Patent Application
20110317705
Embodiments of this invention are related to computer networks and more specifically to peer-to-peer communications across symmetric network address translators on computer networks.
The use of routers with a NAT (Network Address Translation) feature can interfere in accessing an internal network from an external network. This can be a particular problem for peer-to-peer applications such as voice communication over the Internet (known as VoIP) and/or online gaming, etc. NAT is an Internet standard that enables a local area network (LAN) to use of one set of private IP addresses for internal traffic and a second set of global IP addresses for external traffic. A node that has NAT capability is often referred as “NAT box”.
A NAT (literally) translates network (IP) address between the two networks. Network Address Port Translation (NAPT) translates not only IP address but also port numbers of a transport layer protocol. Although NAT/NAPT has its good properties, there is a significant side effect. If the translation is performed dynamically, nodes in the external network have no way to know the IP address (and the port number) on the NAT ahead of time to reach a node in the internal network. Unfortunately, this is the most common behavior of NAT in the residential and SOHO routers deployed in the current market.
A NAT can generally be categorized as being Full Cone, Restricted Cone, Port Restricted Cone or Symmetric. A full cone NAT maps all requests from the same internal IP address and port to the same external IP address and port. Furthermore, any external host can send a packet to the internal host through a full cone NAT by sending a packet to the mapped external address. In a restricted cone NAT all requests from the same internal IP address and port are mapped to the same external IP address and port. Unlike a full cone NAT, an external host with IP address X can send a packet to the internal host only if the internal host had previously sent a packet to IP address X. A port restricted cone NAT is like a restricted cone NAT, but the restriction also includes port numbers. Specifically, an external host can send a packet, with source IP address X and source port P, to the internal host only if the internal host had previously sent a packet to IP address X and port P.
In a symmetric NAT all requests from the same internal IP address and port, to a specific destination IP address and port, are mapped to the same external IP address and port. If the same host sends a packet with the same source address and port, but to a different destination, a different mapping is used. Furthermore, only the external host that receives a packet can send a UDP packet back to the internal host. The symmetric NAT tends to be the most problematic type of NAT to traverse. One technique for symmetric NAT traversal is known as “port prediction”, which is described in detail in US Patent Application publication 20070076729A1, which is incorporated herein by reference. In this type of symmetric NAT traversal, a first node is behind a first NAT that is symmetric and a second node that is behind a second NAT. The first node constructs a list of predicted transport addresses on the first NAT and sends a message containing the list of predicted transport addresses to the second node. A connectivity check is performed with the second node using the predicted transport addresses.
It has been estimated that 18% of NATs are Symmetric and a connection failure rate of more than 10% is anticipated without port prediction. Some applications involving NAT traversal may require up to 64 simultaneous connections. It is not clear whether port prediction can reliably work for such applications.
It is within this context that embodiments of the present invention arise.
The teachings of the present invention can be readily understood by considering the following detailed description in conjunction with the accompanying drawings, in which:
Although the following detailed description contains many specific details for the purposes of illustration, anyone of ordinary skill in the art will appreciate that many variations and alterations to the following details are within the scope of the invention. Accordingly, examples of embodiments of the invention described below are set forth without any loss of generality to, and without imposing limitations upon, the claimed invention.
Port prediction technique is used to traverse symmetric NAT (Network Address Translator). If a node behind a symmetric NAT is trying to connect many remote nodes at a time, prediction could fail with a higher probability. Embodiments of the invention avoid the connection failure caused by the case of simultaneous port prediction by putting new requests on hold until a previous request has reached a connectivity check stage. This allows a node multiple simultaneous connections through a symmetric NAT without causing problems with port prediction.
As used herein, the following terms have the meanings shown in Table I below.
As used herein, the following acronyms have the meanings shown in Table II below.
The basic problem of port prediction failure may be understood with respect to
It is noted that a critical time window TC1 exists for the first node 2 between the sending of the binding request 10 and the sending of the first of the check packets 22. During this time port prediction performed by the first node 2 is based on the address information 12 sent by the STUN server 6 in response to the binding request 10. If that address information should change during this critical time, the port prediction may fail. The address information may change if the first node 2 initiates another binding request for communication with a third node (not shown). A similar critical time window TC2 exists for the second node 4 between receipt of the connection request 14 from the first node 2 and the sending of the first check packets 24.
In general, the critical time window for NAT traversal ends when the first set of connectivity check packets are transmitted. Referring to
The Symmetric NAT traversal described with respect to
Embodiments of the present invention overcome problems with multiple port predictions by serializing “critical windows” during the negotiation phase of NAT traversal, while establishing many connections at a time. It is noted that certain applications, such as multi-player online game applications can support the establishment of multiple peer-to-peer connections.
According to an embodiment of the present invention, serialization of the critical time window may be implemented on a node behind a symmetric NAT that attempts to initiate peer-to-peer connections over a network with two or more other nodes. In particular, the first node may perform a port prediction for initiating a communication session with a second node with first node and construct a list of predicted transport addresses on the symmetric NAT and then send a CONNECTION REQUEST message containing the list of predicted transport addresses to the second node. Upon receipt of a provisional response to the CONNECTION REQUEST message. The first node may then perform a check of connectivity between the first node and the second node using the predicted transport addresses, e.g., by sending test packets. The first node may serialize the critical time window by delaying port prediction for communication between the first node and a third node until after the connectivity check has begun.
In a similar fashion, serialization of the critical time window may be implemented on a node behind a symmetric NAT that receives a CONNECTION REQUEST to initiate a peer-to-peer connection over a network while negotiating NAT traversal with one or more other nodes. When such a node receives a CONNECTION REQUEST message from another node it may perform port prediction and send a provisional response to the CONNECTION REQUEST message with a list of predicted addresses and perform a check of connectivity. Port prediction for communication with an additional node may be delayed until after the connectivity check has begun.
Specifically, as shown in the timing diagram of
As discussed above, a connection may fail if it cannot be established within a timeout period TO. Consequently, it is desirable that the timeout period TO is sufficiently long that all N connections can be established. By way of example, assuming that all of the critical time windows TO for each connection are of some be less than some maximum number of connections the timeout period TO should be less than NmaxTC+Tconn where Tconn is the time for the connectivity check and Nmax is some maximum number of possible simultaneous connections.
As shown in
The nodes 102, 104 can be, e.g., server hosts such as audio/video (A/V) chat, multimedia streaming devices, file sharing nodes, online gaming modules etc. Each node 102, 104 may be a general purpose computer that becomes a special purpose computer when running instructions, such as instructions for implementing the steps of the method of
A method of NAT traversal according to embodiments of the present invention may be understood by simultaneously referring to the block diagram of
By way of example, the first node 102 may wish to establish a peer-to-peer connection to the second node 104. The first node 102 allocates a local port 107 for the new peer-to-peer session. Then the first node 102 obtains an external port 112 by sending a binding request 118 from the local port 107 to the STUN server 101. The sending of the binding request begins the critical time window for the first node 102. Since the first node 102 knows that the first NAT 103 is present and is of type Symmetric, it may perform port prediction as indicated at 210 and construct a list of transport addresses 107, 110, 112, 113 and 114. The list may be put into a new CONNECTION REQUEST message. In a preferred embodiment, the first node 102 may send no information about the first NAT 103. Furthermore, the first node 102, not the second node 104, may perform the port prediction. In addition, sending CONNECTION REQUEST messages containing transport addresses is perfectly compatible with the existing ICE methodology.
To prevent a subsequent binding request for connection to another node from interfering with port prediction, the first node 102 may begin a wait period upon sending the binding request 118 as indicated at 209. During the wait period, new binding requests for communication with other nodes may be temporarily put on hold and queued in the order in which they originated.
At step 212, the first node 102 sends the CONNECTION REQUEST message with the list of transport addresses to a SIP proxy server 100 through the already established path 116. The SIP proxy server 100 finds in the message that the final destination is the second node 104 and forwards the CONNECTION REQUEST message through the already establish path 117, and port 111 and finally the message reaches the second node 104 on local port 108. On reception of the CONNECTION REQUEST, the second node 104 may allocate a local port 109 for the future peer-to-peer session, and then the second node 104 may obtain an external port 115 by sending a binding request 119 from the local port 109 to the STUN server 101. To prevent a subsequent binding request for connection to another node from interfering with port prediction, the second node 104 may begin a wait period upon receiving the CONNECTION REQUEST message as indicated at 211. During the wait period, the second node 104 may temporarily put new binding requests for communication with other nodes on hold and queue them in the order in which they originated. Specifically, Node 1 102 may put a request 217 from Node X 125 in a queue in the order in which the request was received. Node 2 104 may similarly queue a request 219 from node Y 127.
Since the second node 104 knows that the second NAT 105 is not Symmetric, it puts the local port 109 and the external port 115 in a new provisional response message and sends it back to the first node 102 at 214 via the SIP proxy server 100 and the first and second signaling paths 117, 116. The transmission of the provisional response terminates the transport exchange phase and starts connectivity check phase. At this stage, the first and second nodes 102, 104 may safely initiate new binding requests for communication with other nodes. Consequently, the wait period for the first and second nodes 102, 104 may end, as indicated at 213, 215 respectively
To check connectivity, both nodes 102 and 104 may start sending STUN packets from their local ports 107, 109 to check connectivity to the transport addresses obtained from the other node at steps 216, 218. When the first node 102 sends a STUN packet 120, the first NAT 103 allocates a new external port 113, and then the packet 120 reaches the external port 115 on the second NAT 105. The first few packets could be discarded at the external port 115 because the second NAT 105 is a port restricted cone NAT and the second node 104 might not yet have sent a packet from the local port 109 to the external port 113 on the first NAT 103. The second node 104 also sends STUN packets 121, 122, 123 to the obtained transport addresses 112, 113 and 114. The packet 121 reached at the port 112 is discarded because the first NAT 103 is a Symmetric NAT and the port 112 is exclusively allocated for the session to the STUN server 101. The STUN packet 123 reached at 114 is also discarded because there is no such external port allocated by the first NAT 103. The STUN packet 122 reached at 113 is forwarded by the first NAT 103 to the local port 107. The first node 102 then sends a response to the second node 104 and the second node 104 finds it has connectivity to the port 113 on the reception of the response. The STUN packet sent from the local port 107 on the first node 102 to the external port 115 on the second NAT 105 is eventually received at the local port 109 of second node 104. The second node 104 then sends a response back to the first node 102.
Upon reception of the response message, at step 220 the first node 102 sends an UPDATE message to the second node 104 via the SIP proxy server 100 to tell the second node 104 that the first node 102 found connectivity to the external port 115. This triggers the second node 104 to send a final response message to the first node 102 at step 222 to finalize the connection establishment process.
Instead of putting a NAT type in the message and having the second node 104 to make a prediction, the first node 102 makes a prediction and puts the predicted external ports 113, 114 along with the external port 112 obtained from STUN server 101, in the new CONNECTION REQUEST message. Thus, the first node 102 provides no information about the first NAT 103 to the second node 104. Such use of ICE method completely eliminates complicated NAT combination logic for “break-out” packets as done in the prior art (US 2004/0139228). Instead, embodiments of the present invention can achieve the same result by performing a connectivity check that is essentially equivalent to the “break-out packet”. Thus, embodiments of the invention allow a system that already uses ICE methodology to add Symmetric NAT traversal capability by simply adding the predicted transport addresses to the connectivity check list.
As described above, the first node 102, i.e., the node attempting to initiate communication with the second node 104 performs a port prediction and puts the predicted ports in the CONNECTION REQUEST message. There are a number of techniques of performing the port prediction. For example, port prediction may be implemented using a port allocation rule discovery process using the following test. The first node 102 sends a STUN binding request to the STUN server 101 without any flags set in the CHANGE-REQUEST attribute and without the RESPONSE-ADDRESS attribute. This causes the STUN server 101 to send responses back to the address and port that the request came from. This test is applied to different combinations of IP addresses and ports in order to figure out the port allocation characteristics of the NAT 103. The STUN server 101 uses two different IP addresses, CA and DA and two different ports Cp and Dp as shown in Table III below.
As can be seen from Table III, the test is performed four times (e.g., from TRY-1 to TRY-4) per local port in this example. All the tests must be done from the same local port. The first node 102 obtains four mapped addresses from the responses. These four mapped addresses are analyzed to determine the port allocation rule and a port increment value ΔP and to evaluation consistency. To look for consistency, the process can be performed multiple times, preferably using a different local port that does not have a NAT binding associated with it. The port allocation rule can be determined by looking at the port numbers obtained from the mapped addresses. If all port numbers are incremented for successive destinations having different port numbers, the port allocation rule is said to be “port sensitive”. If the port increment size from successive destinations having the same IP address (e.g., from TRY-1 to TRY-2 and from TRY-3 to TRY-4) is always zero, but the incremental size from successive destinations having different IP addresses (e.g., TRY-2 to TRY-3) is not zero, the port allocation rule is said to be “address sensitive”. If all port numbers of the obtained mapped addresses are the same, the NAT 103 is a ‘Cone NAT’.
The ΔP value may be determined as follows. For address sensitive allocation, the ΔP value is equal to a port increment size between successive tries for which the destination port is different, e.g., for TRY-2 and TRY-3. The process may be repeated from another local port as shown in Table IV. In this example, TRY-1 through TRY-4 are as in TABLE III and TRY-5-TRY 8 continue the pattern destination IP addresses and port numbers of Table III.
Note that from Table IV it can be seen that where the destination IP address is the same for successive tries, the port numbers in the corresponding mapped addresses are the same. From this it can be determined that the port allocation rule is “address sensitive”. Furthermore it can be seen that the value of ΔP is equal to the port increment between TRY-2 and TRY-3 and is also equal to the port increment sizes between TRY-4 and TRY-5 and between TRY-6 and TRY-7.
For port sensitive allocation, the value of ΔP is the difference between adjoining port numbers of mapped addresses obtained from testing TRY-[N+1] and TRY-[N]. In situations where the first node 102 cannot find consistency in the port increment size for ΔP determination, the application may include an algorithm to determine the ΔP value based on statistical observation, or to decide to give up obtaining a valid ΔP.
If the second NAT 105 is not symmetric, it may be sufficient to perform port prediction, e.g., as part of the NAT discovery step 202 for the just the first node 102. In the case where the second NAT 105 is also a Symmetric NAT, the second node 104 may perform similar port prediction as part of its NAT discovery phase 204.
It is noted that in embodiments of the invention, a node may be configured to serialize the critical time windows for NAT traversal may be serialized for multiple connections to other nodes. Serializing the critical time windows as described above may potentially give rise to a possible pitfall referred to herein as a Symmetric Node Connection Loop. It is noted that queuing connection requests to serialize the critical time windows is not generally necessary for a node that is not behind a symmetric NAT. However, a lockup may occur if critical time window serialization is implemented for three or more nodes, each of which is behind a symmetric NAT. The nature of this pitfall may be appreciated with reference to
In the example depicted in
By way of example, Node C could send a “QUEUED” message Q to Node B. The QUEUED message Q indicates that Node C is behind Symmetric NAT and is waiting for a connection to another node. Node B could wait for a predetermined wait time Tw, then cancel its initiation with Node C and then subsequently initiate again with C after processing the connection request from Node A. The wait time Tw may be greater than or equal to zero and less than a timeout time that Node B would ordinarily wait before giving up on communication with Node C. A little more specifically, C sends the “QUEUED” message Q to B because C knows that C is behind a symmetric NAT and the request was not processed right away but was queued. Node B could wait a relatively shorter time period (e.g. 3 seconds) in response to the “QUEUED” message, then cancel (put off) the connection process to Node C, process other requests in the queue, then eventually initiate again with C. To prevent Node B from retrying to initiate with Node C indefinitely, the number of retries to the same node may be limited, e.g., up to 3 times, after which, Node B could completely give up. The application running on Node B would receive a connect error in such a case.
As an alternative solution to the lockup, the technique described above may be modified in the following way. The critical time windows may be serialized for outgoing connection requests only and slightly deeper port predictions may be performed. For example, instead of generating a port prediction based on the port increment ΔP, additional port predictions may be generated based on 2ΔP, 3ΔP . . . MAP, where M is an integer referred to herein as the “depth” of the prediction. By way of example, suppose it is determined that for a particular symmetric NAT the port increment ΔP=1. A candidate list based on a port prediction of depth M=1 based on this value of ΔP may look like this:
Unlike the use of the “QUEUED” signaling message, the foregoing solution may not address every possible lockup situation, but it may be simpler to implement. Furthermore, since connection processes more likely occur in parallel at calling side, serializing the outgoing connections, but not the incoming connections can avoid the most likely lockup situations. Since the critical time windows for incoming connections are not serialized, port prediction may fail. Increasing the depth of the port prediction, e.g., from M=+1 to M=+3 increases the number of predicted candidates, which decreases the likelihood of port prediction failure.
The examples illustrated below with
The NAT traversal algorithm may be implemented in software or hardware or a combination of both. By way of example,
The memory 402 may be in the form of an integrated circuit, e.g., RAM, DRAM, ROM, and the like). The memory 402 may also be a main memory or a local store of a synergistic processor element of a cell processor. A computer program 403 that includes the frame reconstruction algorithm described above may be stored in the memory 402 in the form of processor readable instructions that can be executed on the processor module 401. The processor module 401 may include one or more registers 405 into which instructions from the program 403. The instructions of the program 403 may include the steps of the method for peer to peer connection over a network, e.g., as described above with respect to
The program 403 may be written in any suitable processor readable language, e.g., C, C++, JAVA, Assembly, MATLAB, FORTRAN and a number of other languages. The apparatus may also include well-known support functions 410, such as input/output (I/O) elements 411, power supplies (P/S) 412, a clock (CLK) 413 and cache 414. The apparatus 400 may optionally include a mass storage device 415 such as a disk drive, CD-ROM drive, tape drive, or the like to store programs and/or data. The apparatus 400 may also optionally include a display unit 416 and user interface unit to facilitate interaction between the device and a user. The display unit 416 may be in the form of a cathode ray tube (CRT) or flat panel screen that displays text, numerals, graphical symbols or images. The display unit 416 may also include a speaker or other audio transducer that produces audible sounds. The user interface 418 may include a keyboard, mouse, joystick, light pen, microphone, or other device that may be used in conjunction with a graphical user interface (GUI). The apparatus 400 may also include a network interface 420 to enable the device to communicate with other devices over a network, such as the internet. These components may be implemented in hardware, software or firmware or some combination of two or more of these.
An embodiment of the invention was tested with a peer-to-peer library that implements NAT traversal features, using a simple connectivity check tool running on top of the peer-to-peer library. Nodes at both ends of the connection were behind Symmetric NATs. A peer-to-peer library node implemented on a PlayStation 3 development station attempted to connect to 64 other nodes running on a Linux PC. Critical time windows for both outbound and inbound connections were serialized. All 64 connections successfully established.
While the above is a complete description of the preferred embodiment of the present invention, it is possible to use various alternatives, modifications and equivalents. Therefore, the scope of the present invention should be determined not with reference to the above description but should, instead, be determined with reference to the appended claims, along with their full scope of equivalents. Any feature described herein, whether preferred or not, may be combined with any other feature described herein, whether preferred or not. In the claims that follow, the indefinite article “A”, or “An” refers to a quantity of one or more of the item following the article, except where expressly stated otherwise. The appended claims are not to be interpreted as including means-plus-function limitations, unless such a limitation is explicitly recited in a given claim using the phrase “means for.”
This Application claims the benefit of priority of commonly-assigned, co-pending U.S. patent application Ser. No. 12/043,080, now U.S. Pat. No. 7,856,506, to Yutaka Takeda, filed Mar. 5, 2008 and entitled TRAVERSAL OF SYMMETRIC NETWORK ADDRESS TRANSLATOR FOR MULTIPLE SIMULTANEOUS CONNECTIONS, the entire contents of which are incorporated herein by reference. This Application claims the benefit of priority of commonly-assigned, co-pending U.S. patent application Ser. No. 12/910,624 to Yutaka Takeda, filed Oct. 22, 2010 and entitled TRAVERSAL OF SYMMETRIC NETWORK ADDRESS TRANSLATOR FOR MULTIPLE SIMULTANEOUS CONNECTIONS, the entire contents of which are incorporated herein by reference.
| Number | Date | Country | |
|---|---|---|---|
| Parent | 12043080 | Mar 2008 | US |
| Child | 13224253 | US | |
| Parent | 12910624 | Oct 2010 | US |
| Child | 12043080 | US |
