Internet Engineering Task Force G. Chen
Internet-Draft China Mobile
Intended status: Informational C. Williams
Expires: December 25, 2014 Consultant
D. Wing
A. Yourtchenko
Cisco Systems, Inc.
June 23, 2014
Happy Eyeballs Extension for Multiple Interfaces
draft-ietf-mif-happy-eyeballs-extension-05
Abstract
Currently the interface selection in multi-interface environment is
exclusive - only one interface can be used at the time, frequently
needing manual intervention. Happy Eyeballs in MIF would make the
selection process smoother by using the connectivity checks over a
pre-filtered interfaces according to defined policy. This would
choose the fastest interface with an automatic fallback.
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 December 25, 2014.
Copyright Notice
Copyright (c) 2014 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
Chen, et al. Expires December 25, 2014 [Page 1]
Internet-Draft happy-eyeballs-mif June 2014
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 . . . . . . . . . . . . . . . . . . . . . . . . 2
2. Use Cases . . . . . . . . . . . . . . . . . . . . . . . . . . 3
3. Happiness Parameters . . . . . . . . . . . . . . . . . . . . 5
4. HE Behavior in MIF . . . . . . . . . . . . . . . . . . . . . 6
4.1. First Step, Filter . . . . . . . . . . . . . . . . . . . 6
4.2. Second Step, Sort . . . . . . . . . . . . . . . . . . . . 7
5. Implementation Framework . . . . . . . . . . . . . . . . . . 8
6. Additional Considerations . . . . . . . . . . . . . . . . . . 8
6.1. Usage Scope . . . . . . . . . . . . . . . . . . . . . . . 8
6.2. Fallback Timeout . . . . . . . . . . . . . . . . . . . . 8
6.3. DNS Selections . . . . . . . . . . . . . . . . . . . . . 9
6.4. Flow Continuity . . . . . . . . . . . . . . . . . . . . . 10
7. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 10
8. Security Considerations . . . . . . . . . . . . . . . . . . . 10
9. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . 10
10. References . . . . . . . . . . . . . . . . . . . . . . . . . 10
10.1. Normative References . . . . . . . . . . . . . . . . . . 10
10.2. Informative References . . . . . . . . . . . . . . . . . 11
Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 11
1. Introduction
In multiple interface context, the problems raised by hosts with
multiple interfaces have been discussed. The MIF problem
statement[RFC6418] described the various issues when using a wrong
domain selection on a MIF node. Happy Eyeballs (HE) [RFC6555]
described how a dual-stack client can determine the functioning path
to a dual-stack server. It's using stateful algorithm to help
applications quickly determine if IPv6 or IPv4 is the most fast path
to connect a server. That is a good method to achieve smart path
selection. However, the assumption is a single-homed context. The
interaction with multiple interfaces is deferred for further study.
[I-D.ietf-mif-mpvd-arch] has proposed a multiple provisioning domain
architecture. This memo has been proposed to extend happy eyeballs
algorithm to fit into the multiple interfaces architecture. Several
additional considerations have been elaborated to analyze the user
demands and initiate HE-MIF connections. It allows a node with
multiple interfaces picking a fast flow path.
Chen, et al. Expires December 25, 2014 [Page 2]
Internet-Draft happy-eyeballs-mif June 2014
2. Use Cases
The section enumerates several concrete use cases in existing
networks.
Case 1: WiFi is broken
o [Scenario] A MIF node has both 3/4G and WiFi interface. When the
node enters a WiFi area, a common practice would always prefer
WiFi because it' cheap and fast-speed normally.
o [Problem] User assumes the WiFi is working, because the node
already got IP address from WiFi. However, he can't run
applications due to Internet connectivity being unavailable. This
may be an authentication required coming into play, or unstable
Layer 2 conditions. In order to figure out the problems, users
have to turn off the WiFi manually.
o [Workaround] Users can indicate their desire with some setting on
the phone. For instance, they may prefer to wait a little bit of
time but not forever. After the timer is expired, users would
finally give up the WiFi path and try to establish connection over
3G path. Users may won't want the wait time too short, because
the 3G path for most people is more expensive than WiFi path.
Case 2: VPN (Virtual Private Network) scenario
o [Scenario] In some cases, a node has multiple interface because of
VPN. Users would only have interests to connect a corporate
network inside VPN. While, connecting to Internet would work
outside the VPN.
o [Problem] That is normally a implementation consideration that
unmanaged interface may be considered less trustworthy than
managed. It results in trusted interfaces having the highest
priority. This setting may steer all traffic to VPN interface.
When this is a traffic heading to a corporate site, everything is
fine. But sometimes, the connections out to Internet sites may
suffer from long-distance path delays.
o [Workaround] It's desirable if routing could be bound to each
interface. However, a node following weak host model[RFC1122]
takes routing tables as node-scoped. Some sophisticated VPN
clients may enable split tunneling to access a public network
(e.g. the Internet) and a local LAN simultaneously. VPN may adopt
different security policies to applications, for example disable
split tunneling to enforce security tunnel only to a particular
application. It may be useful to facilitate the implementation by
Chen, et al. Expires December 25, 2014 [Page 3]
Internet-Draft happy-eyeballs-mif June 2014
performing parallel IP connectivity checks before selecting an
interface.
Case 3: 3G/LTE tethering scenario
o [Scenario] Many mobile phones are equipped with software to offer
tethered Internet access. It shares their Internet connection
with another Internet-capable mobile phone or other devices over
Wi-Fi.
o [Problem] The tethering WiFi link is not free , i.e., it might be
3/4G backhaul. The policy of "always WiFi" leads to all traffic
being sent over the tethering WiFi. Usually, such tethering WiFi
link puts sharing limitation to access nodes. It could cause
contention on both that WiFi link and the backhaul 3G link, while
it be higher cost than going on the 3/4G that is built in the
handset.
o [Workaround] To solve that, it is necessary for the node to be
aware of not only the link layer information, but also services
information, like billable or free. That could help to facilitate
the execution of the algorithm. Same concern has been documented
in Section 4.4 of [RFC6418])
Case 4: Policy Conflict
o [Scenario] A node has WiFi and 3/4G access simultaneously. In
mobile network, IPv6-only may be preferable since IPv6 has the
potential to be simpler than dual-stack. WiFi access still
remains on IPv4.
o [Problem] The problem is caused by policy confliction. The
transition to IPv6 is likely to encourage IPv6 and prefer
IPv6[RFC6724]. If the 3/4G path has IPv6 on it and the WiFi does
not, a suboptimal interface might be chosen from the cost saving
perspective.
o [Workaround] Users interests should be well understood and
considered before interface selection. The different
preconditions may impact subsequent behaviors. Users concern
about high-reliability or high-speed or less-cost should make
different choice. A flexible mechanism should be provided allow
to make smart decision.
Chen, et al. Expires December 25, 2014 [Page 4]
Internet-Draft happy-eyeballs-mif June 2014
3. Happiness Parameters
This section provides the design proposal for HE-MIF. Two sets of
"Happiness" parameters have been defined. It serves upper
applications and initiates HE-MIF connections to below level API
subsequently. Going through the process, MIF nodes could pick an
appropriate interface which would correspond to user demands. The
two sets of "Happiness" parameters are called Hard set and Soft set
respectively.
o Hard Set: It contains parameters which have mandatory indications
that interface behavior should comply with. This might provide an
interface for applications constraints or delivering operator's
policies. Basically, parameters in Hard set should be easy-to-use
and easy-to-understand. The users would directly use those. When
several hard parameters were conflicted, user's preference should
be prioritized.
* User's preference: users would express preferences which may
not have a formally technical language , like "No 3G while
roaming", "Only use free WiFi", etc.
* Operator policy: operators would deliver the customized
policies in particular network environments due to geo-location
or services regulation considerations. One example in 3GPP
network is that operator could deliver policies from access
network discovery and selection function (ANDSF).
o Soft Set: It contains factors involving in the selection of the
fastest path. The following is considered as for the
justification.
* PVD-ID (Provisioning Domain Identity): PVD-aware node may
decide to use one preferred PVD or allow use mulitple PVDs
simultanenously for applications. The node behavior should be
consistent with MPVD architecture.
* Next hop: [RFC4191] allows configuration of specific routes to
a destination.
* DNS selection: [RFC6731] could configure nodes with information
to indicate DNS server address for a particular namespace.
* Source address selection: the information provided by [RFC6724]
would be considered.
Chen, et al. Expires December 25, 2014 [Page 5]
Internet-Draft happy-eyeballs-mif June 2014
* Other factors: There is a common practice may impact interface
selection, e.g. WiFi is preferable. Such conventional
experiences should also be considered.
4. HE Behavior in MIF
Corresponding to the two sets of parameters, a HE-MIF node may take a
two-steps approach. One is to do "Hard" decision to synthesize
policies from different actors (e.g., users and network operator).
In a nutshell, that is a filter which will exclude the interfaces
from any further consideration. The second is to adjust how we make
a connection on multiple interfaces after the filter. It's sorting
behavior. In the multiple provisioning domain architecture, a PVD
aware node takes connectivity tests as described in Section 5.3 of
[I-D.ietf-mif-mpvd-arch]. A PVD agnostic node take other parameters
in the Soft Set to proceed the sort process.
Those two steps are described as following sub-sections. It should
be noted that HE-MIF doesn't prescribe such two-step model. It will
be very specific to particular cases and implementations. For
example, if one interface on a particular PVD is left after the first
step, the process would be ceased.
4.1. First Step, Filter
One goal of filter is to reconcile multiple selection policies from
users or operators. Afterwards, the merged demands would be mapped
to a set of candidate interfaces, which is judged as qualified.
Decision on reconciliation of different policies will depend very
much on the deployment scenario. An implementation may not be able
to determine priority for each policies without explicit
configuration provided by users or administrator. For example, an
implementation may by default always prefer the WiFi due to cost
saving consideration. Whereas, users may dedicatedly prefer 3/4G
interface to seek high-reliability or security benefits even to
manually turn off WiFi interface. The decision on mergence of
policies may be made by implementations, by node administrators, even
by other standards investigating customer behavior. However, it's
worth to note that a demand from users should be normally considered
higher priority than from other actors.
The merged policies would serve as a filter principle doing iterate
across the list of all known interfaces. Qualified interface would
be selected to sort processing at next step.
Chen, et al. Expires December 25, 2014 [Page 6]
Internet-Draft happy-eyeballs-mif June 2014
4.2. Second Step, Sort
Sort process guarantees fast interface selection with fallback
capacities. As stated in [I-D.ietf-mif-mpvd-arch], a PVD-aware node
shall perform connectivity test and, only after validation of the
PVD, consider using it to serve application connections requests. A
common practise is to probe a pre-configured URL to check network
connectivity status as soon as a node access a network at bootstrap
or changes to different networks, e.g. Windows Vista, Windows 7,
Windows Server 2008 and iOS. If anything is abnormal, it assumes
there is a proxy on the path. This status detection is recommended
to be used in HE-MIF to detect DNS interception or HTTP proxy that
forces a login or a click-through. Unexamined PVDs or interfaces
should be accounted as "unconnected". It should not join the sort
process.
Afterwards, two phases normally are involved in a sort process, i.e.
name resolving and connection establishment. Parameters in a soft
set should considered at this stage.
When a node initiates name resolution requests, it should check if
there is a matched PVD ID for the destination name. A PVD agnostic
node may request DNS server selection DHCP option[RFC6731] for
interface selection guidance. Those information may weight a
particular interface to be preferred one prior to others sending
resolving requests. If the node can't find useful information in the
Soft Set, DNS queries would be sent out on multiple interfaces on
relevant PVDs in parallel to maximize chances for connectivity. Some
additional discussions of DNS selection consideration of HE-MIF are
described in Section 6.3.
Once a destination address was resolved, a connection is to be setup.
For the given destination address, a PVD-aware node selects a next-
hop and source address associated with that PVD in the name
resolution process. A PVD agnostic node may receive certain next hop
in RA message[RFC4191] , the node selects best source address
according to the[RFC6724] rules. When destination and source pairs
are identified, it should be treated with higher priority compared to
others and choose to initiate the connection in advance. This could
avoid thrashing the network, by not making simultaneous connection
attempts on multiple interfaces. After making a connection attempt
on the preferred pairs and failing to establish a connection within a
certain time period (see Section 6.2), a HE-MIF implementation will
decide to initiate connection attempt using rest of interfaces in
parallel. This fallback consideration may make subsequent connection
attempts successful on non-preferable interfaces.
Chen, et al. Expires December 25, 2014 [Page 7]
Internet-Draft happy-eyeballs-mif June 2014
The node would cache information regarding the outcome of each
connection attempt. Cache entries would be flushed periodically. A
system-defined timeout may take place to age the state. Maximum on
the order of 10 minutes defined in [RFC6555] is recommended to keep
the interface state changes synchronizing with IP family states.
If there are no specific Soft Set provided, all selected interfaces
should be equally treated. The connections would initiate on several
interface simultaneously. The goal here is to provide fast
connection for users, by quickly attempting to connect using one of
interfaces. Afterwards, the node would do the same caching and
flushing process as described above.
5. Implementation Framework
The simplest way for the implementation is within the application
itself. The mechanism described in the document would not require
any specific support from the operating system beyond the commonly
available APIs that provide transport service. It could also be
implemented as high-level API approach, linking to MIF-API
[I-D.ietf-mif-api-extension]. A number of enhancements could be
added, making the use of the high-level APIs much more productive in
building applications.
6. Additional Considerations
6.1. Usage Scope
Connection-oriented transports (e.g., TCP, SCTP) could be directly
applied as scoped in [RFC6555]. For connectionless transport
protocols (e.g., UDP), a similar mechanism can be used if the
application has request/ response semantics (e.g., as done by
Interactive Connectivity Establishment (ICE) to select a working IPv6
or IPv4 media path[RFC6157])."
6.2. Fallback Timeout
When the preferred interface was failed, HE-MIF would trigger
fallback process to start connection initiation on several candidate
interfaces. It should set a reasonable wait time to comfort user
experience. Aggressive timeouts may achieve quick interface
handover, but at the cost of traffic that may be chargeable on
certain networks, e.g. the handover from WiFi to 3/4G would bring a
bill to customers. Considering the reasons, it is recommended to
prioritize the input from users(e.g. real customers or applications)
through user interface. For default-setting on a system, a hard
error[RFC1122] in replied ICMP could serve as a trigger for the
fallback process. When the ICMP soft error is present or non-
Chen, et al. Expires December 25, 2014 [Page 8]
Internet-Draft happy-eyeballs-mif June 2014
response was received, it's recommended that the timeout should be
large enough to allow connection retransmission. [RFC1122] states
that such timer MUST be at least 3 minutes to provide TCP
retransmission. Several minutes delay may not inappropriate for user
experiences. A widespread practice[RFC5461] sets 75 seconds to
optimize connection process.
More optimal timer may be expected. The particular setting will be
very specific to implementations and cases. The memo didn't try to
provide a concrete value due to following concerns.
o RTT(Round-Trip Time) on different interfaces may vary quite a lot.
A particular value of timeout may not accurately help to make a
decision that this interface doesn't work at all. On the
contrary, it may cause a misjudgment on a interface, which is not
very fast. In order to compensate the issues, the timeout setting
based on past experiences of a particular interface may help to
make a fair decision. Whereas, it's going beyond the capability
of Happy Eyeballs [RFC6555]. Therefore, it leaves a particular
implementation.
o In some cases, fast interface may not be treated as "best". For
example, a interface could be evaluated in the principle of
bandwidth-delay, termed "Bandwidth-Delay-Product ". Happy
Eyeballs measures only connection speed. That is, how quickly a
TCP connection is established . It does not measure bandwidth. If
the fallback has to take various factors into account and make
balanced decision, it's better to resort to a specific context and
implementation.
6.3. DNS Selections
In the sort process, HE-MIF prioritizes PVD-ID match or
[RFC6731]inputs to select a proper server. It could help to address
following two cases.
o A DNS answer may be only valid on a specific provisioning domain,
but DNS resolver may not be aware of that because DNS reply is not
kept with the provisioning from which the answer comes. The
situation may become worse if asking internal name with public
address response or asking public name with private address
answers.
o Some FQDNs can be resolvable only by sending queries to the right
server (e.g., intranet services). Otherwise, a response with
NXDOMAIN is replied. Fast response is treated as optimal only if
the record is valid. That may cause messy for data connections,
since NXDOMAIN doesn't provide useful information.
Chen, et al. Expires December 25, 2014 [Page 9]
Internet-Draft happy-eyeballs-mif June 2014
By doing HE-MIF, it can help to solve the issues of DNS interception
with captive portal. The DNS server modified and replied the answer
with the IP address of captive portal rather than the intended
destination address. In those cases, TCP connection may succeed, but
Internet connectivity is not available. It results in lack of
service unless user has authenticated. HE-MIF recommended using
network connectivity status probes to examine a pre-configured URL
for detecting DNS interception on the path (see more in Section 4.2).
The node will be able to automatically rely upon other interfaces to
select right DNS servers by excluding the unexamined interfaces.
6.4. Flow Continuity
Interface changing should only happen at the beginning of new session
in order to keep flow continuity for ongoing TCP session. Dynamic
movement of traffic flows are beyond the scope of this document.
7. IANA Considerations
This memo includes no request to IANA.
8. Security Considerations
The security consideration is following the statement in [RFC6555]and
[RFC6418].
9. Acknowledgements
The authors would like to thank Margaret Wasserman, Hui Deng, Erik
Kline, Stuart Cheshire, Teemu Savolainen, Jonne Soininen, Simon
Perreault, Zhen Cao, Dmitry Anipko and Ted Lemon for their helpful
comments.
10. References
10.1. Normative References
[RFC1122] Braden, R., "Requirements for Internet Hosts -
Communication Layers", STD 3, RFC 1122, October 1989.
[RFC4191] Draves, R. and D. Thaler, "Default Router Preferences and
More-Specific Routes", RFC 4191, November 2005.
[RFC6555] Wing, D. and A. Yourtchenko, "Happy Eyeballs: Success with
Dual-Stack Hosts", RFC 6555, April 2012.
Chen, et al. Expires December 25, 2014 [Page 10]
Internet-Draft happy-eyeballs-mif June 2014
[RFC6724] Thaler, D., Draves, R., Matsumoto, A., and T. Chown,
"Default Address Selection for Internet Protocol Version 6
(IPv6)", RFC 6724, September 2012.
[RFC6731] Savolainen, T., Kato, J., and T. Lemon, "Improved
Recursive DNS Server Selection for Multi-Interfaced
Nodes", RFC 6731, December 2012.
10.2. Informative References
[I-D.ietf-mif-api-extension]
Liu, D., Lemon, T., Ismailov, Y., and Z. Cao, "MIF API
consideration", draft-ietf-mif-api-extension-05 (work in
progress), February 2014.
[I-D.ietf-mif-mpvd-arch]
Anipko, D., "Multiple Provisioning Domain Architecture",
draft-ietf-mif-mpvd-arch-01 (work in progress), May 2014.
[RFC5461] Gont, F., "TCP's Reaction to Soft Errors", RFC 5461,
February 2009.
[RFC6157] Camarillo, G., El Malki, K., and V. Gurbani, "IPv6
Transition in the Session Initiation Protocol (SIP)", RFC
6157, April 2011.
[RFC6418] Blanchet, M. and P. Seite, "Multiple Interfaces and
Provisioning Domains Problem Statement", RFC 6418,
November 2011.
Authors' Addresses
Gang Chen
China Mobile
53A,Xibianmennei Ave.,
Xuanwu District,
Beijing 100053
China
Email: phdgang@gmail.com
Chen, et al. Expires December 25, 2014 [Page 11]
Internet-Draft happy-eyeballs-mif June 2014
Carl Williams
Consultant
El Camino Real
Palo Alto, CA 94306
USA
Email: carlw@mcsr-labs.org
Dan Wing
Cisco Systems, Inc.
170 West Tasman Drive
San Jose, CA 95134
USA
Email: dwing@cisco.com
Andrew Yourtchenko
Cisco Systems, Inc.
De Kleetlaan, 7
Diegem B-1831
Belgium
Email: ayourtch@cisco.com
Chen, et al. Expires December 25, 2014 [Page 12]
