Internet DRAFT - draft-pppext-mmp-discovery
draft-pppext-mmp-discovery
Draft-pppext-mmp-discovery-00.txt G. Malkin/Bay Networks
November 1997
Multi-link Multi-node PPP
Abstract
This document specifies a standard way for Multi-link PPP to operate
across multiple nodes. Both the mechanism by which the Bundle Head
is discovered and the PPP fragment encapsulation are specified.
Status of this Memo
This document is an Internet Draft. 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. Internet Drafts may be updated, replaced, or obsoleted by
other documents at any time. It is not appropriate to use Internet
Drafts as reference material or to cite them other than as a "working
draft" or "work in progress."
Please check the I-D abstract listing contained in each Internet
Draft directory to learn the current status of this or any other
Internet Draft.
It is intended that this document will be submitted to the IESG for
consideration as a standards document. Distribution of this document
is unlimited.
Acknowledgements
I would like to thank Joe Frazier for filling in some of the details
and reviewing this document.
1. Introduction
Multi-link PPP [MP] allows a dial-in user to open multiple PPP
connections to a given host. In general, this is done on an on-
demand basis. That is, a secondary link, or multiple secondary
links, are established when the data load on the primary link, and
Malkin Expires: 12May98 [Page 1]
Internet Draft MMP November 1997
any previously established secondary links, nears capacity. As the
load decreases, the secondary link(s) may be disconnected.
Many dial-in hosts which support multi-link PPP dial the same phone
number for all links. This implies that there exists a rotary at the
Point Of Presence (POP) which routes incoming calls to a bank of
modems. These may be physically independent modems connected to
Remote Access Server (RAS) and a rotary of analog phone lines, or a
RAS with internal modems connected to analog lines or a T1/E1 or
T3/E3 channel. In any case, a given RAS can only handle just so many
simultaneous connections. A typical POP may need to support hundreds
of connections, but no RAS today can handle that many. This creates
a problem when a user's primary PPP connection is established to one
RAS in a POP and a secondary connection is established to another.
This may occur because the first RAS has no available modems, or
because incoming calls are assigned to ports in a round-robin
fashion, for example, and the second call is simply assigned to
another RAS.
The solution to this problem is to provide a mechanism by which a RAS
can determine if a Multi-link PPP connection is a primary or
secondary and, if a secondary, where the Bundle Head (the process
within a RAS which reassembles the PPP fragments transmitted over the
primary and secondary links) resides. If the Bundle Head resides on
a different RAS, a protocol must be used to transfer the PPP
fragments to the RAS containing the Bundle Head so that the PPP frame
can be reassembled.
Section 2 of this document specifies the Discovery Mechanism.
Section 3 specifies the Transfer Protocol. Section 4 specifies the
configuration parameters needed for the Discovery Protocol.
2. Bundle Head Discovery Mechanism
When a user dials into a RAS and negotiates Multi-link PPP (MP)
during the Link Control Protocol (LCP) phase, the RAS must determine
which one of the following three cases exists:
1- This is the primary (first) link of the MP connection. In this
case, the RAS should create the Bundle Head.
2- This is a secondary link of the MP connection and the Bundle Head
resides on this RAS. In this case, the RAS should add the link to
the Bundle (standard MP).
3- This is a secondary link of the MP connection and the Bundle Head
resides on a different RAS. In this case, the RAS should
Malkin Expires: 12May98 [Page 2]
Internet Draft MMP November 1997
establish a path (see section 3) to the RAS that has the Bundle
Head, and use that path to transfer MP fragments.
In operation, a RAS will make the determination for case 2 first
(because it is the easiest and requires no communication with other
RASes. If the Bundle Head is not local, the Discovery Protocol is
used to determine where the Bundle Head is, if it exists at all.
2.1 Packet Format
See "IANA Considerations" (section 6) for UDP port number assignment.
A Discovery Message has the following format:
+------+------+------------+----------+----======----+
| type |length| IP address | checksum | endpoint ID |
+------+------+------------+----------+----======----+
where:
type - 1 octet
Message type: 1-query, 2-response.
length - 1 octet
The length (in octets) of the endpoint ID.
IP address - 4 octets
The IP address of the RAS transmitting the message.
checksum - 2 octets
The ones-complement checksum of the endpoint ID. This is the
standard IP checksum except that the checksum is not self-
inclusive (i.e., the checksum is done only over the endpoint ID).
A value of zero indicates that no checksum has been generated.
See "Endpoint Identifier Matching" (section 2.2) for the use of
this field.
endpoint ID - variable length
The endpoint identifier of the connection. From the discovery
protocol's point of view, this is an opaque value. However, to
ensure multi-vendor interoperability, the format of this field
must be defined. The descriptions of, and legal values for, the
fields in the endpoint ID are defined in [MP].
Malkin Expires: 12May98 [Page 3]
Internet Draft MMP November 1997
+------+------+--==--+------+------+--==--+------+--==--+
|remote|remote|remote|local |local |local |user | user |
|EPD |EPD |EPD |EPD |EPD |EPD |name | name |
|class |length|data |class |length|data |length| data |
+------+------+--==--+------+------+--==--+------+--==--+
Notes:
EPD = EndPoint Descriminator.
remote = dial-in host.
local = RAS.
class and length fields are 1-octet in length.
data fields are of variable (including zero) length.
2.2 Endpoint Identifier Matching
Comparing endpoint IDs can be time consuming. First, the classes of
the EPDs must be determined, then the values compared. These
comparisons might be fast arithmetic compares or slow octet-wise
compares of 20-octet long values. To improve performance, because
the protocol is time-driven, the checksum field may be used for a
fast comparison.
When a Bundle Head is created, the checksum is created and stored
along with the Endpoint ID. When a Query or Response Message is
generated, the checksum is created and stored in the message. When a
RAS receives a message, it can do a quick comparison of the checksum
in the message to the checksums in its tables. If a checksum does
not match, the Endpoint ID cannot match. However, if a checksum does
match, the Endpoint IDs must be properly compared to verify the
match.
Obviously, there is a cost associated with creating the checksums,
but they are created only once per message and once for each Bundle
Head creation. However, the comparisons occur multiple times in
multiple RASes for each new secondary connection. Therefore, there
is a net savings in processing.
2.3 Protocol Operation
Throughout this section, configurable variables are specified by
their names (e.g., ROBUSTNESS refers to the number of transmits).
The Discovery Protocol begins by sending to MULTICAST, ROBUSTNESS
Query Messages at QUERY_INTERVAL intervals. If no Response Message
for that Request is received within QUERY_INTERVAL of the last
broadcast (a total time of ROBUSTNESS * QUERY_INTERVAL), the RAS
assumes that this is the primary link and begins to build the Bundle
Head. It then sends to MULTICAST a Response Message (in case another
Malkin Expires: 12May98 [Page 4]
Internet Draft MMP November 1997
link comes up after the time-out but before the Bundle Head is
built). If a Response Message is received (i.e., a Bundle Head
exists on another RAS), no additional Query Messages are sent and the
RAS establishes a path the RAS containing the Bundle Head.
If a RAS receives a Query Message for an MP connection for which it
has the Bundle Head, it sends a unicast Response Message to the
querier. Note that no repetition of the Response Message is
necessary because, if it is lost, the querier's next query message
will trigger a new Response Message.
2.4 Contention Handling
If, while sending Query Messages, a Query Message for the same MP
connection is received, it indicates that the Dial-in Node has
brought up multiple links simultaneously. The resolution to this
contention is to elect the bundle head. To do this, each RAS waits
until all Query Messages have been sent (ROBUSTNESS *
QUERY_INTERVAL). At that time, the RAS with the lowest IP address,
as taken from the Query Message, becomes the Bundle Head. That RAS
then sends TWO response messages, with a QUERY_INTERVAL interval, and
indicates to the MP process that a Bundle Head should be formed.
When the other RAS(es) receive the Response Message, they cease
broadcasting (if they haven't already sent all ROBUSTNESS Query
Messages), stop listening for additional Response Messages, and
indicate to their respective MP processes where the Bundle Head
resides.
2.5 MP Operation
MP must use the following algorithm to ensure that there are no
windows of vulnerability during which multiple Bundle Heads might be
created for the same MP connection.
When an MP link is negotiated, MP first checks to see if it already
has the Bundle Head for this connection (i.e., is this a secondary
link). If it does, it should attach to it and not initiate a
discovery. As an optimization, if MP does not have a Bundle Head for
this connection, but does have a existing secondary link for it, MP
should attach to the known Bundle Head without initiating discovery.
If MP knows of no Bundle Head for this connection, it should initiate
a discovery. If the discovery should locate a Bundle Head, it should
attach to the indicated bundle head. If no Bundle Head is found, MP
should create a Bundle Head.
Malkin Expires: 12May98 [Page 5]
Internet Draft MMP November 1997
3. Transfer Protocol
The Layer 2 Tunneling Protocol (L2TP) [L2TP] will be used to transfer
PPP fragments from a RAS containing a secondary link to the RAS
containing the Bundle Head. By specifying the use of an existing
protocol, it is neither necessary to create nor implement a new
protocol.
4. Configuration
There are two required configuration switches and two conditional
configuration switches. None of the switches are optional.
4.1 Robustness - required
This switch sets the number of transmits (repetitions) for Query
Messages. It may be set between 1 and 15. The default is 3. Be
aware that lower settings may create windows of vulnerability.
Higher settings may cause MP timeouts, but may be needed on very
lossy or congested networks.
4.2 Query Interval - required
This switch sets the interval between Query Messages and the interval
between MULTICAST Response Messages. It should be calibrated in
deciseconds (1/10 second) and may be set between 1 and 15. The
default is 1. Be aware that higher settings may cause MP timeouts,
but may be needed on very slow systems/networks.
4.3 Multicast - conditional
This switch selects between the use of the forwardable and
nonforwardable IP multicast addresses for the Discovery Mechanism.
For systems which do not support IP multicast, and therefore use the
limited broadcast address, this switch should not be implemented.
The default value should be nonforwardable. See "IANA
Considerations" (section 6) for multicast address assignments.
4.4 TTL - conditional
This switch sets the IP Time-To-Live (TTL) of all Discovery packets.
For systems which are using the limited broadcast address, this
switch should not be implemented and the TTL should be set to 1. The
default value should be 1.
Malkin Expires: 12May98 [Page 6]
Internet Draft MMP November 1997
5. Security Considerations
No security is designed into the Discovery Mechanism. When using the
non-forwardable multicast address (or limited broadcast address), the
discovery packets are restricted to a single LAN. If the LAN is
physically secure, there is no need for software security. If the
forwardable multicast address is used, but the range is limited to a
small, physically secure network (e.g., a POP), there is still no
need for software security. If the discovery packets are allowed to
cross an internet (and this is NOT recommended for timing reasons),
authentication of RASes may be done with IPSEC. For increased
security on a LAN or in a POP, IPSEC may still be used.
L2TP security is discussed in [L2TP].
6. IANA Considerations
UDP port number: 581
Forwardable multicast address: TBA
Non-forwardable multicast address: TBA
7. References
[MP] "The PPP Multilink Protocol (MP)," K. Sklower, et al., RFC
1990, August 1996.
[L2TP] "Layer 2 Tunneling Protocol 'L2TP'," K. Hamzeh, et al.,
draft-ietf-pppext-l2tp-08.txt, November 1997.
Author's Address
Gary Scott Malkin
Bay Networks
8 Federal Street
Billerica, MA 01821
Phone: +1 (978) 916-4237
Email: gmalkin@baynetworks.com
Malkin Expires: 12May98 [Page 7]
