| Network Working Group | S. Salsano |
| Internet-Draft | A. Detti |
| Intended status: Informational | N. Blefari-Melazzi |
| Expires: December 22, 2013 | M. Cancellieri |
| Univ. of Rome "Tor Vergata" | |
| June 20, 2013 |
ICTP - Information Centric Transport Protocol for CONET ICN
draft-salsano-ictp-02
Let us consider an Information Centric Networking (ICN) solution, in which an End Node requests for a content sending "content requests" (or "interest packets"). The content is provided back to the requestor by the "origin" node or by an intermediate node that had cached the content. The content is usually divided into "chunks" that can be individually requested, sent back to the requester, cached into intermediate nodes. The sending rate of content requests can be adjusted in order to perform congestion control, implementing a receiver driven transport protocol. As it can be useful to have large chunks (significantly larger than the Maximum Tranfer Unit across the network), the transport protocol should also be used to further segment the chunks rather than relying to IP fragmentation. In this memo we define ICTP (Information Centric Transport Protocol), a receiver driven transport protocol for ICN, which relies on the CONET ICN solution described in a companion draft.
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 http://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 December 22, 2013.
Copyright (c) 2013 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 (http://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.
[I-D.CONET] proposes an approach to Information Centric Networking [Koponen07][Jacobson09]. It proposed to extends the IP protocol suite by defining a new IP Protocol type (CONET). Then it proposes two ways of carring ICN related information in IP packets, one uses a new IP Option (both for IPv4 [RFC0791] and IPv6 [RFC2460]), the other one only relies on the IP payload. The ICN related information is used by network nodes and end nodes to support networking based on content rather than (or better in addition to) end-point addresses. Further information on the proposed solution can also be found in [CONET11].
In this memo we define a receiver driven transport protocol for CONET (COntent NETwork) ICN, called ICTP - Information Centric Transport Protocol. The transport protocol is able to provide a reliable transfer of the content and to perform congestion control in TCP-friendly way. A discussion about the definition of a transport protocol for ICN can be found in [ICTP12].
As shown in Figure 1, the CONET architecture proposed in [I-D.CONET] foresees End-Nodes, Serving Nodes and CONET nodes. End-Nodes request for content. Serving Nodes provide content. CONET nodes: i) forward content requests from End-Nodes to Serving Nodes; ii) deliver content from Serving Nodes to End-Nodes; iii) may cache content and therefore provide it to End-Nodes without contacting the Serving Node.
requests for content
------------------->
content is provided
<-------------------
+----+ +----+ +----+
| | --| |------| |
+----+\ / +----+ +----+
\ +----+ +----+ /
----| |------| |/
+----+ +----+
End-Node legacy Intermediate Border Serving
IP router Node Node Node
| |
+---------CONET next hop----------->+
Figure 1: CONET architecture
In this section we recall the basic aspects of the CONET ICN solution, as needed to introduce the proposed receiver driven Information Centric Transport Protocol (ICTP).
The figure below shows the CONET protocol stack. CONET protocol is divided in two sub-layers, whose data unit are respectively denoted as "Carrier Packets" and "CONET Information Units". Two types of CONET Information Units are currently defined ("Interests CIU" and "Named Data CIU"). The CONET Information Unit Type field in the ICN information header differentiates among the two types. A CONET Information Unit (CIU) can be split into different Carrier Packets. Each Carrier Packet is transported by an IP packet.
+--------+--------+--------+ \
| CONET Information Units | |
+--------+--------+--------+ |
|
+--------+--------+--------+ |- CONET protocol
| Carrier Packets | |
+--------+--------+--------+ |
|
+--------+--------+--------+ /
| IP (opt. CONET IP option)|
+--------+--------+--------+
Figure 2: CONET protocol layers
The generic structure of a Carrier Packet (CP) is reported hereafter:
+-------------------------+
| CP Payload header |
+-------------------------+
| CP Payload |
+-------------------------+
| CP Path state |
+-------------------------+
"Interest CIU" are used by End-Nodes to request for content. The Interest CIUs contain the identifier of the content called ICN-ID and transported within the ICN information header. Optionally, the ICN information header explicitly carry a "Chunk Sequence Number" to identify one of the chunk in which a content has been split. Another possibility is that the chunk number is carried within the ICN-ID itself. A Serving Node or a CONET node that had previously cached the information can reply to the content request by sending a "Named-data CIU". These Named-data CIU will also contain the ICN-ID in the ICN information header.
The CP payload header contains the length of the CP Payload and allows to identify the start of the CP Path state field. The use of CP Path state field was explained in [I-D.CONET] and is out of scope here.
The information contained in the CP Payload header is specific for each CIU type. The information transported in the CP Payload header can be used to implement the functionalities of a transport protocol, providing a reliable transmission of content and performing congestion control. In this document we define the CP Payload header for Interest and Data CIUc using the ICTP protocol.
An end-node that wants to retrieve a content (or better a Chunk of a content) issues an Interest CIU, the ICN-ID and (optionally) the Chunk Sequence Number of the required Content are respectively transported in the ICN Identifier (ICN-ID) field and in the CSN field of the ICN information header. Assuming for simplicity that the Interest CIU will fit into a single Carrier Packet, the Interest CIU will be included in the Carrier Packet that in turn is inserted into an IP packet. The ICTP comes into play because a Chunk of content may need to be fragmented in more than one Carrier Packet, as the chunk size can be much larger than the layer 2 MTU. For example with a chunk size of 64 KB or 256 KB and an MTU of 1500 bytes, the chunks need to be split in tens or hundreds of packets. In this case the RDTP allows to specify in the request the specific segment of the chunk that is required.
The structure of the interest CP payload header is reported hereafter:
+-------------------------+
|TTrrrrrPI| ..Left Edge...|
+-------------------------+
| ...Right Edge... |
+-------------------------+
Flags:
TT : Transport protocol type. It allows to define different transport protocols. 0 indicates ICTP which is defined in this draft. 1-3 are reserved and can be used to indicate different transport protocols. The rest of the bits in this field and in the following bytes may have a different semantic depending on the transport protocol, we are providing here only the defintion for TT=0 i.e. the ICTP protocol. Therefore the number of transport protocols is NOT limited to 4.
P : Prefetch flag - This flag indicates that this packet comes from a receiver that asks to perform prefetch on the content chunks.
I : Ask Chunk Info flag - If this flag is set, the serving node is requested to add chunk-related information to the data CIU payload, particularly the chunk size.
Left edge/Right edge : These fields contain respectively the value of first and the last byte of the requested chunk segment. The fields are encoded with the EVLE (Efficient Variable Lenght Encoding) mechanim described below
The structure of the data CP payload header is reported hereafter:
+-------------------------+
|TTrFSSSS| ...Left Edge...|
+-------------------------+
| ...Right Edge... |
+-------------------------+
| (Optional) Chunk size |
+-------------------------+
Flags:
TT : Transport protocol type. It allows to define different transport protocols. 0 indicates ICTP which is defined in this draft. 1-3 are reserved and can be used to indicate different transport protocols. The same considerations apply that have been reported above for the TT subfield in the interest CIU payload header.
F : Final segment flag - If this bit is set to 1 this carrier packet carries the last segment of a chunk.
SSSS : Chunk size flag - This flag describes the size of a chunk as follows:
0 : unspecified (it may have been already indicated
in a previous data)
1 : the chunk size is trasnported in the optional
Chunk size field, encoded with the EVLE variable length
encoding described below
2-16 : let n be the value from 2 to 16, it can represent
14 different chunk sizes from 2KBytes to 8Mbyte
with the following relation:
chunk size = 2 ^ (9+n)
Left edge/Right edge : These fields contains respectively the value of first and the last byte of the transported chunk segment. The fields are encoded with EVLE variable length encoding described below. These fields carry the actual value of the segment transported that may differ from the request.
Some of the fields described above are encoded using a variable lenght encoding that we denote as "Efficient Variable Lenght Encoding". The same encoding is used for the Chunk Sequence Number (CSN) field in the ICN information header, as described in [I-D.CONET]. To help the reader, we report the definition of the encoding hereafter. An EVLE field is represented with a variable number of bytes. An initial bit pattern determines the length of the EVLE field.
1 byte EVLE (7 bits range)
+--------+
|0 |
+--------+
2 bytes EVLE (15 bit range)
+--------+--------+
|10 |
+--------+--------+
3 bytes EVLE (21 bit range)
+--------+--------+--------+
|110 | | |
+--------+--------+--------+
4 bytes EVLE (28 bit range)
+--------+--------+--------+--------+
|1110 | | | |
+--------+--------+--------+--------+
5 bytes EVLE (32 bit range)
+--------+--------+--------+--------+
|11110000| | | |
+--------+--------+--------+--------+
| |
+--------+
6 bytes EVLE (40 bit range)
+--------+--------+--------+--------+
|11110001| | | |
+--------+--------+--------+--------+
| | |
+--------+--------+
As explained in in [I-D.CONET], binary patterns from 11110010 to 11111111 are reserved and could be used to extend the EVLE range if needed. With the above definion the maximum value is 2^40, roughly 1 Tera.
The transport protocol described in this draft mimics the TPC mechanisms described in [RFC2581], with the required adaptations for a receiver driven approach. In fact, while in TCP the sender sends data segment and implements retransmissions and congestion control based on the reception of ACKs, in ICN based content download the receiver asks for content sending the Interests and implements re-sending of Interests and congestion control based on the reception of Data.
The receiver bases its flow control on the received data CP. The out of sequence recognition is based on the expected chunk number and segment bytes. When three out-of-sequence data CP are received the slow start threshold (ssthresh, see [RFC2581]) is set to half the window. Then the window is set to ssthresh plus the 3 data carrier packet already received and the interest for the missing segment is retransmitted.
To manage the increase of congestion window slow start and congestion avoidance mechanism are performed. During slow start, (e.g. the beginning of the connection), the congestion window increases of an amount equal to the received data CP. This corresponds to an "exponential" growth of the window. During congestion avoidance, the growth of the window is aproximately "linear" with time. Let us define a "segment" as the portion of content transported in a data CP. We define the congestion window cwnd in segments. During slow start cwnd = cwnd + 1 for each received data CP. During congestion avoidance cwnd = cwnd + 1/cwnd for each received data CP.
In a ICN based network, we can have applications separately requesting the download of each chunk of content. In our ICTP approach this means that we can only request the segments of a given chunk after that we have received the request for the chunk coming from the application. The application may implement retransmission and congestion control, therefore the ICTP could receive requests for more than a chunk of the same content in parallel. Anyway the interaction between the congestion control performed at application level and the congestion control performed at ICTP level could prove inefficient. Therefore we believe that the API offered to the application should allow the application to request for a whole content (or for the content starting from a given chunk number). Then the ICTP protocol can handle the retrieval of all the (rest of) the content. Therefore the ICTP protocol offers also a "prefetch option". Whit the prefetch option active, if the congestion window has space left, the mechanism allow to issue requests for segments of next expected chunks, without waiting for an explicit request from application level.
When the transmission start, the receiver must issue a request for at least the chunk size. Receiver should set the flags in carrier packet accordingly to Section 3.1. The data ciu requested will carry at least the chunk size, receiver should use this information to create data structure best suited for the chunk size.
A goal of the ICTP transport protocol is to compete fairly with TCP. This allows a fair sharing of reources when TCP flows and ICN flows compete over network links.
We acknowledge the financial support by the EU in the context of the CONVERGENCE research project.
This document requires the allocation of one IP protocol number by the IANA.
This document requires the allocation of one IP option by the IANA if the solution of using IP options is adopted
This document requires that IANA will maintain the registry of CONET namespaces.
Security considerations to be provided
| [RFC2629] | Rose, M.T., "Writing I-Ds and RFCs using XML", RFC 2629, June 1999. |
| [RFC0791] | Postel, J., "Internet Protocol", STD 5, RFC 791, September 1981. |
| [RFC2460] | Deering, S.E. and R.M. Hinden, "Internet Protocol, Version 6 (IPv6) Specification", RFC 2460, December 1998. |
| [Koponen07] | T. Koponen et al., , "A data-oriented (and beyond) network architecture", Proc. of ACM SIGCOMM 2007 , 2007. |
| [Jacobson09] | V. Jacobson, et al., , "Networking named content", Proc. of ACM CoNEXT 2009 , 2009. |
| [CONET11] | A. Detti, et al., , "CONET: A Content Centric Inter-Networking Architecture", ACM SIGCOMM Workshop on Information-Centric Networking (ICN-2011), Toronto, Canada , August 2011. |
| [ICTP12] | S. Salsano, et al., , "Transport-layer issues in Information Centric Networks", ACM SIGCOMM Workshop on Information-Centric Networking (ICN-2012), Helsinki, Finland , August 2012. |
| [I-D.CONET] | Detti, A., Salsano, S. and N. Blefari-Melazzi, "IP protocol suite extensions to support CONET Information Centric Networking", Internet-Draft draft-detti-conet-ip-option-05, June 2013. |
| [RFC2581] | Allman, M., Paxson, V. and W. Stevens, "TCP Congestion Control", RFC 2581, April 1999. |