Abstract

This document describes the CCNx protocol, the transport protocol for a communications architecture called Content-Centric Networking (CCN) built on named data. CCN has no notion of host at its lowest level — a packet "address" names content, not location. The CCNx protocol efficiently delivers named content rather than connecting hosts to other hosts. Every packet of data may be cached at any CCNx router — combined with intrinsic support for multicast or broadcast delivery this leads to a very efficient use of the network when many people are interested in the same content.

Introduction

This document describes the CCNx protocol which provides location-independent delivery services for named data packets. The services include multihop forwarding for end-to-end delivery, flow control, transparent and automatic multicast delivery using buffer storage available in the network, loop-free multipath forwarding, verification of content integrity regardless of delivery path, and carriage of arbitrary application data. Applications run the CCNx protocol over some lower-layer communications service capable of transmitting packets. There are no restrictions on the nature of the lower-layer service: it may be a physical transport or another network or transport protocol. For example, applications will typically run the CCNx protocol on top of UDP to take advantage of existing IP connectivity. Since content is named independent of location in the CCNx protocol, it may also be preserved indefinitely in the network, providing effectively a form of distributed filesystem service.

The CCNx protocol is general, supporting a wide range of network applications. It may be natural to think of stored content applications such as distribution of video or document files, but the CCNx model also supports real-time communication and discovery protocols and is general enough to carry conversations between hosts such as TCP connections. The protocol supports a broad range of applications by leaving the choice of naming conventions to the application. This document specifies the common functions that are independent of the contents of the names and data and the semantics of exchanges. In addition to this document, therefore, a complete specification of the use of the CCNx protocol for a particular application will require additional specification of the naming rules, data formats, and message semantics. Such specifications of application protocols on top of the CCNx protocol (plus possibly API specifications) are called profiles.

The CCNx protocol is designed for end-to-end communication between applications and so it is intended to be integrated into application processing rather than being implemented as a separate layer.

Definitions

node

a CCNx network entity that implements forwarding and buffering

party

any entity in the network using the CCNx protocol to communicate. Note that parties are not just machines, but also applications using the protocol.

message

a CCNx packet. We use this term to avoid confusion with the lower-layer packet that may be carrying CCNx messages. A single lower-layer packet (for example a single UDP packet) MAY contain more than one CCNx message.

Message Format and Encodings

Unlike many other protocols, the CCNx protocol does not have any fixed-length fields. Instead, CCNx data formats are defined by XML schemas and encoded with explicitly identified field boundaries. This design permits field values of arbitrary length, optional fields that consume no packet space when omitted, and nested structures. The use of XML structure does not imply that field values are text strings nor does it require that messages be encoded as human-readable text. Most fields are defined to contain arbitrary binary values, including those that identify content.

The wire format of CCNx messages is an efficient binary encoding of XML structures called ccnb, which defines such things as the byte order. There is also a text XML encoding which is useful for debugging, testing, presentation in documentation, etc. but MUST NOT be used on the wire.

Content Identification

The CCNx protocol accomplishes transfers of content by name, irrespective of the identities or locations of machines involved.

CCNx content names are hierarchically-structured, consisting of a number of components. The hierarchical structure is like that of IP addresses, but CCNx names and name components have arbitrary lengths and the component divisions are explicitly identified unlike in the classless model of IP addresses. The protocol does not depend on anything but the hierarchical structure of the names so they may contain arbitrary binary data such as encrypted data.

CCNx content names are not interpreted in the operation of the CCNx protocol itself, just matched. All assignment of meaning to names or their component parts comes from application, institution, and/or global conventions reflected in prefix forwarding rules.

A CCNx name sometimes identifies a specific chunk of data, but in other cases identifies a collection of data by naming a point in the name tree under which there may be multiple pieces of data. A name identifying a collection of data is analogous to the address of a network in the host addressing scheme of IP, where the network address can be seen to identify the collection of hosts attached to that network. In the case of naming a collection, the CCNx name is a prefix of the name of every piece of content in the collection, just as an IPv4 network address gives a prefix of the IP addresses of member hosts. For this reason, a CCNx name may be referred to as a name prefix or simply prefix.

See CCNx Name for details of the structure and representation of names.

The CCNx name of a chunk of data always contains as its final, most specific component a value that is derived from the data, called the digest component. Since it is derived from the data itself, the digest component is redundant and so is not transmitted. For more details see Implicit Digest Component.

CCNx Message Types

There are two message types in the CCNx protocol: Interest and Data (also called Content or Content Object).

The Interest message is used to request data by name. An Interest message can identify a chunk of content to retrieve very specifically. Alternatively, an Interest message can provide a name prefix and other qualifications to restrict what data is acceptable from the collection named by the prefix.

The Content Object is used to supply data. A Content Object message not only contains a data payload but the identifying name (without the implicit digest component), a cryptographic signature, and identification of the signer (called the publisher) along with other information about the signing. Formally, a Content Object message is an immutable binding of a name, a publisher, and a chunk of data. Every Content Object message is REQUIRED to contain a valid signature. In this way, all data communicated with the CCNx protocol is attested.

Any CCNx party MAY choose to verify the signature on any Content Object message that it receives. Applications that are going to make use of the data MUST verify first. Verification MAY require public keys which are not themselves included in the Content Object message to be verified. A CCNx party MUST discard a Content Object message that fails to verify.

The CCNx protocol does not provide a separate mechanism for key distribution, since keys can be distributed just like any other data using the general features of the protocol. Profiles for key distribution are separate specifications, not part of the CCNx protocol itself. Signature verification can confirm that a Content Object message has not been corrupted in transit (intentionally or otherwise) since it was originally signed, and that it was signed by the identified publisher, but these verifications are not the same as determining that the publisher in question is a trustworthy source of data for a particular application purpose. CCNx parties SHOULD use trust management practices to decide whether to trust data from particular publishers for particular purposes, but such practices are not specified as part of the CCNx protocol itself. The CCNx protocol is designed to ensure attestation of data, without constraining decisions about key distribution and trust management.

Basic Exchange

Communication using the CCNx protocol is receiver-controlled. A consumer of data transmits an Interest message over available connectivity and any party receiving the message and having data that matches, or satisfies, the request (according to the specifications in the Interest Message) may transmit a matching Content Object message. Data MUST only be transmitted in response to an Interest that matches the Data. An Interest message MAY be transmitted using broadcast or multicast facilities of the underlying transport in order to reach many potential sources of data with minimal bandwidth cost.

A party MUST transmit at most one Content Object message in response to a single received Interest message, even if the party has many Content Objects that match. This one-for-one mapping between Interest and Data messages maintains a flow balance that allows the receiver to control the rate at which data is transmitted from a sender, and avoids consuming bandwidth to send data anywhere it is not wanted.

Nodes SHOULD implement suppression mechanisms to minimize the potential for two different nodes to transmit two Content Object messages in response to a single Interest received by both nodes (for example via broadcast or multicast). The suppression mechanisms SHOULD include randomizing response times and detecting the fact that another node has broadcast or multicast a response so the Interest may be discarded. This version of the CCNx protocol does not specify suppression rules.

A receiver that desires to retrieve a large collection of data requiring multiple Content Object messages must transmit a sequence of Interest messages. For pipelining, a receiver MAY transmit multiple Interest messages without waiting for data in response to each one before sending the next. This is only possible if the receiver knows enough about the names of the Content Objects to construct different Interests in advance. Sending multiple Interest messages with identical requests will usually cause the same Content Object to be transmitted repeatedly, since senders MUST respond to Interests based only on what content they have available at the time, and not on any memory of what Content Objects they have previously transmitted.

The CCNx protocol does not assume that underlying transport of messages is reliable. To provide reliable delivery, Interest messages that are not satisfied in some reasonable period of time must be retransmitted. A receiver MUST maintain a timer on unsatisfied Interests for which it still wants the data, and retransmit them when the timer expires. This version of the CCNx protocol does not specify timer values.

These rules govern the basic Interest/Data exchange for local communication, that is communication between peers which directly receive message transmissions from each other. Since the CCNx protocol may be used on top of any packet transport, this definition of local is quite broad, including pairs of nodes physically far apart but using a long-distance connection such as a TCP connection as a tunnel to transmit CCNx messages directly. Multihop CCNx protocol communication requires forwarding of CCNx message which is specified in terms of the node model described in the next section.

CCNx Node Model

A full CCNx node (as opposed to a limited CCNx party such as a single application) contains the following data structures to provide buffering/caching and loop-free forwarding.

Content Store (CS)

A buffer memory organized for retrieval by prefix match lookup on names. Since CCNx Content Object messages are self-identifying and self-authenticating, each one is potentially useful to many consumers. The CS SHOULD implement a replacement policy that maximizes the possibility of reuse such as Least Recently Used (LRU) or Least Frequently Used (LFU). The CS MUST also implement the Staleness Bit. The CS MAY retain Content Object messages indefinitely but is not required to take any special measures to preserve them; the CS is a cache, not a persistent store.

Face

A face is a generalization of the concept of interface: a face may be a connection to a network or directly to an application party. A face may be configured to send and receive broadcast or multicast packets on a particular network interface, or to send and receive packets using point-to-point addressing in the underlying transport, or using a tunnel (for example a TCP tunnel). A face may also be the connection to a single application process running on the same machine, via an encapsulation like UDP or an OS-specific interprocess communication path. All messages arrive through a face and are sent out through a face.

Forwarding Information Base (FIB)

A table of outbound faces for Interests, organized for retrieval by longest prefix match lookup on names. Each prefix entry in the FIB may point to a list of faces rather than only one.

Pending Interest Table (PIT)

A table of sources for unsatisfied Interests, organized for retrieval by longest prefix match lookup on names. Each entry in the PIT may point to a list of sources. Entries in the PIT MUST timeout rather than being held indefinitely.

Note that the tables listed above may be interconnected through a single index to minimize the cost of lookup operations at the core of CCNx message processing and forwarding. Such an index must be ordered or prioritized to achieve the effect of the lookup order specified below.

Processing Interest Messages

An Interest Message is processed according to the following sequence:

  1. Lookup is performed on CS. If a matching ContentObject is found, it will be transmitted out the arrival face as a response to the Interest Message. Note that to match a ContentObject must satisfy all of the specifications given in the Interest Message. Note also that it may be that multiple ContentObjects simultaneously match in which case the specification in the Interest Message will be used to determine which object to return. When a match is found in the CS, processing stops and the Interest Message is discarded since it has been satisfied.

  2. Lookup is performed on PIT. If a matching Interest Message is found in the PIT it means that an equivalent Interest Message has already been forwarded and is pending. The arrival face of the new Interest Message is added to the list of sources of unsatisfied Interests in the PIT entry and the the Interest Message is discarded.

  3. Lookup is performed on FIB. If a matching prefix is found in the FIB, an entry is created in the PIT identifying the arrival face of the Interest Message and the message is transmitted according to the strategy rules to one or more of the outbound faces registered for the prefix in the FIB. Note that one of the outbound faces may actually be connected to a local agent that uses the semantics of the name to dynamically configure new faces.

  4. If no match is found in the previous steps then the node has no way to satisfy the Interest Message at present. It may be held for a short time before being discarded, since the creation of a new FIB entry may provide way to satisfy it.

Processing Content Messages

A ContentObject is processed according to the following sequence:

  1. Lookup is performed on CS. If a matching ContentObject is found it means that the newly arrived ContentObject is a duplicate which can safely be discarded, because any Interest Messages have already been satisfied and new ones will be satisfied out of the CS.

  2. Lookup is performed on PIT. If there is a match in the PIT, the ContentObject is transmitted on all of the source faces for the Interests represented in the PIT. A node MAY perform verification of the ContentObject before forwarding it, and MAY apply various policy restrictions.

  3. If no match is found in the previous steps, then the content is unsolicited. A node MUST NOT forward unsolicited data and SHOULD discard unsolicited data but MAY store it in the CS in case it is subsequently requested.

Strategy Rules

Unlike in IP, CCNx FIB entries may point to multiple next-hop destinations simultaneously. The self-identifying nature of Interest Messages and ContentObjects means that loops can be avoided without establishing a spanning tree allowing only one destination for any particular prefix at any particular node. The possibility of multiple outbound faces means that different strategies may be employed to select among them when forwarding messages. A node MUST implement some strategy rule, even if it is only to transmit an Interest Message on all listed outbound faces in sequence. A node MAY implement many different strategies. Strategy rules SHOULD be specified by FIB entries: in this case the FIB entry MAY contain effectively a constrained program for handling the forwarding of Interest Messages addressing the particular prefix.

A node MUST discard PIT entries that have not been refreshed by the arrival of new Interest Messages within some timeout. The new Interests confirm that the potential recipients are still interested in receiving content. A node MUST retransmit Interest Messages periodically for pending PIT entries. A node MAY use different timeouts for refreshing the liveness of received Interest Messages (downstream interests) from those used for retransmitting pending Interest Messages (upstream interests), and MAY use different timeouts associated for different faces. In this way, each node SHOULD match to both to downstream and upstream parties.