PCP T. Reddy
Internet-Draft P. Patil
Intended status: Standards Track Cisco
Expires: November 19, 2015 M. Isomaki
Nokia
D. Wing
Cisco
May 18, 2015

Optimizing NAT and Firewall Keepalives Using Port Control Protocol (PCP)
draft-ietf-pcp-optimize-keepalives-06

Abstract

This document describes how Port Control Protocol is useful in reducing NAT and firewall keepalive messages for a variety of applications.

Status of This Memo

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 November 19, 2015.

Copyright Notice

Copyright (c) 2015 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.


Table of Contents

1. Introduction

Many types of applications need to keep their Network Address Translator (NAT) and Firewall (FW) mappings alive for long periods of time, even when they are otherwise not sending or receiving any traffic. This is typically done by sending periodic keep-alive messages just to prevent the mappings from expiring. As NAT/FW mapping timers may be short and unknown to the endpoint, the frequency of these keepalives may be high. An IPv4 or IPv6 host can use the Port Control Protocol (PCP)[RFC6887] to flexibly manage the IP address and port mapping information on NATs and Firewalls to facilitate communications with remote hosts. This document describes how PCP can be used to reduce keepalive messages for both client-server and peer-to-peer type of communication.

The mechanism described in this document is especially useful in cellular mobile networks, where frequent keepalive messages make the radio transition between active and power-save states causing congestion in the signaling path. The excessive time spent on the active state due to keepalives also greatly reduces the battery life of the cellular connected devices such as smartphones or tablets. [I-D.ietf-v6ops-mobile-device-profile] recommends cellular hosts to be PCP-compliant in order to save battery consumption exacerbated by keepalive messages.

2. Notational Conventions

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].

This note uses terminology defined in [RFC5245] and [RFC6887].

3. Overview of Operation

3.1. Application Scenarios

PCP can help both client-server and peer-to-peer applications to reduce their keepalive rate. The relevant applications are the ones that need to keep their NAT/FW mappings alive for long periods of time, for instance to be able to send or receive application messages in both directions at any time.

A typical client-server scenario is depicted in Figure 1. A client, who may reside behind one or multiple layers of NATs/FWs, opens a connection to a globally reachable server, and keeps it open to be able to receive messages from the server at any time. The connection may be a connection-oriented transport protocol such as TCP or SCTP or connection-less transport protocol such as UDP. Protocols operating in this manner include the Session Initiation Protocol (SIP) [RFC3261], the Extensible Messaging and Presence Protocol (XMPP) [RFC3921], the Internet Mail Application Protocol (IMAP) [RFC2177] with its IDLE command, the WebSocket protocol [RFC6455] and the various HTTP long-polling protocols. There are also a number of proprietary instant messaging, Voice over IP, e-mail and notification delivery protocols that belong in this category. All of these protocols aim to keep the client-server connection alive for as long as the application is running. When the application has otherwise no traffic to send, specific keepalive messages are sent periodically to ensure that the NAT/FW state in the middle does not expire. The client can use PCP to keep the required mappings at the NAT/FWs and use application keepalives to keep the state on the Application Server/Peer as mentioned in Section 3.4.

             

     PCP          PCP
    Client       Server      __________
+-----------+   +------+    /          \   +-----------+
|Application|___| NAT/ |____| Internet |___|Application|    
|  Client   |   |  FW  |    |          |   |   Server  | 
+-----------+   +------+    \__________/   +-----------+     
               (multiple
                layers)

       ------------> PCP

       ----------------------------------------->
               Application keepalive

        

Figure 1: PCP with Client-Server applications

There are also scenarios where the long-term communication association is between two peers, both of whom may reside behind one or more layers of NAT/FW. This is depicted in Figure 2. The initiation of the association may have happened using mechanisms such as Interactive Communications Establishment (ICE), perhaps first triggered by a "signaling" protocol such as SIP or XMPP or WebRTC [I-D.ietf-rtcweb-overview]. Examples of the peer-to-peer protocols include RTP and WebRTC data channel. A number of proprietary VoIP or video call or streaming or file transfer protocols also exist in this category. Typically the communication is based on UDP, but TCP or SCTP may be used. If there is no traffic flowing, the peers have to inject periodic keepalive packets to keep the NAT/FW mappings on both sides of the communication active. Instead of application keepalives, both peers can use PCP to control the mappings on the NAT/FWs to reduce the keepalive frequency as explained in Section 3.4.

             

     PCP          PCP                        PCP          PCP
    Client       Server      __________     Server       Client
+-----------+   +------+    /          \   +------+   +-----------+
|Application|___| NAT/ |____| Internet |___| NAT/ |___|Application|    
|   Peer    |   |  FW  |    |          |   |  FW  |   |    Peer   | 
+-----------+   +------+    \__________/   +------+   +-----------+  
               (multiple                  (multiple
                layers)                    layers)

       ------------> PCP                   PCP <------------

       <--------------------------------------------------->
                       Application keepalive

        

Figure 2: PCP with Peer-to-Peer applications

3.2. NAT Topologies and Detection

Before an application can reduce its keepalive rate, it has to make sure it has all of the NATs and firewalls on its path under control. This means it has to detect the presence of any PCP-unaware NATs and firewalls on its path to the Internet.

3.2.1. PCP based detection

PCP itself is able to detect unexpected NATs between the PCP client and PCP server as depicted in Figure 3. The PCP client includes its own IP address and UDP port within the PCP request. The PCP server compares them to the source IP address and UDP port it sees on the packet. If they differ, there are one or more additional NATs between the PCP client and PCP server, and the server will return an error. Unless the application has some other means (like UPnP) to control these PCP unaware NATs, it has to fall back to its default keepalive mechanism.

          

     PCP           PCP       PCP
    Client       Unaware    Aware       __________
+-----------+   +------+   +------+    /          \   +-----------+
|Application|___| NAT  |___| NAT/ |____| Internet |___|Application|    
|  Client   |   |      |   |  FW  |    |          |   |   Server  | 
+-----------+   +------+   +------+    \__________/   +-----------+ 

      <-----------///---------->   
          PCP based detection 
               

Figure 3: PCP unaware NAT between PCP client and PCP server

3.2.2. Application based detection

Figure 4 shows a topology where one or more PCP unaware NATs are deployed on the exterior of the PCP capable NAT/FWs. To detect this, the application client must have the capability to request from its application server or peer what IP and transport address it sees. If those differ from the IP and transport address given by the PCP aware NAT/FW then the application client can determine that there is at least one PCP unaware NAT on the path. In this case, the application client has to fall back to its default keepalive mechanism.

             

     PCP          PCP        PCP
    Client       Aware     Unaware      __________
+-----------+   +------+   +------+    /          \   +-----------+
|Application|___| NAT/ |___| NAT  |____| Internet |___|Application|    
|  Client   |   |  FW  |   |      |    |          |   |   Server  | 
+-----------+   +------+   +------+    \__________/   +-----------+    

      <------------>
            PCP 

      <---------------------///---------------------------> 
                 Application based detection
               

        

Figure 4: PCP unaware NAT external to the last PCP aware NAT

3.3. Detection of PCP unaware firewalls

PCP and application based detection mechanisms explained in Section 3.2.1 and Section 3.2.2 are based on change in the address and will not detect PCP unaware firewalls. In order to detect a PCP unaware firewall, the application client sends a Session Traversal Utilities for NAT (STUN) [RFC5389] Binding request to the STUN server. If STUN server supports the STUN extensions defined in [RFC5780] then it returns its alternate IP address and alternate port in OTHER-ADDRESS attribute in the STUN Binding response. The client then uses PCP to send MAP request with FILTER option to PCP server to permit STUN server to reach the client using the STUN servers alternate IP address and alternate port. The client then sends a Binding request to the primary address of the STUN server with the CHANGE-REQUEST attribute set to change-port and change-IP. This will cause the server to send its response from its alternate IP address and alternate port. If the client receives a response then the client is aware that on path firewall devices are PCP aware. If the client does not receive a response then the client is aware that there could be one or more on path PCP unaware firewall devices. The application client will perform the tests separately for each transport protocol. If no response is received, the client will then repeat the test at most three times for connectionless transport protocols.

             

   PCP            PCP        PCP
  Client         Aware     Unaware      __________
+-----------+   +------+   +------+    /          \   +-----------+
|Application|___| NAT/ |___| FW   |____| Internet |___|   STUN    |    
|  Client   |   | FW   |   |      |    |          |   |   Server  | 
+-----------+   +------+   +------+    \__________/   +-----------+    

      <--------------------------------------------------->
            STUN 

      <------------>
            PCP 
                              X<---------------------------
                 STUN based detection
               

        

Figure 5: PCP unaware firewall

This procedure can be adopted by other protocols to detect PCP unaware firewalls.

3.4. Keepalive Optimization

If the application determines that all NATs and firewalls on its path to the Internet support PCP, it can start using PCP instead of its default keepalives to maintain the NAT/FW state. It can use PCP PEER Request with the Requested Lifetime set to an appropriate value. The application may still send some application-specific heartbeat messages end-to-end to refresh state on the application server, which typically requires keepalives far less frequently than NATs /FWs do.

Processing the lifetime value of the PEER Opcode is described in Sections 10.3 and 15 of [RFC6887]. Sending a PEER request with a very short Requested Lifetime can be used to query the lifetime of an existing mapping. PCP recommends that lifetimes of mapping created or lengthened with PEER be longer than the lifetimes of implicitly-created NAT and firewall mappings. Thus PCP can be used to reduce power consumption by making PCP PEER message interval longer than what the application would normally use to the keep the middle box state alive, and strictly shorter than the server state refresh interval.

An example of savings with PCP is described in Appendix B.

4. Keepalive Interval Determination Procedure when PCP unaware Firewall or NAT is detected

If a PCP unaware NAT/firewall is detected, then a client can use the following heuristics method to determine the keepalive interval:

  1. The client sends a STUN Binding request to the STUN server. This connection is called the Primary Channel. STUN server will return its alternate IP address and alternate port in OTHER-ADDRESS in the Binding response [RFC5780].
  2. The client then sends a STUN Binding request to the STUN server using alternate IP address and alternate port. This connection is called the Secondary Channel.
  3. The Client will initially set the default keepalive interval for NAT/FW mappings to 60 seconds (FWa).
  4. After FWa seconds the Client will send a Binding request to the STUN server using the Primary Channel with the CHANGE-REQUEST attribute set to change-port and change-IP. This will cause the STUN server to send its response from the Secondary channel.
  5. If the client receives response from the server then it will increase the keepalive interval value FWa = (old FWa) + (old FWa)/2. This indicates that NAT/FW mappings are alive.
  6. Steps 4 and 5 will be repeated until there is no response from the STUN server. If there is no response from the STUN server then the client will use the old FWa value as Keepalive interval to refresh FW/NAT mappings.

The above procedure will be done separately for each transport protocol. For connectionless transport protocols such as UDP, if 2 seconds elapse without a response from the STUN server then the client will repeat step 4 at most three times to handle packet loss.

This procedure can be adopted by other protocols to use Primary and Secondary channels, so that the client can determine the keepalive interval to refresh FW/NAT mapping. This procedure only serves as a guideline and if applications already use some other heuristic to determine the keepalive interval, they can continue with the existing logic. For example Teredo determines the Refresh interval using the procedure in "Optional Refresh Interval Determination Procedure" (Section 5.2.7 of [RFC4380]).

Note: The keepalive interval learnt using the above method can be inaccurate if a firewall is configured with an application-specific inactivity timeout.

To improve reliability, applications SHOULD continue to use PCP to lengthen the FW/NAT mappings even if the above mechanism is used to a detect PCP unaware NAT/firewall. This ensures that PCP aware FW/NATs do not close old mappings with no packet exchange when there is a resource-scarcity situation.

5. Application-Specific Operation

This section describes how PCP is used with specific application protocols.

5.1. SIP

For connection-less transports the User Agent (UA) sends a STUN Binding request over the SIP flow as described in section 4.4.2 of [RFC5626]. The UA then learns the External IP Address and Port using a PCP PEER request/response. If the XOR-MAPPED-ADDRESS in the STUN Binding response matches the external address and port provided by PCP PEER response then the UA optimizes the keepalive traffic as described in Section 3.4. There is no further need to send STUN Binding requests over the SIP flow to keep the NAT Binding alive.

If the XOR-MAPPED-ADDRESS in the STUN Binding response does not match the external address and port provided by the PCP PEER response then PCP will not be used to keep the NAT bindings alive for the flow that is being used for the SIP traffic. This means that multiple layers of NAT are involved and intermediate NATs are not PCP aware. In this case the UA will continue to use the technique in section 4.4.2 of [RFC5626].

For connection-oriented transports, the UA sends a STUN Binding request multiplexed with SIP over the TCP connection. STUN multiplexed with other data over a TCP or TLS-over-TCP connection is explained in section 7.2.2 of [RFC5389]. The UA then learns the External IP address and port using a PCP PEER request/response. If the XOR-MAPPED-ADDRESS in the STUN Binding response matches the external address and port provided by the PCP PEER response, then the UA optimizes the keepalive traffic as described in Section 3.4.

If the XOR-MAPPED-ADDRESS in the STUN Binding response does not match the external address and port provided by the PCP PEER response, then PCP will not be used to keep the NAT bindings alive. In this case the UA performs a keepalive check by sending a double-CRLF (the "ping") then waits to receive a single CRLF (the "pong") using the technique in section 4.4.1 of [RFC5626].

5.2. HTTP

Web Applications that require persistent connections use techniques such as HTTP long polling and Websockets for session keep alive as explained in section 3.1 of [I-D.isomaki-rtcweb-mobile]. In such scenarios, after the client establishes a connection with the HTTP server, it can execute server side scripts such as PHP residing on the server to provide the transport address and port of the HTTP client seen at the HTTP server. In addition, the HTTP client also learns the external IP Address and port using a PCP PEER request/response.

If the IP address and port learned from the server matches the external address and port provided by the PCP PEER response then the HTTP client optimizes keepalive traffic as described in Section 3.4.

If the IP address and port do not match, then PCP will not be used to keep the NAT bindings alive for the flow that is being used for the HTTP traffic. This means that there are NATs or HTTP proxies between the PCP server and the HTTP server. The HTTP client will have to resort to use existing techniques for keep alive. Please see Appendix A for an example server side PHP script to obtain the client source IP address.

The HTTP protocol allows intermediaries such as transparent proxies to be involved and there is no way for the client to know that a request/response is relayed through a proxy.

5.3. Media and data channels with ICE

The ICE agent learns the External IP Addresses and Ports using the PCP MAP opcode. If server reflexive candidates learnt using STUN [RFC5389] and external IP addresses learnt using PCP are different then candidates learnt through both STUN and PCP are encoded in the ICE offer and answer . When using the Recommended Formula explained in section 4.1.2.1 of [RFC5245] to compute priority for the candidate learnt through PCP, the ICE agent MUST use a preference value greater than the server reflexive candidate and hence tested before the server reflexive candidate. The recommended type preference value is 105 for candidates discovered using PCP and is explained in section 4.2 of [RFC6544].

The ICE agent, in addition to the ICE connectivity checks, performs the following:

  1. The ICE agent checks if the XOR-MAPPED-ADDRESS from the STUN Binding response received as part of ICE connectivity check matches the External IP address and Port provided by PCP MAP response.
  2. If the match is successful then PCP will be used to keep the NAT bindings alive. The ICE agent optimizes keepalive traffic by refreshing the mapping via a new PCP MAP request containing information from the earlier PCP response.
  3. If the match is not successful then PCP will not be used for keep NAT binding alive. The ICE agent will use the technique in section 4.4 of [RFC6263] to keep NAT bindings alive. This means that multiple layers of NAT are involved and intermediate NATs are not PCP- aware.

Some network operators deploying a PCP Server may allow PEER but not MAP. In such cases the ICE agent learns the external IP address and port using a STUN Binding request/response during ICE connectivity checks. The ICE agent also learns the external IP Address and port using a PCP PEER request/response. If the IP address and port learned from the STUN Binding response matches the external address and port provided by the PCP PEER response then the ICE agent optimizes keepalive traffic as described in Section 3.4.

5.4. Detecting Flow Failure

Using the Rapid Recovery technique in section 14 of [RFC6887] upon receiving a PCP ANNOUNCE from a PCP server, a PCP client becomes aware that the PCP server has rebooted or lost its mapping state. The PCP client issues new PCP requests to recreate any lost mapping state and thus reconstructs lost mappings fast enough that existing media, HTTP and SIP flows do not break. If the NAT state cannot be recovered the endpoint will find the new external address and port as part of the Rapid Recovery technique in PCP itself and reestablish a connection with the peer.

5.5. Firewalls

PCP allows applications to communicate with firewall devices with PCP functionality to create mappings for incoming connections. In such cases PCP can be used by the endpoint to create an explicit mapping on firewall in order to permit inbound traffic. The endpoint can further use PCP to send keepalives to keep the firewall mappings alive.

5.5.1. IPv6 Network with Firewalls

For scenarios where the client uses the ICE Lite implementation explained in section 2.7 of [RFC5245], the ICE Lite endpoint will not generate its own ICE connectivity checks, by definition. As part of the call setup, the ICE Lite endpoint would gather its host candidates and relayed candidate from a TURN server and send the candidates in the offer to the peer endpoint. On receiving the answer from the peer endpoint, the ICE Lite endpoint sends a PCP MAP request with FILTER opcode to create a dynamic mapping in the firewall to permit ICE connectivity checks and subsequent media traffic from the remote peer. This way, the ICE Lite endpoint and its network are protected from unsolicited incoming UDP traffic, and can still operate using ICE Lite (rather than full ICE).

5.5.2. Mobile Network with Firewalls

Some mobile networks are also making use of a firewall to protect their customers from various attacks like downloading malicious content. The firewall is usually configured to block all unknown inbound connections as explained in section 2.1 of [I-D.chen-pcp-mobile-deployment]. As described in Section 3.4, in such cases, PCP can be used by mobile devices to create an explicit mapping on the firewall to permit inbound traffic and optimize the keepalive traffic. This would result in saving of radio and power consumption of the mobile device while protecting it from attacks.

6. IANA Considerations

This document has no actions for IANA.

7. Security Considerations

The security considerations in [RFC5245] and [RFC6887] apply to this use.

8. Acknowledgements

Authors would like to thank Dave Thaler, Basavaraj Patil, Anca Zamfir, Reinaldo Penno, Suresh Kumar, Dilipan Janarthanan and Mohamed Boucadair for their valuable inputs.

9. References

9.1. Normative References

[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, March 1997.
[RFC5245] Rosenberg, J., "Interactive Connectivity Establishment (ICE): A Protocol for Network Address Translator (NAT) Traversal for Offer/Answer Protocols", RFC 5245, April 2010.
[RFC5389] Rosenberg, J., Mahy, R., Matthews, P. and D. Wing, "Session Traversal Utilities for NAT (STUN)", RFC 5389, October 2008.
[RFC5626] Jennings, C., Mahy, R. and F. Audet, "Managing Client-Initiated Connections in the Session Initiation Protocol (SIP)", RFC 5626, October 2009.
[RFC5780] MacDonald, D. and B. Lowekamp, "NAT Behavior Discovery Using Session Traversal Utilities for NAT (STUN)", RFC 5780, May 2010.
[RFC6263] Marjou, X. and A. Sollaud, "Application Mechanism for Keeping Alive the NAT Mappings Associated with RTP / RTP Control Protocol (RTCP) Flows", RFC 6263, June 2011.
[RFC6544] Rosenberg, J., Keranen, A., Lowekamp, B. and A. Roach, "TCP Candidates with Interactive Connectivity Establishment (ICE)", RFC 6544, March 2012.
[RFC6887] Wing, D., Cheshire, S., Boucadair, M., Penno, R. and P. Selkirk, "Port Control Protocol (PCP)", RFC 6887, April 2013.

9.2. Informative References

[I-D.chen-pcp-mobile-deployment] Chen, G., Cao, Z., Boucadair, M., Ales, V. and L. Thiebaut, "Analysis of Port Control Protocol in Mobile Network", Internet-Draft draft-chen-pcp-mobile-deployment-04, July 2013.
[I-D.ietf-rtcweb-overview] Alvestrand, H., "Overview: Real Time Protocols for Browser-based Applications", Internet-Draft draft-ietf-rtcweb-overview-13, November 2014.
[I-D.ietf-v6ops-mobile-device-profile] Binet, D., Boucadair, M., Ales, V., Chen, G., Heatley, N., Chandler, R., Michaud, D., Lopez, D. and W. Haeffner, "An Internet Protocol Version 6 (IPv6) Profile for 3GPP Mobile Devices", Internet-Draft draft-ietf-v6ops-mobile-device-profile-21, March 2015.
[I-D.isomaki-rtcweb-mobile] Isomaki, M., "RTCweb Considerations for Mobile Devices", Internet-Draft draft-isomaki-rtcweb-mobile-00, July 2012.
[RFC2177] Leiba, B., "IMAP4 IDLE command", RFC 2177, June 1997.
[RFC3261] Rosenberg, J., Schulzrinne, H., Camarillo, G., Johnston, A., Peterson, J., Sparks, R., Handley, M. and E. Schooler, "SIP: Session Initiation Protocol", RFC 3261, June 2002.
[RFC3921] Saint-Andre, P., "Extensible Messaging and Presence Protocol (XMPP): Instant Messaging and Presence", RFC 3921, October 2004.
[RFC4380] Huitema, C., "Teredo: Tunneling IPv6 over UDP through Network Address Translations (NATs)", RFC 4380, February 2006.
[RFC6455] Fette, I. and A. Melnikov, "The WebSocket Protocol", RFC 6455, December 2011.

Appendix A. Example PHP script

<html>
Connected to <?PHP echo gethostname(); ?> on port <?PHP echo
getenv(SERVER_PORT)?> on <?PHP echo date("d-M-Y H:i:s");?> 
Pacific Time
<p>
Your IP address is: <?PHP echo getenv(REMOTE_ADDR); ?>,
port <?PHP echo getenv(REMOTE_PORT); ?>
</p>;
</html>

Appendix B. Savings with PCP

The following example illustrates the savings in keepalive messages with PCP.

             

     PCP               PCP        
    Client            Server         __________
+-----------+        +------+       /          \   +-----------+
|Application|________| NAT/ |_______| Internet |___|Application|    
|  Client   |        |  FW  |       |          |   |   Server  | 
+-----------+        +------+       \__________/   +-----------+    

With Application Heartbeat (without PCP):

      <---------------------///---------------------------> 
        Application heartbeat (Max Interval = 30 seconds)
      <---------------------///---------------------------> 
        Application heartbeat (Max Interval = 30 seconds)
      <---------------------///---------------------------> 
        Application heartbeat (Max Interval = 30 seconds)
      <---------------------///---------------------------> 
        Application heartbeat (Max Interval = 30 seconds)
             ....
             ....
             ....
             ....

With PCP:
      <------------------>
        PCP PEER request 
     (Max Lifetime = 3600 seconds)
             ....
             ....              
      <------------------>
        PCP PEER request 
     (Max Lifetime = 3600 seconds)
    

Figure 6: Savings with PCP

In the example above, let's suppose normally an application would need to send a heartbeat every 30s to keep mappings active on the NAT/firewall device. In 24 hours, in the absence of PCP, the number of packets sent by the application to keep those mappings active would be (86400/30) = 2880 packets.

If the same application uses PCP PEER to create a mapping, with a lifetime of 3600 seconds, on a PCP controlled NAT/firewall device, the number of packets sent by the application to keep those mappings active would be (86400/3600) = 24 packets.

With the above assumptions, using PCP saves 99.16% of keepalive traffic. As the number of applications running on a host increase, savings in cost of sending application heartbeats are significant with the use of PCP.

        PCP                PCP              PCP
       Client             Proxy/           Server
                          Server                     __________
   +-----------+         +------+         +------+  /          \   +-----------+
   |Application|_________| NAT/ |_________| NAT/ |__| Internet |___|Application|
   |  Client   |         |  FW  |         |  FW  |  |          |
   +-----------+         +------+         +------+  \__________/   +-----------+
                           <-multiple NAT/FW->

      <------------------>  <------------------>
        PCP PEER request      PCP PEER request    

If there are multiple PCP-aware NAT/firewall devices on a client's path to the internet, e.g., PCP servers at a home gateway and also at a CGN, the savings with PCP are the same. The PCP server at the home gateway is a PCP proxy that can create associated mappings on the PCP server at the CGN. The client will only have to communicate with the PCP proxy, and receives a single mapping lifetime that needs to be refreshed.

Authors' Addresses

Tirumaleswar Reddy Cisco Systems, Inc. Cessna Business Park, Varthur Hobli Sarjapur Marathalli Outer Ring Road Bangalore, Karnataka 560103 India EMail: tireddy@cisco.com
Prashanth Patil Cisco Systems, Inc Bangalore, India EMail: praspati@cisco.com
Markus Isomaki Nokia Keilalahdentie 2-4 FI-02150 Espoo, Finland EMail: markus.isomaki@nokia.com
Dan Wing Cisco Systems, Inc. 170 West Tasman Drive San Jose, California 95134 USA EMail: dwing@cisco.com