Internet Engineering Task Force J. Palet
Internet-Draft M. Diaz
Expires: July 28, 2005 Consulintel
P. Savola
CSC/FUNET
January 24, 2005
Analysis of IPv6 Tunnel End-point Discovery Mechanisms
draft-palet-v6ops-tun-auto-disc-03.txt
Status of this Memo
This document is an Internet-Draft and is subject to all provisions
of section 3 of RFC 3667. By submitting this Internet-Draft, each
author represents that any applicable patent or other IPR claims of
which he or she is aware have been or will be disclosed, and any of
which he or she become aware will be disclosed, in accordance with
RFC 3668.
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 July 28, 2005.
Copyright Notice
Copyright (C) The Internet Society (2005).
Abstract
To be able to automate setting up IPv6-in-IPv4 tunnels, it is
important to be able to automatically determine the tunnel end-point
for the tunneling mechanism. This memo presents a short analysis and
conclusions on the different approaches for discovering the IPv6
Palet, et al. Expires July 28, 2005 [Page 1]
Internet-Draft Analysis of TEP Discovery Mechanisms January 2005
tunnel endpoint on a node.
Table of Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1 Manual Configuration . . . . . . . . . . . . . . . . . . . 3
2. Applicability of Tunnel Endpoint Discovery . . . . . . . . . . 4
2.1 Scope of Tunnel Endpoint Discovery . . . . . . . . . . . . 4
2.2 Assumptions about Network Topologies . . . . . . . . . . . 5
3. Analysis of Solutions . . . . . . . . . . . . . . . . . . . . 5
3.1 Anycast-based Solutions . . . . . . . . . . . . . . . . . 5
3.2 DNS-based Solutions . . . . . . . . . . . . . . . . . . . 7
3.2.1 Storing the TEP Information . . . . . . . . . . . . . 8
3.2.2 Prefixing the DNS Search Path . . . . . . . . . . . . 8
3.2.3 IP-address Query from Reverse DNS . . . . . . . . . . 9
3.3 DHCP-based Solutions . . . . . . . . . . . . . . . . . . . 10
3.4 PPP-based Solutions . . . . . . . . . . . . . . . . . . . 11
3.5 SLP-based Solutions . . . . . . . . . . . . . . . . . . . 11
3.6 Combined Solutions . . . . . . . . . . . . . . . . . . . . 11
4. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . 12
5. Security Considerations . . . . . . . . . . . . . . . . . . . 13
6. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 13
7. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . 13
8. References . . . . . . . . . . . . . . . . . . . . . . . . . . 13
8.1 Normative References . . . . . . . . . . . . . . . . . . . 13
8.2 Informative References . . . . . . . . . . . . . . . . . . 14
Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . 15
A. More Discussion of Anycast Discovery . . . . . . . . . . . . . 15
B. Centralized Broker-based Solutions . . . . . . . . . . . . . . 16
Intellectual Property and Copyright Statements . . . . . . . . 17
Palet, et al. Expires July 28, 2005 [Page 2]
Internet-Draft Analysis of TEP Discovery Mechanisms January 2005
1. Introduction
It is important to make setting up IPv6 connectivity simpler, so that
IPv6-ignorant novice users can get the benefit of IPv6 transparently,
without the user even having to know that IPv6 connectivity has been
obtained.
While this has become possible with Teredo and 6to4, they do not
provide well for managed infrastructure, where the addresses come
from the service provider's prefix.
This document presents a short analysis and conclusions for different
options to automatically determine the IPv6-in-IPv4 (with or without
UDP encapsulation) tunnel end-points to a tunnel server, so that the
set up of tunnels could be automated.
Note that the other end-point ("tunnel server") typically also needs
to have a means to configure the client's end-point, but that is
assumed to be solved by the tunnel server mechanism [1][7], and
beyond the scope of this memo.
Some form of automatic discovery already exists in some already
specified mechanisms; the generic discovery is out of scope, but we
contrast the approaches to already-deployed methods when appropriate.
For example, 6to4 [2] uses global anycast [3] and/or vendor's branch
of DNS, Teredo [8] uses vendor's branch of DNS, and ISATAP [9] uses
search-path -prefixed DNS.
First, we look at manual configuration, and why it is not considered
sufficient.
1.1 Manual Configuration
Users typically expect to be able to manually configure the tunnel
endpoint information, and the implementations obviously should allow
that.
Some implementations may also provide some default values (e.g.,
using a vendor branch of DNS, as described in Section 3.2). This is
a non-interoperability issue, and may also be a good idea.
Often the ISPs also provide CD-ROMs or other material to the
(non-knowledgeable) users which will automatically reset the network
connectivity settings to the values used by the ISP. Such
configuration could also include the tunnel endpoint if the ISP would
like to roll it out to every customer.
This raises the question whether it is strictly required to have an
Palet, et al. Expires July 28, 2005 [Page 3]
Internet-Draft Analysis of TEP Discovery Mechanisms January 2005
automatic discovery process, rather than wait and rely on the ISPs to
do a massive roll-out.
We do not consider manual configuration sufficient for two main
reasons (XXX: feel free to send some more..):
1. Many deployments are expected to happen at pre-production stage,
and those service providers would not yet be ready to integrate
the configuration in their CD-ROM etc. material.
2. The CD-ROM materials have mainly been targeted to the users who
set up their (IPv4) connectivity which either requires software
or out-of-band configuration. There is no reason to require
out-of-band configuration if the discovery could be done in a
feasible manner.
2. Applicability of Tunnel Endpoint Discovery
2.1 Scope of Tunnel Endpoint Discovery
There are three main areas of applicability for tunnel endpoint
discovery:
1. No discovery: always assume the client pre-configures the
endpoint information.
2. Discovery of the end-point at the "care-of" ISP only; that is,
discovery is only supported inside the ISP of the network the
user plugs into.
3. Discovery of the endpoint everywhere; contrast to the use of 6to4
anycast address. This is very problematic administratively,
financially and technically, because the IPv6 prefix is not
provider-independent as with 6to4.
We decree 3) out of scope for this study; this is too extensive a
problem to be solved here. Therefore, we concentrate on 2).
In addition to automatic discovery, the implementations should
naturally provide a manual configuration option. The manual
configuration could be used to override the automatic discovery
process or to configure a tunnel server at "home ISP" which the
discovery would not find if the user is visiting a network where no
tunnel server exists (compare to configuring MIPv6 Home Agent).
When there are multiple inputs to which tunnel server to use,
implementations will have to make a policy decision which one to
Palet, et al. Expires July 28, 2005 [Page 4]
Internet-Draft Analysis of TEP Discovery Mechanisms January 2005
pick; this is rocket science, and some implementations (e.g.,
Microsoft 6to4/ISATAP) already something like this. The main options
are:
a. First try local discovery, if it fails, try manual config,
b. Manual config first, then try local discovery, or
c. Only local discovery or only manual config.
2.2 Assumptions about Network Topologies
The following assumptions about network topologies apply to the
discovery process:
1. The CPE device can run either in bridged or routed mode.
2. In routed mode, the router typically is doing NAT with private IP
addresses. The router is also a DHCPv4 server, so DHCPv4 tunnel
endpoint option is not relayed to the user.
3. The router may also inject its own DNS search string (e.g.,
"home", "lan") instead relaying the one received from the ISP,
though how often that happens is unknown.
4. The user may also have deployed NAT boxes of his/her own.
3. Analysis of Solutions
Several possible solutions to discovering the tunnel end-point can be
imagined; this section describes them in detail.
3.1 Anycast-based Solutions
An "anycast" (shared-unicast by some terminology: see [10]) address
identifies a group of hosts, usually server hosts. When a client
sends a datagram to a shared-unicast address, it is delivered to one
of the shared-unicast servers based on the routing topology and
metrics.
There are two possible ways of using "anycast": as a global service
(where a shared-unicast prefix is the same for everyone, and
advertised in the Inter-domain routing) or as a local service, where
the service provider is sharing one of its own addresses on multiple
nodes for example for load-balancing or redundancy reasons. As local
anycast is invisible to the users, it is not further discussed here.
Palet, et al. Expires July 28, 2005 [Page 5]
Internet-Draft Analysis of TEP Discovery Mechanisms January 2005
A packet to a shared-unicast address may end up being delivered to
more than one node. In addition, there is no guarantee that two
consecutive datagrams sent from the same host towards the same
shared-unicast address are going to be delivered to the same node.
However, when the routing topology is stable and metrics are
well-designed, the packets are regularly delivered to the same nodes.
Operational issues relating to managemeng of anycast services have
been described in [4].
A global anycast address could be leveraged in two fundamentally
different ways:
1. Use the anycast address only for the initial handshake, to
establish a stable unicast address of the end-point (and possibly
to perform some initial negotiation, e.g., nonces). All the
subsequent packets are sent to the unicast address which is
included in the payload of the reply. An example of such use is
in [5].
2. Use the anycast address for all the communications (e.g., as with
6to4).
The former approach is much more suitable in this situation as the
IPv6 address/prefix of the tunnel service depends on the operator of
the service. The cost is at least one additional roundtrip.
The failure modes of the initial handshake anycast are described in
Appendix A.
The advantages of the anycast approach are:
o Works well also in the presence of NATs and does not require any
other components like DNS or DHCP.
o The routing stability and leaks are not a major concern if the
anycast address is only used for initial discovery. In other
words, the worst that could happen is that if the initial
discovery does not work correctly at the moment, and the user is
either cannot get any service or directed to a tunnel server which
does not offer any service.
The drawbacks are:
o Setting up an internal anycast route advertisement (e.g., in an
enterprise) is likely a little bit more difficult than adding a
name in the DNS or configuring DHCP.
o The use of non-local prefixes may require changes in firewall IP
Palet, et al. Expires July 28, 2005 [Page 6]
Internet-Draft Analysis of TEP Discovery Mechanisms January 2005
prefix, access lists, etc.
o The failure modes are a bit more complex than, e.g., just looking
up a domain name, as one will have to send 1-2 packets to the
anycast address to see if a working tunnel server is found or not.
In summary, from the automatic methods, a global anycast based
solution, using the address only for initial handshake, seems like a
very promising approach, but has routing operations issues which need
to be considered.
3.2 DNS-based Solutions
As DNS is globally deployed and easy to use, it could provide a means
for discovering the end-point address, either based on the forward or
reverse tree.
There are roughly four kinds of different approaches:
1. "(forward) global name": the systems look up a globally unique
name, like www.tunnel-server.net which would point to the global
anycast address. This is not considered further as this does not
solve any problem in itself, because the clients could have been
configured with IP address instead.
2. "(forward) vendor branch": the operating system vendors may
provide a DNS record which is looked up (contrast to
"6to4.windows.microsoft.com."), giving the vendor some control
over already deployed systems. This could in practice only be
used to configure the global anycast address, because the authors
don't expect the vendors would provide a tunnel server to all the
customers all over the world. Therefore this approach is not
considered further.
3. "(forward) prefixing the search path" [6]: one could look up a
service-specific special string, like "_tunnel-server", appended
by the DNS search path, e.g., "isp.example.com", resulting in a
query of "_tunnel-server.isp.example.com".
4. "(reverse) querying the IP addresses for TEP information": for
example looking up a special record for the assigned IP address.
It is also a question where to store the information; the main
suggestions have been at least A/CNAME, SRV, NAPTR and new type of
records. We first discuss these options.
Approaches 3) and 4) are a bit more complicated and have different
tradeoffs, so they are elaborated after looking at how to store the
Palet, et al. Expires July 28, 2005 [Page 7]
Internet-Draft Analysis of TEP Discovery Mechanisms January 2005
information.
3.2.1 Storing the TEP Information
Forward-tree global name and service branch would obviously use
A/CNAME records.
Forward-tree search path prefixing could use A/CNAME records.
Architecturally a bit "cleaner" approach would be to use SRV records,
which also provide a bit more fine-grained means for load
distribution. NAPTR provide even more extended load distribution,
but it is not clear what the benefit would be. A new RR could also
be defined, but there seems to be no particular reason to do so, and
a lot of drawbacks in the process.
Reverse-tree IP address lookup would likely have to define a new
record type.
The main benefit of using A/CNAME records would be the applications
can use simple getaddrinfo() lookups, instead of having to write
their own or use a non-standardized DNS record lookup functions
(e.g., getrrsetbyname).
3.2.2 Prefixing the DNS Search Path
Prefixing the search path bears a bit more analysis; we discussed how
to store the information above, and now look at where to store the
records (i.e., the prefix to use, and what to do with the conflicts).
This approach makes two assumptions; there are cases of both when
these do not hold:
1. There is a decent mapping between the DNS hierarchy and the
routing topology.
2. The information propagated to the end nodes in the "DNS search
path" is relevant to figure out the domain name of the whoever is
providing the tunnel service.
The main advantages of the solution are:
o The discovery process is simple, and is already used for example
by ISATAP (but without NATs in the middle).
o Adding the service is very simple, as the ISP is only required to
add one address in their DNS zone.
However, the main drawbacks of this approach are:
Palet, et al. Expires July 28, 2005 [Page 8]
Internet-Draft Analysis of TEP Discovery Mechanisms January 2005
o Some routers and middleboxes may not propagate the search path,
but to insert their own, and this approach does not work in that
situation. It is not clear how widespread this is. (In case the
NAT would insert a new search path, e.g., "lan", they also should
have be authorative for the zone, otherwise the root servers get
bombarded by lookups.)
o In some cases in global enterprises, forward DNS may not map as
well to the physical topology as IP addresses through reverse DNS
would. (NB: then ISATAP would have the same issue as well.)
In summary, it is questionable how well prefixing the search path
works under these circumstances, and in particular how common
propagating (or not) the search path is.
3.2.3 IP-address Query from Reverse DNS
The node might also try to find out its tunnel endpoint information
by querying its own IP address in the reverse DNS for a certain DNS
record type.
The name needs to be stored somewhere. The options are basically
queries like (for IP address 192.0.2.1):
1. "QNAME=_tunnel.1.2.0.192.in-addr.arpa. QTYPE=A"
2. "QNAME=1.2.0.192.in-addr.arpa. QTYPE=TEP"
In the first case, an arbitrary subname would have to be defined; one
could query these for A, PTR, or some other records directly.
In the second case, where the PTR records for the name might already
exist one should probably use a new record type, though something
like NAPTR has been used in private experiments.
Advantages:
o IP address maps very well to the topology.
o Querying a new record type is an architecturally relatively clean
approach.
Drawbacks:
o Does not work well with NAT; would require that the ISPs
prepopulate all the private address space records.
o Major management problems. Wildcards cannot be used because they
Palet, et al. Expires July 28, 2005 [Page 9]
Internet-Draft Analysis of TEP Discovery Mechanisms January 2005
would be in the middle of the QNAME or would match other records
(such as PTR) because wildcards are QTYPE-agnostic.
o Reverse DNS is not necessarily managed by those who would like to
configure this service.
o A new record type number would be available (for test deployments,
interoperability etc.) only after an RFC has been published.
In some cases, the reverse records are generated by scripts which
could be modified to also add these records. However, the presence
of such scripts and the ability to modify them cannot be assumed.
In short, storing information in the reverse DNS does not look like a
good approach either.
3.3 DHCP-based Solutions
In most situations, the users receive the IPv4 information from an
IPv4 DHCP server. Consequently, one of the parameters to be provided
by the DHCP server could be the tunnel end-point address, e.g., as
described in [11].
This approach has several drawbacks:
o It requires standardizing new parameters/options on this protocol
and also upgrading the DHCP client/server implementations to
support this feature.
o It will not work if DHCP client is not used, e.g., in many
dial-up scenarios, where only PPP is used; DHCP is not used in
some (advanced) xDSL setups which use static routing. Also, some
managed networks do not use DHCP. Still, in many cases, DHCP is
used between a customer and the ISP.
o If a router is providing local DHCP information (e.g., an ADSL
router), the tunnel end-point information would have to be
automatically "proxied" to the "local DHCP", or manually
configured on the router to propagate to the hosts in the case
that the router is not activating the tunnel itself.
o It requires manual configuration/update of the ISP's DHCP servers
when there are changes to the tunnel end-points, similar to
updating DNS, NTP, etc., server information.
In short, DHCP-based solutions seem unacceptable because a NAT/router
does not automatically pass the information to the nodes in an
"opaque" manner.
Palet, et al. Expires July 28, 2005 [Page 10]
Internet-Draft Analysis of TEP Discovery Mechanisms January 2005
3.4 PPP-based Solutions
In the case of PPP-like connections, specific PPP parameters could be
passed to the clients, as part of the AAA signaling process.
This solution has the same drawbacks as DHCP-based solution.
Further, there has been resistance to making extensions to PPP (e.g.,
passing IPv6 prefix options), so it is an open question whether this
information could be passed as a standardized PPP option at all.
3.5 SLP-based Solutions
The Service Location Protocol [12] provides a framework for the
discovery and selection of network services. Tunnel-End-Point for
IPv6 tunnels could be defined as a network service which will have
also assigned a specific service name.
SLP has a number of drawbacks:
o SLP is not really widely implemented or deployed.
o It requires multicast infrastructure or the additional deployment
of Directory Agents (DA) for Service Agent (SA) discovery.
o If DA is deployed and the network has not multicast support, some
way for discovery the DA is required. DHCP could be used [12] but
this has the same issues why a DHCP-based approach is not
sufficient (in Section 3.3).
o It requires the implementation of a User Agent (UA) on the
client's host. This is neither always possible nor feasible.
o It can not offer any kind of load-balancing if more than one TEP
is deployed.
In short, SLP seems to come awfully short in matching the
requirements for the solution.
3.6 Combined Solutions
Many solutions can be combined with each other, but because the
clients and servers must have a minimum mandatory-to-implement
mechanism, it is better if only one externally visible mechanism can
be used.
Manual configuration override option is of course a good addition to
any discovery mechanism.
Palet, et al. Expires July 28, 2005 [Page 11]
Internet-Draft Analysis of TEP Discovery Mechanisms January 2005
Anycasting a locally determined TEP address (e.g., through DNS or
DHCP) is a useful technique for load balancing purposes. This is
invisible to the clients.
As has been experienced with 6to4, it might be possible to use a
vendor's DNS branch to configure a global anycast address, but this
the former requires no interoperability, so it's an implementation's
internal matter.
4. Conclusions
Manual configuration should always be provided, but the question is
whether the configuration distributed by ISPs (e.g., in CD-ROMs etc)
is sufficient.
Global anycast for initial discovery looks like a promising solution,
but has routing operations issues which need to be considered.
Both DNS forward and reverse based solutions suffer from various
problems when a NAT/router is present.
Centralized brokers are a non-starter.
DHCP and PPP do not seem to be usable due to restrictions on
environment where they work. SLP is not sufficiently deployed,
implemented or otherwise feasible.
The following table summarizes the pros/cons of different approaches
presented in this memo.
+--------------------+----------------------+-----------------------+
| | Pros | Cons |
+--------------------+----------------------+-----------------------+
| Anycast | *** | * |
| Centralized Broker | - | *** |
| Forward DNS w/ | ** | ** |
| Prefixing | | |
| Reverse DNS | ** | *** |
| DHCP | * | *** |
| PPP | - | *** |
| SLP | - | *** |
+--------------------+----------------------+-----------------------+
Qualification of pros/cons:
- : None
* : Few
Palet, et al. Expires July 28, 2005 [Page 12]
Internet-Draft Analysis of TEP Discovery Mechanisms January 2005
** : Medium
*** : High
5. Security Considerations
If the tunnel end-point discovery is done in an insecure fashion, so
that an attacker could influence the discovery process, the attacker
could be able to hijack all the IPv6 communications. This must be
kept in mind when analyzing the different discovery solutions, and
spelled-out explicitly in the requirements, if the threats are to be
mitigated in tunneling mechanisms somehow (e.g., using a return
routability procedures).
In particular, the potential weaknesses of DNS bear some
consideration.
6. IANA Considerations
This document requests no action for IANA.
[[note to RFC-editor: this section can be removed upon publication.]]
7. Acknowledgements
This memo was written as a consequence of experience using IPv6 when
traveling, number of talks during IETF meetings and specially the
work with the unmanaged, ISP and enterprise v6ops design teams.
The authors would also like to acknowledge inputs from Carl Williams,
Brian Carpenter, Jeroen Massar, Alain Durand, Tim Chown, and Florent
Parent. This work has been partially funded by the European
Commission under the Euro6IX project.
8. References
8.1 Normative References
[1] Parent, F., "Goals for Registered Assisted Tunneling",
Internet-Draft draft-ietf-v6ops-assisted-tunneling-requirements-01
, October 2004.
[2] Carpenter, B. and K. Moore, "Connection of IPv6 Domains via IPv4
Clouds", RFC 3056, February 2001.
[3] Huitema, C., "An Anycast Prefix for 6to4 Relay Routers",
RFC 3068, June 2001.
Palet, et al. Expires July 28, 2005 [Page 13]
Internet-Draft Analysis of TEP Discovery Mechanisms January 2005
[4] Lindqvist, K. and J. Abley, "Operation of Anycast Services",
Internet-Draft draft-kurtis-anycast-bcp-00, October 2004.
[5] Thaler, D. and L. Vicisano, "IPv4 Automatic Multicast Without
Explicit Tunnels (AMT)",
Internet-Draft draft-ietf-mboned-auto-multicast-03, October
2004.
[6] Faltstrom, P. and R. Austein, "Design Choices When Expanding
DNS", Internet-Draft draft-iab-dns-choices-00, October 2004.
8.2 Informative References
[7] Durand, A., Fasano, P., Guardini, I. and D. Lento, "IPv6 Tunnel
Broker", RFC 3053, January 2001.
[8] Huitema, C., "Teredo: Tunneling IPv6 over UDP through NATs",
Internet-Draft draft-huitema-v6ops-teredo-04, January 2005.
[9] Templin, F., Gleeson, T., Talwar, M. and D. Thaler, "Intra-Site
Automatic Tunnel Addressing Protocol (ISATAP)",
Internet-Draft draft-ietf-ngtrans-isatap-23, December 2004.
[10] Hagino, J. and K. Ettican, "An analysis of IPv6 anycast",
Internet-Draft draft-ietf-ipngwg-ipv6-anycast-analysis-02, June
2003.
[11] Kim, P. and S. Park, "DHCP Option for Configuring IPv6-in-IPv4
Tunnels", Internet-Draft draft-daniel-dhc-ipv6in4-opt-05,
October 2004.
[12] Guttman, E., Perkins, C., Veizades, J. and M. Day, "Service
Location Protocol, Version 2", RFC 2608, June 1999.
[13] Brisco, T., "DNS Support for Load Balancing", RFC 1794, April
1995.
[14] Johnson, D., Perkins, C. and J. Arkko, "Mobility Support in
IPv6", RFC 3775, June 2004.
[15] Durand, A. and S. Yamamoto, "Service Discovery using NAPTR
records in DNS",
Internet-Draft draft-yamamoto-naptr-service-discovery-00,
October 2004.
Palet, et al. Expires July 28, 2005 [Page 14]
Internet-Draft Analysis of TEP Discovery Mechanisms January 2005
Authors' Addresses
Jordi Palet Martinez
Consulintel
San Jose Artesano, 1
Alcobendas - Madrid
E-28108 - Spain
Phone: +34 91 151 81 99
Fax: +34 91 151 81 98
Email: jordi.palet@consulintel.es
Miguel Angel Diaz Fernandez
Consulintel
San Jose Artesano, 1
Alcobendas - Madrid
E-28108 - Spain
Phone: +34 91 151 81 99
Fax: +34 91 151 81 98
Email: miguelangel.diaz@consulintel.es
Pekka Savola
CSC/FUNET
Espoo
Finland
Email: psavola@funet.fi
Appendix A. More Discussion of Anycast Discovery
We do a little bit of more extensive analysis of anycast-based
solution here to get a better understanding of its operational
properties.
At this point, we only discuss the different failure modes of initial
handshake anycast, and see that these can be solved with robust
specification and implementation.
a. A new server starts advertising the address but is refusing to
serve the users: established tunnels continue to work, new
tunnels cannot be established; tracerouting to the server
identifies where the culprit is.
b. The current anycast server goes down: tunnels established to its
unicast address go down after the event is detected. New tunnels
Palet, et al. Expires July 28, 2005 [Page 15]
Internet-Draft Analysis of TEP Discovery Mechanisms January 2005
use the next anycast server (if available) or no server at all
(in which case the tunnels may get re-created when the server
comes back up).
c. There is no server in the local ISP, and the anycast prefix has
been filtered out or does not exist globally: the initial
discovery follows the default route, and either gets discarded
(and the discovery process notices this after a timeout) or a
router returns a network/host unreachable ICMP error message.
d. There is no server locally, but someone else is advertising it
here: if the service works, that's OK (though if the service
works but is bad quality or 100's of milliseconds away is
undesirable; robust implementations may check the RTT). If the
server fails to serve, that's also OK -- the discovery process
has to be robust against this.
Appendix B. Centralized Broker-based Solutions
This solution is described in the appendix because it does not
actually solve the discovery problem, but has been proposed on the
assumption that it would.
Inside a single administrative domain, it would also be possible to
deploy a centralized server or a "broker" knowing the status of all
the associated end-points. Furthermore, it could redirect the users
to the correct end-points. This mechanism would still need another
complementary approach to actually discover the centralized broker.
This approach is highly assumptive of the tunneling set-up mechanism,
and likely requires the implementation of lengthy redirection or
negotiation features which do not work well through a NAT.
Applying a centralized model over multiple administrative domains,
e.g., having a single server for the whole Internet, would be
administratively and management-wise unfeasible.
A global centralized broker is completely unfeasible. The only
benefit of a local broker is the ability to perform more fine-grained
load balancing or policing, and does not solve the actual problem,
because the same methods need to be applied to discover the
centralized broker.
Palet, et al. Expires July 28, 2005 [Page 16]
Internet-Draft Analysis of TEP Discovery Mechanisms January 2005
Intellectual Property Statement
The IETF takes no position regarding the validity or scope of any
Intellectual Property Rights or other rights that might be claimed to
pertain to the implementation or use of the technology described in
this document or the extent to which any license under such rights
might or might not be available; nor does it represent that it has
made any independent effort to identify any such rights. Information
on the procedures with respect to rights in RFC documents can be
found in BCP 78 and BCP 79.
Copies of IPR disclosures made to the IETF Secretariat and any
assurances of licenses to be made available, or the result of an
attempt made to obtain a general license or permission for the use of
such proprietary rights by implementers or users of this
specification can be obtained from the IETF on-line IPR repository at
http://www.ietf.org/ipr.
The IETF invites any interested party to bring to its attention any
copyrights, patents or patent applications, or other proprietary
rights that may cover technology that may be required to implement
this standard. Please address the information to the IETF at
ietf-ipr@ietf.org.
Disclaimer of Validity
This document and the information contained herein are provided on an
"AS IS" basis and THE CONTRIBUTOR, THE ORGANIZATION HE/SHE REPRESENTS
OR IS SPONSORED BY (IF ANY), THE INTERNET SOCIETY AND THE INTERNET
ENGINEERING TASK FORCE DISCLAIM ALL WARRANTIES, EXPRESS OR IMPLIED,
INCLUDING BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE
INFORMATION HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED
WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
Copyright Statement
Copyright (C) The Internet Society (2005). This document is subject
to the rights, licenses and restrictions contained in BCP 78, and
except as set forth therein, the authors retain all their rights.
Acknowledgment
Funding for the RFC Editor function is currently provided by the
Internet Society.
Palet, et al. Expires July 28, 2005 [Page 17]