Chopan - Compressed HTTP Over PANs
draft-frank-6lowapp-chopan-00.txt

Status of this Memo

This Internet-Draft is submitted to IETF in full conformance with the provisions of BCP 78 and BCP 79.

Internet-Drafts are working documents of the Internet Engineering Task Force (IETF), its areas, and its working groups. Note that other groups may also distribute working documents as Internet-Drafts.

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.”

The list of current Internet-Drafts can be accessed at http://www.ietf.org/ietf/1id-abstracts.txt.

The list of Internet-Draft Shadow Directories can be accessed at http://www.ietf.org/shadow.html.

This Internet-Draft will expire on March 15, 2010.

Copyright Notice

Copyright (c) 2009 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 in effect on the date of publication of this document (http://trustee.ietf.org/license-info). Please review these documents carefully, as they describe your rights and restrictions with respect to this document.

Abstract

This document describes a method for compressing HTTP messages into a binary format to be transmitted using UDP over 6LoWPAN wireless networks.

Table of Contents

1.  Introduction
    1.1.  Requirements notation
    1.2.  Security Considerations
    1.3.  Terminology
2.  Datagram Format
    2.1.  Format Notation
    2.2.  Request Format
    2.3.  Response Format
    2.4.  Compressed Headers
    2.5.  Mime Type Codes
    2.6.  Example
3.  UDP Transmission
4.  Transaction-Id
5.  Caching
    5.1.  Cache Control
    5.2.  ETag Validation
    5.3.  Interception Proxy Caching
    5.4.  Sleeping Nodes
    5.5.  Cache Refresh
    5.6.  Caching non-GET Methods
6.  HTTP to Chopan Gateways
7.  Security
8.  Normative References
§  Author's Address

1.  Introduction

The Pervasive Internet is a vision that everyday devices with microprocessors are woven into the fabric of the Internet. One of the critical emerging technologies in this domain is 6LoWPAN which enables low cost, low power devices to communicate using the Internet Protocol. 6LoWPAN is the first step towards building the Pervasive Internet. Chopan defines the next step: integrating 6LoWPAN devices with the World Wide Web to leverage the massive investment in existing URI and HTTP infrastructure.

Chopan is derived from HTTP with these changes:

• UDP: utilizes UDP packets instead of TCP as the underlying transport protocol
• Binary compression: HTTP headers are compressed into a binary format to save bandwidth and buffer space
• Interception Caches: transparent caching is used to minimize PAN traffic and manage sleeping nodes
• Gateways: may be used to translate between full HTTP and Chopan to interoperate with the existing Web infrastructure

1.1.  Requirements notation

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in [RFC2119] (Bradner, S., “Key words for use in RFCs to Indicate Requirement Levels,” March 1997.).

1.2.  Security Considerations

Discussed in Section 7 (Security).

1.3.  Terminology

6LoWPAN: IPv6 for Low power Personal Area Networks described in [RFC4944] (Montenegro, G., “Transmission of IPv6 Packets over IEEE 802.15.4 Networks,” September 2007.).

Compression: translation from of a TCP/HTTP text based message into a compressed binary UDP/Chopan message (gateway functionality).

Decompression: translation from of a binary UDP/Chopan message into a TCP/HTTP text based message (gateway functionality).

Gateway: a node which transparently translates between HTTP and Chopan messages.

HTTP: Hyper Text Transfer Protocol described in [RFC2616] (Fielding, R., “Hypertext Transfer Protocol -- HTTP/1.1,” June 1999.).

PAN: Personal Area Network - an IP sub-network with constrained bandwidth and/or constrained computing devices. This specification is designed for low power PANs running 6LoWPAN, but Chopan is an ideal solution for any network with bandwidth or computing restraints.

Interception Proxy Cache: a node which transparently intercepts HTTP requests to an origin server and returns cached responses on its behalf.

Origin Server: the server on which the master version of resource resides.

Resource: an abstract unit of information identified with a URI and transported over a network using a MIME typed representation.

Sleeping Nodes: battery powered network nodes which spend most of their time in a hibernation state to converse power.

TCP: Transmission Control Protocol described in [RFC0793] (Postel, J., “Transmission Control Protocol,” September 1981.).

UDP: User Datagram Protocol described in [RFC0768] (Postel, J., “User Datagram Protocol,” August 1980.).

UTF-8: Encoding of Unicode characters compatible with ASCII described in [RFC2279] (Yergeau, F., “UTF-8, a transformation format of ISO 10646,” January 1998.)

2.  Datagram Format

Chopan uses a customized binary encoding for HTTP requests and responses to achieve message compression into a UDP packet. A two byte magic number is used to identify the packet as a Chopan message – “h6” for requests and “H6” for responses. Both requests and responses allow for zero or more compressed headers.

Any bytes after the headers in the packet are considered the message-body. The length of the message-body is implied by the packet length (the Content-Length header MAY be omitted). The entire message MUST fit with in a single UDP packet. When running over 6LoWPAN, messages SHOULD fit into a single 802.15.4 frame to avoid fragmentation.

2.1.  Format Notation

Message formats are described as a data structure using the following primitive types:

• u1: an unsigned 8-bit byte
• u2: an unsigned 16-bit integer in network byte order
• str: UTF-8 (Yergeau, F., “UTF-8, a transformation format of ISO 10646,” January 1998.) [RFC2279] encoded text, followed by a null terminator (0x00) byte
• x[]: an sequence of type x which contains zero or more occurrences
• x|y: either x OR y

2.2.  Request Format

A normal HTTP request is composed of a request-line, a set of request-headers, and the message-body. This information is compressed in the following binary format:

request
{
  u2       magic 0x6836 – ASCII "h6"
  method   method
  str      uri
  header[] headers
  u1       zero byte end of headers
  u1[]     message-body
}

method
{
  u1   method-code
  utf  optional string value only if method-code is 0x80
}

The HTTP request-line contains three pieces of information: the method, URI, and version. The URI is encoded as a null-terminated UTF-8 string. Standard request methods are encoded into a byte as follows:

Method         Code   ASCII Char
------------   ----   ----------
DELETE         0x44   D
GET            0x47   G
HEAD           0x48   H
OPTIONS        0x4F   O
POST           0x50   P
PUT (Update)   0x55   U
TRACE          0x54   T
str value      0x80   -

Most standard methods are encoded into a single byte, for example "GET" is encoded into the ASCII byte 'G'. If the method code is 0x80, then it is followed by a null terminated UTF-8 string.

2.3.  Response Format

A normal HTTP response is composed of a status-line, response-headers, and the message-body. This information is compressed in the following binary format:

response
{
  u2       magic  – ASCII "H6"
  u1       status-code
  header[] headers
  u1       zero byte end of headers
  u1[]     message-body
}

The HTTP status code is compressed into a single byte where the top 3-bits represent the 100s decimal digit, and the bottom 5-bits represent the last two decimal digits. Example of binary mappings:

1xx  ->  0x2X, b001x_xxxx
2xx  ->  0x4X, b010x_xxxx
3xx  ->  0x6X, b011x_xxxx
4xx  ->  0x8X, b100x_xxxx
5xx  ->  0xAX, b101x_xxxx
200  ->  0x40  // OK
404  ->  0x84  // Not Found
415  ->  0x4F  // Unsupported Media Type
416  ->  0x50  // Requested range not satisfiable
417  ->  0x51  // Expectation Failed

2.4.  Compressed Headers

Standardized HTTP request and response headers are compressed using predefined binary codes. Compressed headers are encoded as follows:

header
{
  u1     header-code (high bit determines encoding of value)
  u2|str value (u2 or str based on header-code high bit)
}

Headers are encoded using a 8-bit header-code which represents the header name. If the high bit (0x80) is clear in the header-code, then the value is encoded as an unsigned 16-bit integer. If the high bit is set, then the value is encoded as a null terminated UTF-8 string. The u2 value encoding allows compression on a header-by-header basis. Refer to the table below how each header utilizes a u2 value.

If an HTTP header name does not have a standard binary encoding, then it MAY be stripped at the proxy gateway, otherwise it can be passed using its string name. Uncompressed header names are encoded as follows:

uncompressed-header
{
  u1      header-code is 0x7f (u2 val) or 0xff (str val)
  str     name encoded as null terminated string
  u2|str  value
}

The following table defines the header codes for standard HTTP headers. Each code has the high bit clear indicating a u2 value. Mask the code with 0x80 to obtain the str value code:

Header              Code  Notes
------------------  ----  -------------------------------------------
End-Of-Headers      0x00  zero indicates no more headers
Uncompressed        0x7F  name string, u2/string value
Accept              0x01  u2 val: mime type code
Accept-Charset      0x02
Accept-Encoding     0x03
Accept-Language     0x04
Accept-Ranges       0x05
Age                 0x06  u2 val: delta age in seconds
Allow               0x07
Authorization       0x08
Awake-Time          0x09  u2 val: seconds, used with check-in request
Cache-Control       0x0A  u2 val: max-age in seconds
Connection          0x0B  unsupported
Content-Encoding    0x0C
Content-Language    0x0D
Content-Length      0x0E  u2 val: bytes; omit to imply by packet size
Content-Location    0x0F
Content-MD5         0x10
Content-Type        0x11  u2 val: mime type code
Cookie              0x12
Date                0x13
ETag                0x14  u2 val: etag is 4 digit upper case hex str
Expect              0x15  u2 val: uncompressed code 100 is 0x64
Expires             0x16  should be avoided (use max-age)
From                0x17
Host                0x18
If-Match            0x19  u2 val: etag is 4 digit upper case hex str
If-Modified-Since   0x1A  should be avoided (use max-age)
If-None-Match       0x1B  u2 val: etag is 4 digit upper case hex str
If-Range            0x1C
If-Unmodified-Since 0x1D  should be avoided (use max-age)
Last-Modified       0x1E  should be avoided (use age, max-age)
Location            0x1F
Max-Forwards        0x20  u2 val: number of hops
Pragma              0x21  obsolete
Proxy-Authenticate  0x22
Proxy-Authorization 0x23
Range               0x24
Referer             0x25
Retry-After         0x26  u2 val: seconds, used with 202 responses
Server              0x27
Set-Cookie          0x28
Sleep-Time          0X29  u2 val: seconds, used for check-in requests
TE                  0x2A
Transaction-Id      0x2B  u2 val: same as 4 digit upper case hex str
Trailer             0x2C  unsupported
Transfer-Encoding   0x2D
Upgrade             0x2E
User-Agent          0x2F
Vary                0x30
Via                 0x31
Warning             0x32  u2 val: uncompressed code 111 is 0x6F
WWW-Authenticate    0x33

2.5.  Mime Type Codes

The Accept and Content-Type headers may be compressed into an unsigned 16-bit type code using the following table:

Mime Type                 Code    Notes
------------------------  ------  -------------------------------
application/octet-stream  0xA001  used for arbitrary binary files
text/plain                0xB001  charset implied to be UTF-8
text/html                 0xB002  charset implied to be UTF-8
text/xml                  0xB003  charset implied to be UTF-8
text/csv                  0xB004  charset implied to be UTF-8

NOTE: we also need to give thought to what kind of information models we use and how they are represented with existing or new MIME types. For example we might want to use ASN.1 MIBs, binary oBIX, etc...

2.6.  Example

Assume the following HTTP request:

GET /pt07 HTTP/1.1
Host: sensor2086.acme.com
Accept: text/plain
If-None-Match: "3A7F"
Cache-Control: max-age=900

The HTTP request above would be compressed into the following sequence of hexadecimal bytes:

68 36 47 2F 70 74 30 37 00 01 B0 01 1B 3A 7F 0A 03 84 00
^     ^  ^                 ^        ^        ^        ^
|     |  |                 |        |        |        +- End
|     |  +- URI            |        |        +- Cache-Control
|     +- GET               |        +- If-None-Match
+- magic                   +- Accept

Note that we stripped the Host header and compressed Accept, If-None-Match, and Cache-Control into two byte header values.

3.  UDP Transmission

One of the primary characteristics of Chopan is the ability to transmit HTTP requests and responses over UDP. Since HTTP was originally designed to be run over TCP, we must make some design trade-offs to layer the protocol over an unreliable packet based transport.

Chopin follows the standard HTTP request/response model. A client makes a Chopan request to a server with a request message. When the server receives the request, it sends the client back a response message.

Both the request and response messages MUST fit within one UDP packet, as such large message bodies are not supported. However, the Range header may be used to chunk the transfer of resources which do not fit a single UDP packet. When running over 6LoWPAN, messages SHOULD fit into a single 802.15.4 frame to avoid fragmentation.

Because UDP is unreliable, there is no guarantee that a server receives a request, nor that a client receives the response. If a client does not receive a response to its request after a reasonable amount of time, then it SHOULD retry the request up to three times before timing out. It is therefore possible that the server might receive the same request multiple times. A request is "retry safe" if it can be retried multiple times by the client without compromising server state. Idempotent methods like GET and HEAD MUST be retry safe. Methods such as PUT and DELETE should also be retry safe since they atomically modify or delete the resource. Methods like POST are typically not retry safe unless coupled with another mechanism. In the next section we examine an extension to HTTP for making requests retry safe with the Transaction-Id header.

UDP does not guarantee message order. Therefore, it is the client’s responsibility to impose message ordering if required. Message ordering can be maintained by waiting for a response, before sending the next request. When message ordering is not required, the client MAY have multiple simultaneous outstanding requests. This can increase throughput on networks with high latency. If performing concurrent requests, clients SHOULD use the Transaction-Id header to match responses to requests.

4.  Transaction-Id

Due to the unreliable nature of UDP, requests and responses do not have guaranteed delivery or ordering. This can particularly cause problems when a non-idempotent request is received successfully by the server, but the response packet is dropped. In this case the client’s expected behavior is to retry the request which might cause the server to receive the same request multiple times. For methods such as POST which are not implicitly retry-safe, we define a new header called Transaction-Id.

Transaction-Id is a unique identifier generated by the client. The tuple of the client’s IP address, port number, and Transaction-Id should be globally unique within the transaction’s temporal window. Any retries initiated by the client MUST include the same transaction id in the retry requests.

When a server receives a request with a Transaction-Id header, it MUST pass the identifier back to the client via the response’s Transaction-Id header. The server MAY also choose to utilize the Transaction-Id to implement "at-most-once" semantics. It is a server local matter to decide how to apply the transaction id for a given HTTP method and resource.

If a client attempts to request a method on the resource which requires a Transaction-Id header and fails to specify one, then the server SHOULD respond with 400 Bad Request.

5.  Caching

6LoWPAN networks are typified by a gateway device which acts as a router or bridge between the PAN and the external IP network. Often the external IP network is physically connected by high a bandwidth technology such as Ethernet or WiFi. The PAN itself typically has low bandwidth and is composed of resource constrained nodes. Often times the nodes in a PAN are battery powered, and spend most of their time sleeping.

Because of this physical architecture, it is desirable for the more capable nodes in the PAN to serve as caches for the more constrained devices. Effective use of caching enables Chopan to optimize both bandwidth on the PAN and power on constrained devices. In the case of a sleeping node, it allows proxies to immediately return cached representations of resources.

5.1.  Cache Control

HTTP (Fielding, R., “Hypertext Transfer Protocol -- HTTP/1.1,” June 1999.) [RFC2616] defines a sophisticated caching model in sections 13 and 14.9. This model has multiple caching features, often with overlapping functionality. Since Chopan is targeted for resource constrained devices, this specification recommends use of a subset of the HTTP caching model based on resource age and max-age.

It is expected that most resources accessed by Chopan are representations of sensor data. The nature of the sensor data determines its cache life. For example a temperature sensor in a room is likely to change very slowly, so it might have a cache life of fifteen minutes. But a temperature sensor in an oven might have a cache life of only ten seconds before it is considered stale data. Chopan uses existing HTTP caching features to give both the client and server a say in cache management.

When an origin server publishes a resource representation via a GET request, it SHOULD specify the Age header. For example if a resource represents a sensor, and that sensor was read 4 seconds ago, then the Age header should be set to 4 seconds. If the resource has an age less than 1 second, then set the Age header to 0. The Age header SHOULD be compressed into a two byte value if less than 18.2 hours.

In cases when the origin server has knowledge about the cache life of a given resource, it SHOULD set the Cache-Control header with a Max-Age directive. Note that the two byte value encoding of Cache-Control is implied to be Max-Age as a number of seconds. When the server specifies Max-Age, it is directing upstream proxies and clients how long to cache the resource. For example if a resource specifies an Age of 4 seconds, and a Max-Age of 30 seconds, then the resource should be cached for 26 seconds before it is considered stale.

Clients MAY also specify the Cache-Control header with a Max-Age directive on requests. In this case, the client is directing the maximum amount of staleness which may be tolerated. For example if a client requests a resource with a Max-Age of 10 seconds, and the resource has an age of 8 seconds, then the server may respond with the cached resource. If however the resource has an age older then 10 seconds, then the server should refresh its cache. In the case of a proxy cache, this means contacting the origin server. In the case of the origin server, it may require polling the sensor.

A resource is considered stale if its Age is greater than either Max-Age specified by the server or the Max-Age specified by the client. If a server node has a cached version of a resource which is stale, it SHOULD always attempt to refresh its cache. If the cache cannot be refreshed immediately because of normal operation (for example the origin server is a sleeping node), then the stale resource should be returned and the Warning header SHOULD be specified with the 110 status code (response is stale). If cache refresh fails abnormally (for example the origin server cannot be contacted), then the stale resource SHOULD be returned and the Warning header specified with the 111 status code (revalidation failed).

5.2.  ETag Validation

Key to any caching strategy is cache validation, the mechanism used by a client or proxy to refresh its cache with the origin server. Often even though a cached resource has expired, the original resource hasn’t been modified. But in order to avoid re-transmitting the entire resource the client and server must define a mechanism to validate the cached copy. In HTTP this validation may be negotiated using either timestamps or entity tags. Chopen discourages the use of timestamps because often nodes do not support time clocks. Instead entity tags SHOULD be used for cache validation.

An entity tag is an opaque hash of a given resource’s version. It is defined by the origin server using the ETag header. If possible, a two byte etag should be used to allow for optimal compression. If an etag was specified for a cached resource, then clients and proxies SHOULD specify the If-None-Match header on cache refresh. The server SHOULD respond with a 304 (Not Modified) response if the etag has been not modified.

5.3.  Interception Proxy Caching

In Chopan, caching is done transparently to the client via "interception caching". Interception caching is a commonly used technique used to insert HTTP caches between clients and origin servers, without requiring client configuration. Clients send packets to the origin server directly, but as these packets are routed into the PAN, one of the routing nodes processes the message directly on behalf of the origin server. This architecture requires that routing nodes in the PAN are actively examining the packets before they are routed to their destination address.

The downside to using interception caching, is that technically it breaks the encapsulation of the IP stack - routing nodes must become aware of an application level protocol. The upside to this design, is that client nodes do not have be explicitly configured to know about the proxies for every PAN. Since PANs have the potential to add billions of new nodes to the Internet, it seems reasonable to trade-off the purity of IP routing within the PAN to maintain the simplicity of the Internet at large.

Interception caches SHOULD use a combination of the destination port and the packet’s magic two byte marker to sniff Chopan packets. By default we assume Chopan runs on UDP port 80, although proxies SHOULD make this configurable.

The lifecycle of an interception cache request:

1. The client sends a request to the origin server
2. The interception proxy traps the request
3. If the request can be immediately fulfilled by a cached representation of that resource in the proxy, then the proxy responds directly to client on behalf of the origin server using the origin server’s IP address
4. If the proxy has no cached representation of the resource (or the cache has expired), then it makes its own request to the origin server for the resource to update its cache, then performs step 3 to return the cached resource to the client
5. Cache might also be actively refreshed periodically (see Cache Refresh)

This lifecycle assumes that the origin server is a powered device which is awake during normal operation. If the origin server is a battery powered device then the origin server is mostly likely sleeping. This use case is discussed further in Sleeping Nodes section.

Interception proxies SHOULD be transparent to the client. However, when a proxy communicates directly with the origin server it has a choice to forward the client’s original packet (with the client’s IP address), or to initiate a new request (with the proxy’s IP address). Proxy’s SHOULD initiate new requests using the proxy’s own IP address. This means that origin servers are effectively responding directly to the proxy with no knowledge of the original client request. The disadvantage of this model is that it breaks end-to-end communication principles of the Internet. However this model provides significant advantages:

• On 6LoWPAN it keeps IP addressing to intra-PAN nodes which results in better compression (since we don’t need to pass through the external IP address);
• It ensures that the response gets routed directly back to the proxy for caching;
• Gateways which are translating TCP/HTTP into UDP/Chopan do not have UDP packets from the client to begin with (rather they are translating from a TCP stream)
• Sleeping nodes which require active cache refresh must be polled directly by the proxy

5.4.  Sleeping Nodes

PANs commonly include battery powered nodes which spend most of their time sleeping to conserve power. These nodes periodically wake up to check sensors, perform computation, and catch up on network communications. Because of their nature, sleeping nodes do not make for reliable origin servers. Chopan handles this use case by fronting all sleeping nodes with interception caches. This allows all requests for resources on the sleeping nodes to be transparently brokered by proxies. Proxies then synchronize their caches with the sleeping nodes periodically during a "check-in" process.

The lifecycle for interception caching of sleeping nodes follows the standard interception model detailed above. However, when a request is made for a resource the proxy doesn’t have in its cache, the request cannot be immediately fulfilled. In this case the proxy SHOULD return a 202 Accepted response indicating that background processing is required before the request can be completed (waiting for the sleeping node to wake up). The Retry-After header SHOULD be set indicating the number of seconds before the request should be tried again. The retry time should be based on the time it will take the sleeping node to wake up, check-in, and give the proxy a chance to refresh its cache. The Retry-After header can be estimated from the Awake-Time and Sleep-Time headers (see below).

Sleeping nodes MUST be configured to check-in with their proxy or proxies when they wake up. This is done by sending a POST request to the "/ci" URI of each proxy. When a proxy node receives a check-in request, it SHOULD respond with 200 OK response. The sleeping node SHOULD use standard retry/timeout mechanism to ensure that the check-in is received by the proxy. After the sleeping node has checked-in, then the proxy SHOULD poll for all the resources in its cache which require refreshing. This will include all new pending resources which resulted in 202 responses. After the sleeping node has given the proxy a chance to refresh its cache, it can go back to sleep.

Sleeping nodes SHOULD specify the Awake-Time and Sleep-Time headers in their check-in request. The Awake-Time header specifies how long the node expects to stay awake to give the proxy a chance for cache refresh. The Sleep-Time indicates how long the node expects to sleep before the next check-in. A proxy should expect the next check-in after the sum of Awake-Time and Sleep-Time has elapsed - this period can then be used for estimating the proxy’s Retry-After header.

5.5.  Cache Refresh

Chopan proxies can take an active or a passive approach to cache refresh. In a passive model, stale resources are allowed to expire and are eventually flushed from the cache. New requests for the resources are forwarded to the origin server, and the response is used to refresh the cache. On the other hand, the proxy can actively poll origin servers to refresh cached resources independent of client requests.

For sleeping nodes, proxies MUST actively refresh their cache. This is required because there are only limited windows of opportunity while the node is awake for the proxy to refresh resources.

When the origin server is a powered node, either active or passive cache refresh may be used. Using active refresh to proactively keep caches refreshed can potentially decrease the latency of external requests.

Cached resources can be in one of the following states:

• Pending: these are resources on sleeping nodes which have resulted in a 202 response. Eventually we expect to poll the node on check-in and turn them into fresh resources or invalid resources.
• Fresh: these are resources with an Age less than both the client’s and server’s configured Max-Age.
• Stale: these are resources with an Age which exceeds either the client’s or server’s Max-Age. The proxy may continue to maintain stale resources in the cache for some period of time.
• Flushed: resources may be flushed from a cache at any time. Normally stale resources are flushed after a timeout period. However LRU caches may flush fresh resources if buffer space is exceeded.
• Invalid: some proxies may maintain a cached representation of a resource to indicate an error condition. This is helpful when a proxy receives a request for a sleeping node and returns 202, then after the check-in discovers the origin server returns 404 for the resource. In this case the proxy SHOULD temporarily cache an error return so that the client’s next poll will receive a 404 instead of another 202.

5.6.  Caching non-GET Methods

In most circumstances, clients make GET requests to retrieve representations of resources. In this case, proxies are caching the response which contains that resource representation. However clients may also perform POST, PUT, or DELETE requests. In the case where the origin server is a powered node, these requests SHOULD always be immediately forwarded to the origin server.

However in the case of sleeping nodes, the proxy MUST cache the request itself until the node wakes up and checks-in. Without this functionality it would be impossible to perform these HTTP methods on sleeping nodes. Non-GET methods to sleeping nodes MUST use a Transaction-Id to associate the request with a specified client IP address, port number, and transaction id.

Let’s consider a transaction for a resource POST on a sleeping node:

1. Client POSTs to origin server with a unique transaction id
2. Proxy transparently intercepts the request, caches it, and returns 202
3. Upon check-in the proxy forwards the request then caches the response
4. Client waits for Retry-After, then resubmits POST request using same transaction id
5. Proxy transparently intercepts the request and returns cached response with the transaction id

6.  HTTP to Chopan Gateways

Chopan leverages the HTTP standard in order to provide interoperability with the World Wide Web. Interoperability is achieved by using standard HTTP external to the PAN and using Chopan internal to the PAN. Nodes which perform HTTP-Chopan translation are called Chopan gateways:

• Requests into the PAN are translated from HTTP to Chopan.
• Requests from inside the PAN to the external network are translated from Chopan to HTTP.
• Requests from inside the PAN to other nodes inside the PAN are Chopan end-to-end

Diagram of gateway translations:

          <= External  |  PAN =>
Client -> [HTTP] -> Gateway -> [Chopan] -> Server
Server <- [HTTP] <- Gateway <- [Chopan] <- Client

Gateway translations SHOULD be performed transparently. Clients external the PAN assume they are communicating HTTP directly to the origin server. Gateways intercept these HTTP requests and translate them into Chopan requests. Likewise responses are translated from Chopan back to HTTP.

Because Chopan recommends that translation happens transparently, this means that the gateway must be sniffing incoming packets for TCP/HTTP requests. This design has all the same issues as detailed in Interception Proxy Caching. It is expected that in most implementations the gateway will also perform interception caching, although this specification does not require it.

HTTP to Chopan is referred to as compression, and Chopan to HTTP is referred to as decompression. During the compression process the text format of requests and responses is encoded into Chopan’s binary message format. Each HTTP header is examined and mapped into its binary encoding. Depending on the quality of the PAN link layer, the compression process may strip out HTTP headers, according to these priorities:

• Content type and cache-control headers SHOULD never be stripped
• Standard headers with u2 value encodings or short strings SHOULD be maintained
• Standard headers without u2 value encodings or with longer strings MAY be stripped
• Non-standard headers SHOULD be stripped (assuming typical PAN constraints)

The Chopan compression and stripping of headers is a gateway to origin server matter. This does not free the gateway from faithfully implementing the full HTTP specification and abiding by its conventions. In the cases where HTTP headers or functionality is reduced to meet Chopan constraints, the gateway should compensate so that the client’s perspective is communication with a fully compliant HTTP origin server.

7.  Security

Ideally Internet protocols implement an end-to-end security model between the two endpoint nodes. However it is difficult to implement end-to-end session based security with unreliable packet protocols and sleeping nodes. Rather Chopan, recommends that the security strategy is divided between internal and external PAN nodes.

Internally all PAN nodes should be fully trusted using link layer security such as the AES encryption specified by 802.15.4.

External to the PAN, the gateway should utilize full TCP/HTTP to enable the well known security mechanisms associated with those protocols. This includes TLS/HTTPS and the various HTTP authentication mechanisms.

NOTE: A lot more to think about here...

8. Normative References

Author's Address