with metadata.
As outlined in previous section, the Service Function Path defines an ordered sequence of specific Service Functions instances being used for the interaction between initiator and service functions along the SFP. These service functions are addressed by IP (or any L2/MAC) addresses and defined as next hop information in the network locator maps of traversing SFF nodes.
As outlined in our use case, however, the service provider may want to provision SFC nodes based on dynamically spun up service function instances so that these (now virtualized) service functions can be reached in the SFC domain using the SFC underlay layer.
Following the original model of SFC, any change in a specific execution point for a specific Service Function along the SFP will require a change of the SFP information (since the new service function execution point likely carries different IP or L2 address information) and possibly even the Next Hop information in SFFs along the SFP. In case the availability of new service function instances is rather dynamic (e.g., through the use of container-based virtualization techniques), the current model and realization of SFC could lead to reducing the flexibility of service providers and increasing the management complexity incurred by the frequent changes of (service) forwarding information in the respective SFF nodes. This is because any change of the SFP (and possibly next hop info) will need to go through suitable management cycles.
To address these challenges through a suitable solution, we identify the following requirements:
The next section outlines a solution to address the issue, allowing for keeping SFC information (represented in its SFP) intact while addressing the desired flexibility of the service provider.
The general idea is two-pronged. Firstly, we elevate the definition of a Service Function Path onto the level of ‘name-based interactions’ rather than limiting SFPs to Layer 3 or Layer 2 information only. Secondly, we extend the operations of the SFF to allow for forwarding decisions that take into account such name-based interaction while remaining backward compatible to the current SFC architecture, as defined in [RFC7665]. In the following sections, we outline these two components of our solution.
If the next hop information in the Network Locator Map (NLM) is described using L2/L3 identifier, the name-based SFF (nSFF) may operate as described for [traditional] SFF, as defined in [RFC7665]. On the other hand, if the next hop information in the NLM is described as a name, then the nSFF operates as described in the following sections.
In the following sections, we outline the two components of our solution.
The existing SFC framework is defined in [RFC7665]. Section 4 outlines that the SFP information is representing path information based on Layer 2 or 3 information, i.e., MAC or IP addresses, causing the aforementioned frequent adaptations in cases of execution point changes. Instead, we introduce the notion of a "name-based service function path (nSFP)".
In today’s networking terms, any identifier can be treated as a name but we will illustrate the realization of a “Name based SFP” through extended SFF operations (see Section 6) based on URIs as names and HTTP as the protocol of exchanging information. Here, URIs are being used to name for a Service Function along the nSFP. It is to be noted that the Name based SFP approach is not restricted to HTTP (as the protocol) and URIs (as next hop identifier within the SFP). Other identifiers such as an IP address itself can also be used and are interpreted as a ‘name’ in the nSFP. IP addresses as well as fully qualified domain names forming complex URIs (uniform resource identifiers), such as www.example.com/service_name1, are all captured by the notion of ‘name’ in this document.
Generally, nSFPs are defined as an ordered sequence of the “name” of Service Functions (SF) and a typical name-based Service Function Path may look like: 192.0.x.x -> www.example.com -> www.example2.com/service1 -> www.example2.com/service2.
Our use case in Section 3 can then be represented as an ordered named sequence. An example for a session initiation that involves an authentication procedure, this could look like 192.0.x.x -> smf.example.org/session_initiate -> amf.example.org/auth -> smf.example.org/session_complete -> 192.0.x.x. [Note that this example is only a conceptual one, since the exact nature of any future SBA-based exchange of 5G control plane functions is yet to be defined by standardization bodies such as 3GPP].
In accordance with our use case in Section 3, any of these named services can potentially be realized through more than one replicated SF instances. This leads to make dynamic decision on where to send packets along the SAME service function path information, being provided during the execution of the SFC. Through elevating the SFP onto the notion of name-based interactions, the SFP will remain the same even if those specific execution points change for a specific service interaction.
+---------------------------------------------------------------+
|SERVICE LAYER |
| 192.0.x.x --> www.example.com --> www.example2.com --> |
| || || |
+----------------------||--------------||-----------------------+
|| ||
|| ||
+----------------------||--------------||-----------------------+
| Underlay network \/ \/ |
| +--+ +--+ +--+ +--+ +--+ +--+ |
| | | | | | | | | | | | | |
| +--+ +--+ +--+ +--+ +--+ +--+ |
| Compute and Compute and |
| storage nodes storage nodes |
+---------------------------------------------------------------+
Figure 2: Mapping SFC onto Service Function Execution Points along a Service Function Path based on Virtualized Service Function Instance
The following diagram in Figure 2, describes this name-based SFP concept and the resulting mapping of those named interactions onto (possibly) replicated instances.
In order to forward a packet within a name-based SFP, we need to extend the network locator map as defined in [RFC8300] with the ability to consider name relations based on URIs as well as high-level transport protocols such as HTTP for means of SFC packet forwarding. Another example for SFC packet forwarding could be that of CoAP.
The extended Network Locator Map or name-based Network Locator Map (nNLM) is shown in Figure 3 as an example for www.example.com being part of the nSFP. Such extended nNLM is stored at each SFF throughout the SFC domain with suitable information populated to the nNLM during the configuration phase.
+------+------+---------------------+-----------------------------+
| SPI | SI | Next Hop(s) | Transport Encapsulation (TE)|
+------+------+---------------------+-----------------------------+
| 10 | 255 | 192.0.2.1 | VXLAN-gpe |
| | | | |
| 10 | 254 | 198.51.100.10 | GRE |
| | | | |
| 10 | 253 | www.example.com | HTTP |
-----------------------------------------------------------------
| | | | |
| 40 | 251 | 198.51.100.15 | GRE |
| | | | |
| 50 | 200 | 01:23:45:67:89:ab | Ethernet |
| | | | |
| 15 | 212 | Null (end of path) | None |
+------+------+---------------------+-----------------------------+
Figure 3: Name-based Network Locator Map
Alternatively, the extended network locator map may be defined with implicit name information rather than explicit URIs as in Figure 3. In the example of Figure 4 below, the next hop is represented as a generic HTTP service without a specific URI being identified in the extended network locator map. In this scenario, the SFF forwards the packet based on parsing the HTTP request in order to identify the host name or URI. It retrieves the URI and may apply policy information to determine the destination host/service.
+------+------+---------------------+-----------------------------+
| SPI | SI | Next Hop(s) | Transport Encapsulation (TE)|
+------+------+---------------------+-----------------------------+
| 10 | 255 | 192.0.2.1 | VXLAN-gpe |
| | | | |
| 10 | 254 | 198.51.100.10 | GRE |
| | | | |
| 10 | 253 | HTTP Service | HTTP |
-----------------------------------------------------------------
| | | | |
| 40 | 251 | 198.51.100.15 | GRE |
| | | | |
| 50 | 200 | 01:23:45:67:89:ab | Ethernet |
| | | | |
| 15 | 212 | Null (end of path) | None |
+------+------+---------------------+-----------------------------+
Figure 4: Name-based Network Locator Map with Implicit Name information
It is desirable to extend the SFF of the SFC underlay to handle nSFPs transparently and without the need to insert any service function into the nSFP. Such extended name-based SFF would then be responsible for forwarding a packet in the SFC domain as per the definition of the (extended) nSFP.
In our exemplary realization for an extended SFF, the solution described in this document uses HTTP as the protocol of forwarding SFC packets to the next (name-based) hop in the nSFP. The URI in the HTTP transaction are the names in our nSFP information, which will be used for name based forwarding.
Following our reasoning so far, HTTP requests (and more specifically the plain text encoded requests above) are the equivalent of Packets that enter the SFC domain. In the existing SFC framework, typically an IP payload is assumed to be a packet entering the SFC domain. This packet is forwarded to destination nodes using the L2 encapsulation. Any layer 2 network can be used as an underlay network. This notion is now extended to packets being possibly part of a entire higher layer application, such as HTTP requests. The handling of any intermediate layers such as TCP, IP is left to the realization of the (extended) SFF operations towards the next (named) hop. For this, we will first outline the general lifecycle of an SFC packet in the following subsection, followed by two examples for determining next hop information in Section 6.2.3, finalized by a layered view on the realization of the nSFF in Section 6.2.4.
+----------+
| SF1 | +--------+ +------+
| instance |\ | NR | | SF2 |
+----------+ \ +--------+ +------+
\ || ||
+------------+ \ +-------+ +---------+ +---------+ +-------+
| Classifier |---| nSFF1 |---|Forwarder|---|Forwarder|---| nSFF2 |
+------------+ +-------+ +---------+ +---------+ +-------+
||
+----------+
| Boundary |
| node |
+----------+
Figure 5: High-level architecture
The high-level architecture for name based operation shown in Figure 5 is very similar to the SFC architecture, as described in [RFC7665]. Two new functions are introduced, as shown in the above diagram, namely the name-based Service Function Forwarder (nSFF) and the Name Resolver (NR).
nSFF (name-based Service Function Forwarder) is an extension of the existing SFF and is capable of processing SFC packets based on name-based network locator map (nNLM) information, determining the next SF, where the packet should be forwarded and the required transport encapsulation. Like standard SFF operation, it adds transport encapsulation to the SFC packet and forwards it.
The Name Resolver is a new functional component, capable of identifying the execution end points, where a “named SF” is running, triggered by suitable resolution requests sent by the nSFF. Though this is similar to DNS function, but it is not same. It does not use DNS protocols or data records. A new procedure to determine the suitable routing/forwarding information towards the Nsff (name-based SFF) serving the next hop of the SFP (Service Function Path) is used. The details is described later.
The other functional components such as Classifier, SF are same as described in SFC architecture, as defined in [RFC7665], while the Forwarders shown in the above diagram are traditional Layer 2 switches.
In the proposed solution, the operations are realized by the name-based SFF, called nSFF. We utilize the high-level architecture in Figure 5 to describe the traversal between two service function instances of an nSFP-based transactions in an example chain of : 192.0.x.x -> SF1 (www.example.com) -> SF2 (www.example2.com) -> SF3 -> ... Service Function 3 (SF3)is assumed to be a classical Service Function, hence existing SFC mechanisms can be used to reach it and will not be considered in this example.
According to the SFC lifecycle, as defined in [RFC7665], based on our example chain above, the traffic originates from a Classifier or another SFF on the left. The traffic is processed by the incoming nSFF1 (on the left side) through the following steps. The traffic exits at nSFF2.
+------+------+---------------------+----------------------------+
| SPI | SI | Next Hop(s) | Transport Encapsulation(TE)|
+------+------+---------------------+----------------------------+
| 10 | 255 | 192.0.2.1 | VXLAN-gpe |
| | | | |
| 10 | 254 | 198.51.100.10 | GRE |
| | | | |
| 10 | 253 | www.example.com | HTTP |
| | | | |
| 10 | 252 | www.example2.com | HTTP |
| | | | |
| 40 | 251 | 198.51.100.15 | GRE |
| | | | |
| 50 | 200 | 01:23:45:67:89:ab | Ethernet |
| | | | |
| 15 | 212 | Null (end of path) | None |
+------+------+---------------------+----------------------------+
Figure 6: nNLM at nSFF1
+------+------+---------------------+-----------------------------+
| SPI | SI | Next Hop(s) | Transport Encapsulation (TE)|
+------+------+---------------------+-----------------------------+
| | | | |
| 10 | 252 | www.example2.com | HTTP |
| | | | |
| 40 | 251 | 198.51.100.15 | GRE |
| | | | |
| 50 | 200 | 01:23:45:67:89:ab | Ethernet |
| | | | |
| 15 | 212 | Null (end of path) | None |
+------+------+---------------------+-----------------------------+
Figure 7: nNLM at SFF2
This section outlines the realization of various nSFF forwarding operations in Section 5.6. Although the operations in Section 5 utilize the notion of name-based transactions in general, we exemplify the operations here in Section 5 specifically for HTTP-based transactions to ground our description into a specific protocol for such name-based transaction. We will refer to the various steps in each of the following sub-sections.
Figure 8 shows the protocol layers, based on the high-level architecture in Figure 5.
+-------+ +------+----+ +----+-----+
|App | | | | +--------+ | | |
|HTTP | |--------> | | NR | |nSFF----->|--
|TCP |->| TCP |nSFF| +---/\---+ | | TCP | |
|IP | | IP | | || | | IP | |
+-------+ +------+----+ +---------+ +---------+ +----------+ |
| L2 | | L2 |->|Forwarder|-->|Forwarder|-->| L2 | |
+-------+ +------+----+ +---------+ +---------+ +----------+ |
SF1 nSFF1 nSFF2 |
+-------+ |
| App |/ |
| HTTP | -----------+
| TCP |\
| IP |
| L2 |
+-------+
SF2
Figure 8: Protocol layers
The nSFF component here is shown as implementing a full incoming/outgoing TCP/IP protocol stack towards the local service functions, while implementing the nSFF-NR and nSFF-nSFF protocols based on the descriptions in Section 6.2.3.
For the exchange of HTTP-based service function transactions, the nSFF terminates incoming TCP connections from as well as outgoing TCP connections to local SFs, e.g., the TCP connection from SF1 terminates at nSFF1, and nSFF1 may store the connection information, such as socket information. It also maintains the mapping information for the HTTP request such as originating SF, destination SF and socket ID. nSFF1 may implement sending keep-alive messages over the socket to maintain the connection to SF1. Upon arrival of an HTTP request from SF1, nSFF1 extracts the HTTP Request and forwards it towards the next node, as outlined in Section 6.2. Any returning response is mapped onto the suitable open socket (for the original request) and send towards SF1.
At the outgoing nSFF2, the destination SF2/Host is identified from the HTTP request message. If no TCP connection exists to the SF2, a new TCP connection is opened towards the destination SF2 and the HTTP request is sent over said TCP connection. The nSFF2 may also save the TCP connection information (such as socket information) and maintain the mapping of the socket information to the destination SF2. When an HTTP response is received from SF2 over the TCP connection, nSFF2 extracts the HTTP response, which is forwarded to the next node. nSFF2 may maintain the TCP connection through keep-alive messages.
In this section, we present three key aspects of operations for the realization of the steps in Section 5.6, namely (i) the registration of local SFs (for step 3 in Section 5.6), (ii) the forwarding of SFC packets to and from local SFs (for step 3 and 4 as well as 10 in Section 5.6), (iii) the forwarding to a remote SF (for steps 5, 6 and 7 in Section 5.6) and to the NR as well as (iv) for the lookup of a suitable remote SF (for step 7 in Section 5.6). We also cover aspects of maintaining local lookup information for reducing lookup latency and others issues.
Forwarding between the distributed nSFFs as well as between nSFF and NR is realized over the operator network via a path-based approach. A path-based approach utilizes path information provided by the source of the packet for forwarding said packet in the network. This is similar to segment routing albeit differing in the type of information provided for such source-based forwarding, as described in this section. In this approach, the forwarding information to a remote nSFF or the NR is defined as a ‘path identifier’ (pathID) of a defined length where said "Length" field indicates the full pathID length. The payload of the packet is defined by the various operations outlined in the following sub-sections, resulting in an overall packet being transmitted. With this, the generic forwarding format (GFF) for transport over the operator network is defined in Figure 9 with the length field defining the length of the pathID provided.
+---------+-----------------+------------------------//------------+
| | | // |
| Length | Path ID | Payload // |
|(12 bit) | | // |
+---------+-----------------+--------------------//----------------+
Figure 9: Generic Forwarding Format(GFF)
For the pathID information, solutions such as those in [Reed2016] can be used. Here, the IPv6 source and destination addresses are used to realize a so-called path-based forwarding from the incoming to the outgoing nSFF or the NR. The forwarders in Figure 8 are realized via SDN (software-defined networking) switches, implementing an AND/CMP operation based on arbitrary wildcard matching over the IPv6 source and destination addresses, as outlined in [Reed2016]. Note that in the case of using IPv6 address information for path-based forwarding, the step of removing the transport encapsulation at the outgoing nSFF in Figure 8 is realized by utilizing the provided (existing) IP header (which was used for the purpose of the path-based forwarding in [Reed2016]) for the purpose of next hop forwarding, such as that of IP-based routing. As described in step 8 of the extended nSFF operations, this forwarding information is used as traffic encapsulation. With the forwarding information utilizing existing IPv6 information, IP headers are utilized as TE in this case. The next hop nSFF (see Figure 8) will restore the IP header of the packet with the relevant IP information used to forward the SFC packet to SF2 or it will create a suitable TE (Transport Encapsulation) information to forward the information to another nSFF or boundary node. Forwarding operations at the intermediary forwarders, i.e., SDN switches, examine the pathID information through a flow matching rule in which a specific switch-local output port is represented through the specific assigned bit position in the pathID. Upon a positive match in said rule, the packet is forwarded on said output port.
Alternatively, the solution in [I-D.ietf-bier-multicast-http-response] suggests using a so-called BIER (Binary Indexed Explicit Replication) underlay. Here, the nSFF would be realized at the ingress to the BIER underlay, injecting the SFC packet (plus the NSH) header with BIER-based traffic encapsulation into the BIER underlay with each of the forwarders in Figure 8 being realized as a so-called Bit-Forwarding Router (BFR) [RFC8279].
Given that the proposed solution operates at the 'named transaction' level, particularly for HTTP transactions, forwarding between nSFFs and/or NR SHOULD be implemented via a transport protocol between nSFFs and/or NR in order to provide reliability, segmentation of large GFF packets, and flow control, with the GFF in Figure 9 being the basic forwarding format for this.
Note that the nSFFs act as TCP proxies at ingress and egress, thus terminating incoming and initiating outgoing HTTP sessions to SFs.
Figure 10 shows the packet format being used for the transmission of data, being adapted from the TCP header. Segmentation of large transactions into single transport protocol packets is realized through maintaining a 'Sequence number'. A 'Checksum' is calculated over a single data packet with the ones-complement TCP checksum calculation being used. The 'Window Size' field indicates the current maximum number of transport packets that are allowed in-flight by the egress nSFF. A data packet is sent without 'Data' field to indicate the end of (e.g., HTTP) transaction.
Note that in order to support future named transactions based on other application protocols, such as CoAP, future versions of the transport protocol MAY introduce a 'Type' field that indicates the type of application protocol being used between SF and nSFF with 'Type' 0x01 proposed for HTTP. This is being left for future study.
| 16 bit | 16 bit |
+----------------------------------------------+
| Sequence number |
+----------------------------------------------+
| Checksum | Window Size |
+----------------------------------------------+
| ... |
| Data (Optional) |
+----------------------------------------------+
Figure 10: Transport protocol data packet format
Given the path-based forwarding being used between nSFFs, the transport protocol between nSFFs utilizes negative acknowledgements from the egress nSFF towards the ingress nSFF. The transport protocol NACK packet carries the number of NACKs as well as the specific sequence numbers being indicated as lost in the 'NACK number' field(s), as shown in Figure 11.
| 16 bit | 16 bit |
+----------------------------------------------+
| Number of NACKs | +
+----------------------------------------------+
| NACK number |
+----------------------------------------------+
+ ... NACK Number +
+----------------------------------------------+
Figure 11: Transport protocol NACK packet format
If the indicated number of NACKs in a received NACK packet in non-zero, the ingress nSFF will retransmit all sequence numbers signalled in the packet, while decreasing its congestion window size for future transmissions.
If the indicated number of NACKs in a received NACK packet in zero, it will indicate the current congestion window as being successfully (and completely) being transmitted, increasing the congestion window size if smaller than the advertised 'Window Size' in Figure 10.
The maintenance of the congestion window is subject to realization at the ingress nSFF and left for further study in nSFF realizations.
As outlined in step 3 and 10 of Section 5.6, the nSFF needs to determine if the SF derived from the nNLM is locally reachable or whether the packet needs forwarding to a remote SFF. For this, a registration mechanism is provided for such local SF with the local nSFF. Two mechanisms can be used for this:
1. SF-initiated: We assume that the SF registers its FQDN to the local nSFF. As local mechanisms, we foresee that either a REST-based interface over the link-local link or configuration of the nSFF (through configuration files or management consoles) can be utilized. Such local registration event leads to the nSFF to register the given FQDN with the NR in combination with a system-unique nSFF identifier that is being used for path computation purposes in the NR. For the registration, the packet format in Figure 12 is used (inserted as the payload in the GFF of Figure 9 with the pathID towards the NR).
+---------+-----------------+------------------+
| | | |
| R/D | hash(FQDN) | nSFF_ID |
| (1 bit) | (16 bit) | (8 bit) |
+---------+-----------------+------------------+
Figure 12: Registration packet format
We assume that the pathID towards the NR is known to the nSFF through configuration means.
The NR maintains an internal table that associates the hash(FQDN), the nSFF_id information as well as the pathID information being used for communication between nSFF and NR. The nSFF locally maintains a mapping of registered FQDNs to IP addresses, for the latter using link-local private IP addresses.
2. Orchestration-based: in this mechanism, we assume that SFC to be orchestrated and the chain being provided through an orchestration template with FQDN information associated to a compute/storage resource that is being deployed by the orchestrator. We also assume knowledge at the orchestrator of the resource topology. Based on this, the orchestrator can now use the same REST-based protocol defined in option 1 to instruct the NR to register the given FQDN, as provided in the template, at the nSFF it has identified as being the locally servicing nSFF, provided as the system-unique nSFF identifier.
There are two cases of local SF forwarding, namely the SF sending an SFC packet to the local nSFF (incoming requests) or the nSFF sending a packet to the SF (outgoing requests) as part of steps 3 and 10 in Section 5.6. In the following, we outline the operation for HTTP as an example named transaction.
As shown in Figure 8, incoming HTTP requests from SFs are extracted by terminating the incoming TCP connection at their local nSFFs at the TCP level. The nSFF MUST maintain a mapping of open TCP sockets to HTTP requests (utilizing the URI of the request) for HTTP response association.
For outgoing HTTP requests, the nSFF utilizes the maintained mapping of locally registered FQDNs to link-local IP addresses (see Section 6.2.2 option 1). Hence, upon receiving an SFC packet from a remote nSFF (in step 9 of Section 5.6), the nSFF determines the local existence of the SF through the registration mechanisms in Section 6.2.2. If said SF does exist locally, the HTTP (+NSH) packet, after stripping the TE, is sent to the local SF as step 10 in Section 5.6 via a TCP-level connection. Outgoing nSFF SHOULD keep TCP connections open to local SFs for improving SFC packet delivery in subsequent transactions.
When executing step 3 and 10 in Section 5.6, the SFC packet will be delivered to the locally registered next hop. As part of the HTTP protocol, responses to the HTTP request will need to be delivered on the return path to the originating nSFF (i.e., the previous hop). For this, the nSFF maintains a list of link-local connection information, e.g., sockets to the local SF and the pathID on which the request was received. Once receiving the response, nSFF consults the table to determine the pathID of the original request, forming a suitable GFF-based packet to be returned to the previous nSFF.
When receiving the HTTP response at the previous nSFF, the nSFF consults the table of (locally) open sockets to determine the suitable local SF connection, mapping the received HTTP response URI to the stored request URI. Utilizing the found socket, the HTTP response is forwarded to the locally registered SF.
In steps 5, 6, 7, and 8 of Section 5.6, an SFC packet is forwarded to a remote nSFF based on the nNLM information for the next hop of the nSFP. Section 6.2.5.1 handles the case of suitable forwarding information to the remote nSFF not existing, therefore consulting the NR to obtain suitable information, while Section 6.2.5.2 describes the maintenance of forwarding information at the local nSFF, while Section 6.2.5.3 describes the update of stale forwarding information. Note that the forwarding described in Section 6.2.1 is used for the actual forwarding to the various nSFF components. Ultimately, Section 6.2.5.4 describes the forwarding to the remote nSFF via the forwarder network.
The nSFF communicates with the NR for two purposes, namely the registration and discovery of FQDNs. The packet format for the former was shown in Figure 10 in Section 6.2.2, while Figure 13 outlines the packet format for the discovery request.
+--------------+-------------+ +--------+-----------------//--------+
| | | | | // |
| hash(FQDN) | nSFF_ID | | Length | pathID // |
| (16 bit) | (8 bit) | | (4 bit)| // |
+--------------+-------------+ +--------+-------------//------------+
Path Request Path Response
Figure 13: Discovery packet format
For Path Request:
For Path Response:
A path to a specific FQDN is requested by sending a hash of the FQDN to the NR together with its nSFF_id, receiving as a response a pathID with a length identifier. The NR SHOULD maintain a table of discovery requests that map discovered (hash of) FQDN to the nSFF_id that requested it and the pathID that is being calculated as a result of the discovery request.
The discovery request for an FQDN that has not previously been served at the nSFF (or for an FQDN whose pathID information has been flushed as a result of the update operations in Section 6.2.5.3), results in an initial latency incurred by this discovery through the NR, while any SFC packet sent over the same SFP in a subsequent transaction will utilize the nSFF local mapping table. Such initial latency can be avoided by pre-populating the FQDN-pathID mapping proactively as part of the overall orchestration procedure, e.g., alongside the distribution of the nNLM information to the nSFF.
Each nSFF MUST maintain an internal table that maps the (hash of the) FQDN information to a suitable pathID information. As outlined in step 7 of Section 5.6, if a suitable entry does not exist for a given FQDN, the pathID information is requested with the operations in Section 6.2.5.1 and the suitable entry is locally created upon receiving a reply with the forwarding operation being executed as described in Section 6.2.1.
If such entry does exist (i.e., step 6 of Section 5.6) the pathID is locally retrieved and used for the forwarding operation in Section 6.2.1.
The forwarding information maintained at each nSFF (see Section 6.2.5.2) might need to be updated for three reasons:
For this purpose, the packet format in Figure 14 is sent from the NR to all affected nSFFs, using the generic format in Figure 9.
+---------+-----------------+--------------//----+
| | | // |
| Type | #IDs | IDs // |
| (1 bit) | (8 bit) | // |
+---------+-----------------+----------//--------+
Figure 14: Path update format
The pathID to the affected nSFFs is computed as the binary OR over all pathIDs to those nSFF_ids affected where the pathID information to the affected nSFF_id values is determined from the NR-local table maintained in the registration/deregistration operation of Section 6.2.2.
The pathID may include the type of information being updated (e.g., node identifiers of leaf nodes or link identifiers for removed links). The node identifier itself may be a special identifier to signal “ALL NODES” as being affected. The node identifier may signal changes to the network that are substantial (e.g., parallel link failures). The node identifier may trigger (e.g., recommend) purging of the entire path table (e.g., rather than the selective removal of a few nodes only).
It will include the information according to the type. The included information may also be related to the type and length information for the number of identifiers being provided.
In case 1 and 2, the Type bit is set to 1 (type nSFF_id) and the affected nSFFs are determined by those nSFFs that have previously sent SF discovery requests, utilizing the optional table mapping previously registered FQDNs to nSFF_id values. If no table mapping the (hash of) FQDN to nSFF_id is maintained, the update is sent to all nSFFs. Upon receiving the path update at the affected nSFF, all appropriate nSFF-local mapping entries to pathIDs for the hash(FQDN) identifiers provided will be removed, leading to a new NR discovery request at the next remote nSFF forwarding to the appropriate FQDN.
In case 3, the Type bit is set to 0 (type linkID) and the affected nSFFs are determined by those nSFFs whose discovery requests have previously resulted in pathIDs which include the affected link, utilizing the optional table mapping previously registered FQDNs to pathID values (see Section 6.2.5.1). Upon receiving the node identifier information in the path update, the affected nSFF will check its internal table that maps FQDNs to pathIDs to determine those pathIDs affected by the link problems and remove path information that includes the received node identifier(s). For this, the pathID entries of said table are checked against the linkID values provided in the ID entry of the path update through a binary AND/CMP operation to check the inclusion of the link in the pathIDs to the FQDNs. If any pathID is affected, the FQDN-pathID entry is removed, leading to a new NR discovery request at the next remote nSFF forwarding to the appropriate FQDN.
Once step 5, 6, and 7 in Section 5.6 are being executed, step 8 finally sends the SFC packet to the remote nSFF, utilizing the pathID returned in the discovery request (Section 6.2.5.1) or retrieved from the local pathID mapping table. The SFC packet is placed in the payload of the generic forwarding format in Figure 9 together with the pathID and the nSFF eventually executes the forwarding operations in Section 6.2.1.
This document requests no IANA actions.
The operations in Sections 5 and 6 describes the forwarding of SFC packets between named SFs based on URIs exchanged in HTTP messages. For security considerations, TLS is sufficient between originating node and Nsff, Nsff to Nsff, Nsff to destination. TLS handshake allows to determine the FQDN, which in turn is enough for the service routing decision. Supporting TLS also allows the possibility of HTTPS based transactions.
The authors would like to thank Dirk von Hugo and Andrew Malis for their reviews and valuable comments. We would also like to thank Joel Halpern, the chair of the SFC WG, and Adrian Farrel for guiding us through the IETF Independent Submission Editor (ISE) path.
| [RFC2119] | Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, DOI 10.17487/RFC2119, March 1997. |
| [RFC7665] | Halpern, J. and C. Pignataro, "Service Function Chaining (SFC) Architecture", RFC 7665, DOI 10.17487/RFC7665, October 2015. |
| [RFC8174] | Leiba, B., "Ambiguity of Uppercase vs Lowercase in RFC 2119 Key Words", BCP 14, RFC 8174, DOI 10.17487/RFC8174, May 2017. |
| [RFC8279] | Wijnands, IJ., Rosen, E., Dolganow, A., Przygienda, T. and S. Aldrin, "Multicast Using Bit Index Explicit Replication (BIER)", RFC 8279, DOI 10.17487/RFC8279, November 2017. |
| [RFC8300] | Quinn, P., Elzur, U. and C. Pignataro, "Network Service Header (NSH)", RFC 8300, DOI 10.17487/RFC8300, January 2018. |
| [_3GPP_SBA] | 3GPP, "Technical Realization of Service Based Architecture", 3GPP TS 29.500 0.4.0, January 2018. |
| [_3GPP_SBA_ENHANCEMENT] | 3GPP, "New SID for Enhancements to the Service-Based 5G System Architecture", 3GPP S2-182904 , February 2018. |
| [I-D.ietf-bier-multicast-http-response] | Purkayastha, D., Rahman, A., Trossen, D. and T. Eckert, "Applicability of BIER Multicast Overlay for Adaptive Streaming Services", Internet-Draft draft-ietf-bier-multicast-http-response-00, February 2019. |
| [Reed2016] | Reed, M., Al-Naday, M., Thomas, N., Trossen, D. and S. Spirou, "Stateless multicast switching in software defined networks", ICC 2016, 2016. |
| [Schlinker2017] | Schlinker, B., Kim, H., Cui, T., Katz-Bassett, E., Madhyastha, Harsha., Cunha, I., Quinn, J., Hassan, S., Lapukhov, P. and H. Zeng, "Engineering Egress with Edge Fabric, Steering Oceans of Content to the World", ACM SIGCOMM 2017, 2017. |