[Docs] [txt|pdf] [Tracker] [Email] [Nits]

Versions: 00

Internet Draft                           Stefano Salsano (ed.), CoRiTeL
<draft-salsano-bgrpp-arch-00.txt>              Vincenzo Genova, CoRiTeL
                                                Fabio Ricciato, CoRiTeL
May, 2002                                     Martin Winter, Siemens AG
Expiration: November, 2002                     Bert F. Koch, Siemens AG
                                                  Lila Dimopoulou, NTUA
                                               Eugenia Nikolouzou, NTUA
                                             Petros D. Sampatakos, NTUA
                                              Iakovos S. Venieris, NTUA
                                    Gerald Eichler, T-Systems Nova GmbH

Inter-domain QoS Signaling: the BGRP Plus Architecture

     Status of this Memo

This document is an Internet-Draft and is in full conformance with all
provisions of Section 10 of RFC2026.

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.

Distribution of this memo is unlimited.

     Copyright Notice
Copyright (C) The Internet Society (2002).  All Rights Reserved.

Abstract

DiffServ architecture provides a set of "user plane" tools for IP QoS.
Deployment of DiffServ is starting with static, single-domain solutions.
Dynamic QoS and multi-domain QoS are still open issues, as witnessed by
the activity of the IETF NSIS WG, related to the "signaling plane".
In order to realize true end-to-end QoS services in the Internet,
spanning multiple administrative domains, efficient and scalable
signaling and resource control mechanisms are needed.

This document describes a scalable inter-domain resource control
architecture for DiffServ networks. The architecture is called BGRP
Plus, as it extends the previously proposed BGRP framework.


S. Salsano (ed.)        Expires November 2002                        1

      Inter-domain QoS Signaling: the BGRP Plus Architecture    May-02




Table of Contents

1. Introduction.................................................4
2. Requirements for inter-domain resource reservation...........5
3. Reference Network............................................6
4. Scalability of inter-domain resource reservation.............7
5. Quiet grafting...............................................8
6. Inter-domain QoS and Globally Well Known Services............10
7. BGRPP messages...............................................10
8. BGRPP procedures.............................................11
9. BGRP+ and BGRP...............................................13
10. Acknowledgements............................................14
11. Author Information..........................................14
12. References..................................................15
13. APPENDIX A: PSEUDO CODE for Message Processing..............15






































S. Salsano (ed.)        Expires November 2002                        2

      Inter-domain QoS Signaling: the BGRP Plus Architecture    May-02




1. Introduction

The DiffServ architecture specifies a set of "user plane" mechanisms,
which can be used to provide QoS in IP networks. Intentionally, the IETF
DiffServ WG has not covered any "signaling plane" aspect. QoS signaling
capabilities are indeed needed to extend the provisioning of QoS in IP
networks from a static model towards a dynamic one. The IETF WG NSIS
(Next Steps In Signaling) [1] has been specifically chartered to address
the signaling aspects of QoS in IP networks. The NSIS WG is currently
defining the requirements for the QoS signaling mechanisms, and is
considering a set of reference scenarios. One of these scenarios is the
QoS reservation/ negotiation over administrative boundaries or "inter-
domain" QoS.
A fundamental issue in the definition of an inter-domain QoS model is
the scalability, because the ambitious goal is to support QoS services
on the scale of the global Internet.

This draft proposes an architecture that originates from the BGRP
protocol framework proposed in [2], providing "sink tree" based
aggregation for resource reservations over a network of DiffServ
domains. The aggregated reservations are negotiated between so-called
BGRP agents, which are deployed at each BGP-capable border router of
each DiffServ domain. In this way, each domain can perform some kind of
admission control, taking into account the available resources within
that domain. By aggregating the reservations according to the sink trees
created by the BGP routing protocol [3], the number of reservations and
thus the amount of state information stored in the network can be
reduced.

However, aggregation of reservations is just the first step towards
scalability. To limit the signaling load and the processing power
required in the BGRP agents, it is also necessary to reduce the number
of signaling messages. We propose mechanisms for the early response to
reservation messages, in [2] called "quiet grafting", so that not each
message has to travel edge-to-edge through the DiffServ network region.
The architecture proposed in this draft is called BGRP Plus (BGRPP or
BGRP+).

As regards the applicability of this proposal:
1) The BGRPP architecture has been implemented in the context of the
   AQUILA IST project, where a specific access QoS signaling mechanism
   has been defined [4].
2) The BGRPP architecture is largely compliant to the requirements for
   the QoS signaling under definition by NSIS WG.
3) Considering the IntServ over DiffServ architecture as described in
   [5], then an efficient and scalable resource control is also
   essential within the DiffServ region. BGRPP may be a more scalable
   answer to this need, with respect to the "RSVP-aware DiffServ
   region", which is proposed in [5].




S. Salsano (ed.)        Expires November 2002                        3

      Inter-domain QoS Signaling: the BGRP Plus Architecture    May-02



Our discussion is focused on technical aspects and even if we mention
the Service Level Agreements (SLAs) the discussion of financial and
legal aspects of the inter-domain services is intentionally omitted.


2. Requirements for inter-domain resource reservation

An architecture and related protocol for inter-domain resource
reservation should meet the requirements listed in the following
paragraphs. BGRP is a good candidate to meet them.

2.1 Scalability

The architecture has to scale well with the Internet size and growth.

2.2 Autonomous operation

Admission control should be made autonomously in each domain. Each
domain's network operator can independently administer his network and
configure the behavior of inter-domain resource reservation elements
independently each other.

2.3 Follow established routes

Inter-domain resource reservation shall rely on inter-domain routing
(e.g. BGP) for routing decisions. An interface has to be defined between
inter-domain routing and inter-domain resource reservation to allow the
latter to retrieve routing information (e.g. the NEXT_HOP). We assume
that BGP remains unaffected by the inter-domain resource reservation
architecture.

2.4 Independence of each domain

The inter-domain protocol should be independent of the intra-domain
resource control architecture. A domain may use static provisioning as
well as dynamic resource allocation. However, a common interface between
intra-domain and inter-domain resource control has to be defined.

2.5 Influence of SLA and SLS

In the inter-domain scope, the most important constraint is not an
optimization of network resources but the conformity to the SLA with the
neighbor. Each domain has many neighbors, from just a few up to
hundreds, and the administrator signs a contract with each of them. When
a request for a reservation to another AS reaches a BR, it has to choose
a "next hop" to forward the request: if the next hop belongs to another
AS, the BR should make sure that the request is in compliance with the
SLA.






S. Salsano (ed.)        Expires November 2002                        4

      Inter-domain QoS Signaling: the BGRP Plus Architecture    May-02



3. Reference Network

The architecture described in this document assumes a DiffServ region
consisting of several connected, but administratively separated domains.
The traffic can enter and leave the domains at two different types of
routers:

   - An edge router (ER) connects a domain to a network, which is not
     taking part in the BGRPP resource allocation mechanism, e.g. an
     access network.

   - A border router (BR) connects a domain to another domain, which
     also takes part in the BGRPP resource allocation mechanism.

   ___       ________         ________         ________       ___
      \     /        \       /        \       /        \     /
       \   /          \     /          \     /          \   /
    |---| |---|    |---|   |---|    |---|   |---|    |---| |---|
    | R1|-|ER1|    |BR1|---|BR2|    |BR3|---|BR4|    |ER2|-|R2 |
    |---| |---|    |-- |   |---|    |---|   |---|    |---| |---|
       /   \          /     \          /     \          /   \
   ___/     \________/       \________/       \________/     \___

  Access      Source           Transit        Destination    Access
  network     domain           domain           domain       network

            Figure 1: Reference network configuration

A source or destination domain is a domain with at least one edge
router. A transit domain is a domain with at least two border routers,
which forwards traffic received at one border router to another border
router. All edge routers and border routers are required to run BGP.

Corresponding to each border router, a BGRP agent is instantiated as
shown in the following figure:

                  |-----| |-----|  |-----| |-----|
                  |BGRPP|-|BGRPP|--|BGRPP|-|BGRPP|
                  |-----| |-----|  |-----| |-----|
                     :       :        :       :
   ___       ________:       :________:       :________       ___
      \     /        \       /        \       /        \     /
       \   /         :\     /:        :\     /:         \   /
    |---| |---|    |---|   |---|    |---|   |---|    |---| |---|
    | R1|-|ER1|    |BR1|---|BR2|    |BR3|---|BR4|    |ER2|-|R2 |
    |---| |---|    |-- |   |---|    |---|   |---|    |---| |---|
       /   \          /     \          /     \          /   \
   ___/     \________/       \________/       \________/     \___

  Access      Source           Transit        Destination    Access
  network     domain           domain           domain       network

                       Figure 2: BGRPP agents

S. Salsano (ed.)        Expires November 2002                        5

      Inter-domain QoS Signaling: the BGRP Plus Architecture    May-02




We assume that the source domain is capable of performing some kind of
per flow admission control, taking into account the available resources
up to BR1. The source domain has however no information about the
availability of resources along the further path of the reservation
through other domains.

To perform inter-domain admission control, the source domain determines
the egress border router and contacts the corresponding BGRPP agent.
Through the BGRPP protocol this agent is able to determine, whether
resources are available in each domain along the path and on the links
connecting the domains. Each BGRPP agent cares for the resources on the
next path segment from its associated border router towards the
destination.

Resource reservation in the source and destination domain is not the
task of the BGRPP protocol. However, as we describe later, BGRPP has to
provide mechanisms to enable resource reservation in the destination
domain.

There are mainly two types of messages involved in the reservation set-
up:

   - PROBE messages are initiated at the source domain and are
     forwarded between BGRPP agents along the BGP path. They check the
     conformance of the request with the SLAs between neighbored
     domains. The path is recorded in the message, to enable the
     response to take the same way in the backward direction.
   - GRAFT messages indicate the availability of resources towards the
     destination domain. During message processing, the resource
     availability is checked and the resources are actually reserved in
     each path segment.


4. Scalability of inter-domain resource reservation

BGRPP is an inter-domain protocol to allow resource reservation. The
main issue that BGRPP should tackle is the scalability problem.
This problem is related to two aspects:

- the handling of state information for each reserved flow and
- the processing of reservation messages.

In particular, the scalability problem is related to:

- memory needs for each router that runs reservation protocol;
- CPU usage for message processing;
- bandwidth utilization for messages.

A scalable solution will minimize these three elements.

BGRP as described in [2] mainly addresses the scalability in terms of
the amount of state information kept in each BGRP agent. It aggregates

S. Salsano (ed.)        Expires November 2002                        6

      Inter-domain QoS Signaling: the BGRP Plus Architecture    May-02



reservations along the BGP sink trees and thus achieves a scalability
behavior, where the memory requirements are proportional to the number
of sink trees with simultaneous active reservations.

However, in order to check the resource availability, BGRP messages have
still to travel along the full path to the destination domain, asking
each BGRP agent to check the resource availability for that part of the
network it is responsible for. So the message processing overhead may be
rather large, especially in big backbone networks.

In order to solve this problem, an hint is given in [2]: resources may
be kept in advance at a BGRP agent, so that further requests may be
already terminated at an earlier stage. In this document we provide the
complete functional specification of this mechanism, called "quiet
grafting". Together with the sink-tree-based aggregations, this
mechanism provides a scalable solution for inter-domain resource
reservations.


5. Quiet grafting

5.1 Requirements

When an inter-domain reservation is initiated at a source domain, the
first BGRPP agent constructs a PROBE message indicating the amount of
bandwidth required and the destination address. It is not evident, to
which sink tree this reservation will belong. So the PROBE message is
forwarded hop-by-hop between BGRPP agents along the BGP route to the
destination. Obviously, the last BGRPP agent, which corresponds to the
root of the sink tree, can assign a sink tree id to the reservation. The
GRAFT message sent back will contain this sink tree id. As this message
travels back to the source domain, it installs the necessary sink tree
reservations in the path segments.

To enable an intermediate BGRPP agent to answer a PROBE message
successfully with a positive response, the following conditions have to
be met:

  -  The BGRPP agent must be able to determine the sink tree, to which
     the reservation belongs.
  -  The BGRPP agent must have pre-reserved resources for this sink
     tree, so that he can guarantee, that the resources are available
     on the path segment from the current point to the destination
     domain.
  -  As the last BGRPP agent may no longer be informed about a new
     reservation, the BGRPP agent must provide means to contact the
     destination domain, so that resources can also be reserved on the
     not-BGRPP-controlled path segment from BR4 to ED2 (see figure 2).

The following paragraphs describe the solutions we propose to solve
these requirements.



S. Salsano (ed.)        Expires November 2002                        7

      Inter-domain QoS Signaling: the BGRP Plus Architecture    May-02



5.2 NLRI Labeling for Sink Tree Identification

According to [2], a sink tree is identified by the destination AS number
and a border router id. The network layer reachability information
(NLRI) associated with the root of the tree can be used to identify the
tree at a distant point in the network. As BGP may aggregate this
information into aggregated routes, it is not directly derivable from
BGP routing information. Instead, we propose to propagate the NLRI back
from the root of the sink tree in GRAFT messages and to store this
information in each BGRPP agent, which processes the message.

When a new PROBE message arrives, the BGRPP agent will compare the
destination address with the already stored NLRI thus trying to identify
the sink tree. Successful sink tree identification is a prerequisite to
the following steps, to perform successful quiet grafting.

To reduce memory requirements, BGRPP agents will only store NLRI
information for those sink trees, where actual reservations exist.

5.3 Pre-reservation

When a BGRPP agent processing a PROBE message was able to determine the
sink tree, he will now check, whether he has pre-reserved resources on
this sink tree. If this is the case, he can generate a GRAFT message and
terminate the PROBE at this stage. If there are not sufficient resources
available, the BGRPP agent will further forward the GRAFT message to the
next BGRPP agent, as determined by the NEXT_HOP attribute of the BGP
path to the destination.

The amount of pre-reserved resources that is available at a given time
for a sink tree is called resource cushion. A possible solution to build
this resource cushion at each BGRPP agent is the "delayed resource
release" mechanism. This means, that a BGRPP agent will not immediately
release unused resources, but instead keep them in an attempt to satisfy
further requests. Resources are however released, when they are unused
for some time.

We will describe the proposed algorithm in a separate document and
provide information about its performance in various scenarios.

5.4 Signaling in the last domain

When a PROBE message is answered "in advance", the last domain may not
be informed about the new request and cannot reserve resources within
that domain. To include the resource availability in the last domain in
the overall admission control, we propose to back-propagate a reference
to the interface to intra-domain resource control at the root of the
sink tree within the GRAFT message. This information is also stored
within each BGRPP agent that handles this message.

The originating BGRPP agent can then use this reference to directly
request the required resources in the destination domain.


S. Salsano (ed.)        Expires November 2002                        8

      Inter-domain QoS Signaling: the BGRP Plus Architecture    May-02



This reference is also necessary for the release of a reservation: since
resources are not immediately released, when the original end-user
request is removed, possibly the destination domain will not be informed
at all about this event. For this purpose it is necessary that the
originator of the request can directly inform the destination domain.

6. Inter-domain QoS and Globally Well Known Services

In the previous sections we have mentioned "reservations" which are
originated by a source domain (and propagated up to the destination
domain) and are characterized by a bandwidth parameter. Actually, a
reservation should be also be characterized by the required service and
other parameters besides the bandwidth could be needed. We assume that a
set of "Globally Well Known Services" is defined to characterize the
required service. For the sake of simplicity, in the following
description of messages and procedures, we also assume that the
bandwidth is the only needed parameter.


7. BGRPP messages

7.1 PROBE message
It consists of a reservation request and destination network
information. It travels from origin AS towards destination AS and
collects routing information. It contains the information of the
requested amount of resources.

The fields of PROBE message are:

   Sender:   It is a BGRPP agent identifier, it is composed of AS id
             and the IP address of BR that has sent the PROBE.
   ProbeId:  The origin AS chooses the ProbeId. The ProbeID scope is
             local to the originating BGRPP agent and only used there
             to match the GRAFT (or ERROR) message. In other words,
             only the originating AS stores a "PROBE state" while
             Intermediate ASs nodes does not need correlate PROBE and
             GRAFT.
   GwksId:   GWKS id
   Destination: IP address prefix of the host or network that is
             destination of request
   RequiredBW:  Requested bw [bit/s]
   TreeId:   Id of sink tree, it is NULL if the sink tree is unknown.
             This information is actually redundant, as each node could
             check weather the Destination matches an existing sink
             tree. By avoiding this check, it enhances the
             performances.
   Path:     Record of the route in terms of BGRPP agents from origin
             AS


7.2 GRAFT message
When the destination AS (or a transit AS if it can do quiet grafting)
receives a PROBE and it can accept the request, it returns a GRAFT

S. Salsano (ed.)        Expires November 2002                        9

      Inter-domain QoS Signaling: the BGRP Plus Architecture    May-02



towards the origin AS, using the Path information collected by PROBE
message. This message carries sink tree information. It also carries the
amount of resources that are reserved by the node sending the GRAFT.

The fields of GRAFT message are:
   Sender:   It is a BGRPP agent identifier, it is composed of AS id
             and the IP address of BR that has sent the GRAFT.
   Id:       Msg id to match GRAFT with PROBE, echoed from PROBE
   GwksId:   GWKS id
   Destination: Ip address prefix of the host or network that is
             destination of request, echoed from PROBE
   ReservedBW:  Reserved bw [bit/s]
   TreeId:   Id of sink tree
   DestResMgr:  Reference to destination domain reservation manager for
             reservation in the last domain
   AddressPrefixList:   NLRI
   Path:     Record of the route in terms of BGRPP agents to be
             followed back

7.3 REFRESH message
It contains the indication of the current amount of needed BW.
It is sent by a previous (upstream) node to reduce the amount of
requested resource. A REFRESH with zero bandwidth is used to tear down a
reservation.
REFRESH messages must never be used by a previous (upstream) node to
increase the amount of requested resources.


   Sender:   It is a BGRPP agent identifier, it is composed of AS id
             and the IP address of BR that has sent the REFRESH.
   GwksId:   GWKS id
   ActualBw: actual reserved bandwidth
   TreeId:   id of sink tree

7.4 ERROR message
It indicates that some error has occurred. It contains a description of
the specific error case. For example if the resources are not available,
an ERROR message is sent and propagated backwards up to the originator
of the request.

7.5 TEAR message
It was defined in the BGRP architecture. Currently, it has no use in the
BGRPP architecture, because it is replaced by the use of refresh
messages.

8. BGRPP procedures

8.1 State information in BGRPP agents

In order to process the BGRPP protocol messages, the BGRPP agent stores
the following sink tree status information:



S. Salsano (ed.)        Expires November 2002                       10

      Inter-domain QoS Signaling: the BGRP Plus Architecture    May-02



For each sink tree (with at least one reservation) and for each service,
the next ("outgoing") hop and a list of previous ("incoming") hops is
stored. The value of reserved resources for the outgoing hop and for
each incoming hop is stored (this information is replicated per each
GWKS).

For each sink tree, the NLRI is stored, to enable sink tree
identification for quiet grafting (see 5.2). Additionally, a reference
to the intra-domain resource control of the destination domain is
stored, to enable signaling to the last domain (see 5.4).

TABLE 1 - Sink tree x state information
-----------------------------------------------------------------------
   Next hop       IP address of next hop BGRPP agent
   Outgoing.res[g]Reserved BW for the sink tree x towards the Next hop
                  for the GWKS g
   Previous hop[i]IP addresses of previous hop BGRPP agents
   Incoming[i].res[g]   Reserved BW for the sing tree assigned to each
                  Previous hop for the GWKS g
   NLRI           NLRI for the sink tree
   Intra-dom. RC Ref. Reference to the intra-domain resource control of
                  the destination domain
-----------------------------------------------------------------------

The resource cushion is the difference between the Outgoing.res and the
sum of the Incoming[i].res.


8.2 Message handling

(In the following the dependence from the GWKS g will always be
omitted).
The message handling for a transit AS is described, the procedures for
Originating AS and Terminating AS can be derived. In APPENDIX A a more
complete pseudo-code description of the message handling is provided.

- PROBE messages

The PROBE messages can be:
    - rejected because SLA/SLS does not match or there are no resources
      on outgoing link or on intra-domain links
    - accepted with Quiet Grafting
    - forwarded to downstream node with no change in RequiredBW

When a PROBE is received a preliminary admission control is performed.
If the BGRPP agent is at an ingress Border Router, it checks for SLA
between the requester AS and its own AS and decides if the request can
be accepted. The BGRPP agent could also check if intra-domain resource
manager could accept the new request. If the BGRPP agent is associated
to an egress BR it has to check the SLA with the AS of next hop BR. It
can also check for the resources on the outgoing link towards the next
hop BR.


S. Salsano (ed.)        Expires November 2002                       11

      Inter-domain QoS Signaling: the BGRP Plus Architecture    May-02



If any of these checks fail, the request cannot be accepted and an ERROR
message is sent to previous hop.
If the checks are OK, the node (both the ingress and the egress BR)
first verifies if the Destination of the PROBE matches an existing sink
tree. In this case it is first checked if the resource cushion can fit
the RequiredBW. In case of positive answer, Quiet Grafting is performed:
the Incoming[PH].res related to the previous hop is increased by the
RequiredBW value and a GRAFT message is sent to the Previous hop. If the
Destination of the PROBE does not match an existing sink tree, or the
resource cushion is not enough, the PROBE is forwarded and the state
information of the node is not touched in any way.

- GRAFT messages

The resource reservation is performed when a GRAFT arrives. A GRAFT
tells the receiving node the downstream node can accept a given amount
of BW for a given sink tree. Now the receiving node should decide if it
has the resources to send this amount of BW towards the next hop BR and
if the SLA matches. In particular, an Ingress BR should check the
Incoming SLS with upstream AS and should consider the availability of
intra-domain resources towards egress BR. An Egress BR should check the
Outgoing SLS towards the downstream AS and should consider the resources
on the outgoing link towards the Ingress BR of the downstream AS. (The
nodes could have already performed these checks during PROBE phase, but
the situation could have changed).
If the resources are available and SLA/SLS matches, the graft is
accepted and the node:
1) Increases the Outgoing.res by the ReservedBW value.
2) Increases the Incoming[PH].res of the Previous hop PH by the
   ReservedBW value.
3) Propagates the GRAFT to the Previous hop.
If the resources are not available or (incoming/outgoing) SLA/SLS not
matching, the GRAFT cannot be accepted. The node will immediately send
an ERROR towards the first originator of the PROBE. If the reason of
refusal is outgoing SLA not matching or unavailability of resources, the
node has also to release reserved BW to next hop because it can not
utilize it. To this purpose the node sends a REFRESH message to its next
hop. If the reason of refusal is incoming SLA not matching the node can
maintain the additional reserved BW towards its next hop to enlarge
resource cushion.

9. BGRP+ and BGRP

It is worth to recall the extensions of this draft with respect to the
original BGRP proposal:
  -  The architecture includes the aspects related to the interactions
     with intra-domain resource reservation mechanisms
  -  The quiet grafting mechanisms simply mentioned in the BGRP
     proposal has been fully specified
  -  The mechanisms to match a flow into the sink-tree by means of the
     NLRI information and to distribute this NLRI information where
     needed have been specified


S. Salsano (ed.)        Expires November 2002                       12

      Inter-domain QoS Signaling: the BGRP Plus Architecture    May-02



  -  The issue of the missing signaling in the last domain due to quiet
     grafting has been addressed and resolved
  -  The semantic content of the messages has been described
  -  The procedures for message handling have been given (both in text
     and using a pseudo-code)


10. Acknowledgements

Part of this work has been funded under the European Commission 5th
framework IST program.

The authors would like to acknowledge all their colleagues in the AQUILA
project for their important contribution to this work.


11. Author Information

Stefano Salsano
CoRiTeL - Consorzio di Ricerca sulle Telecomunicazioni
Via Anagnina, 203
00040 Morena Roma - ITALY
e-mail: salsano@coritel.it

Vincenzo Genova, CoRiTeL
e-mail: genova@coritel.it

Fabio Ricciato, CoRiTeL
e-mail: ricciato@coritel.it

Martin Winter,
Siemens AG, ICN WN CC EK A19
81359 Munich - Germany
e-mail: martin.winter@icn.siemens.de

Bert F. Koch, Siemens AG
e-mail: bert.koch@icn.siemens.de

Lila Dimopoulou
NTUA - National Technical University of Athens
Heroon Polytechniou 9, 157 73, Athens GREECE
e-mail: lila@telecom.ntua.gr

Eugenia Nikolouzou, NTUA
e-mail: enik@telecom.ntua.gr

Petros D. Sampatakos, NTUA
e-mail: psampa@telecom.ntua.gr

Iakovos S. Venieris, NTUA
e-mail: ivenieri@cc.ece.ntua.gr

Gerald Eichler, T-Systems Nova GmbH

S. Salsano (ed.)        Expires November 2002                       13

      Inter-domain QoS Signaling: the BGRP Plus Architecture    May-02



e-mail: Gerald.Eichler@t-systems.com

12. References

 [1]M. Brunner (Editor), "Requirements for QoS Signaling Protocols",
    draft-ietf-nsis-req-01.txt, April 2002, Work in Progress
 [2]P. Pan, E. Hahne, H. Schulzrinne: "BGRP: Sink-Tree-Based
    Aggregation for Inter-Domain Reservations", Journal of
    Communications and Networks, Vol. 2, No. 2, June 2000, pp. 157-167,



    http://www.cs.columbia.edu/~pingpan/papers/bgrp.pdf
 [3]Y. Rekhter, T.J. Watson, T. Li: "A Border Gateway Protocol 4 (BGP-
    4)", RFC 1771, March 1995
 [4]AQUILA IST Project http://www.ist-aquila.org/
 [5]Y. Bernet et al., "Integrated Services Over Diffserv Networks", RFC
    2998, November 2000


13. APPENDIX A: PSEUDO CODE for Message Processing

13.1 PROBE processing
// State variables for PROBE processing:
treeOutgoing.res
treeIncoming[i].res

// information related to messages:
probe.required
new_graft.reserved
new_probe.required
new_error.failedBW

// Local variables:
cushion
request

// PROBE processing:

// perform BGP lookup to determine the BgrpNeighbour

// test incoming SLA
if (no resources in SLA)
   new_error.failedBW=probe.requiredBW
   send error message to previous hop
   exit
endif

// try to identify the sink tree (DestinationDomain) for the message
if (PROBE does not contain a tree id)
   // perform lookup in NLRI table to find the tree
   if (tree found)
      insert tree id in PROBE
   endif //
endif


S. Salsano (ed.)        Expires November 2002                       14

      Inter-domain QoS Signaling: the BGRP Plus Architecture    May-02



//check quiet grafting possibility
request = probe.required
cushion=treeOutgoing.res-sum(treeIncoming[i].res)
if (cushion>=request && tree found)
   // perform quiet grafting
   treeIncoming[IN].res += request
   // send graft message to previous hop
   new_graft.reserved=request
else
   // quiet grafting is not possible
   // test outgoing SLA
   if (no resources in SLA)
      // send error message to previous hop
      new_error.failedBW=probe.required
   else
      // prepare and forward PROBE message to next hop
      // update route hop
      new_probe.required=request
   endif
endif

13.2 GRAFT processing
// State variables:
treeOutgoing.res
treeIncoming[IN].res


// information related to messages:
graft.reserved
new_graft.reserved
new_error.failedBW


// GRAFT processing:

// check outgoing SLA and at BGRPP ingress node request intradomain
resource
if (no resources available)
   // send error message to previous hop
   new_error.failedBW= graft.reserved
   // send refresh message to next hop
   new_refresh.actualBW=treeOutgoing.res
else
   // update reserved bandwidth
   treeOutgoing.res += graft.reserved
endif

// check incoming SLA
if (no resources available)
   // send error message to previous hop
   new_error.failedBW= graft.reserved
else
   // update reserved bandwidth

S. Salsano (ed.)        Expires November 2002                       15

      Inter-domain QoS Signaling: the BGRP Plus Architecture    May-02



   treeIncoming[IN].res += graft.reserved
   // prepare and forward GRAFT message to previous hop
   update route hop
   new_graft.reserved = graft.reserved
endif

13.3 REFRESH processing
// State variables:
treeIncoming[IN].res
treeOutgoing.res

// information related to messages:
refresh.actualBW
refresh.sender
new_refresh.actualBW

// REFRESH processing:

if(refresh.sender is next hop for tree)
   // next hop knows more than I
   if (refresh.actualBW > treeOutgoing.res)
      error situation
      //Send to next hop a REFRESH message
      new_refresh.actualBW=treeOutgoing.res

   // next hop thinks less than I
   if (refresh.actualBW < treeOutgoing.res)
      error situation
else
   // previous hop wants less than I know (like a TEAR)
   if (refresh.actualBW < treeIncoming[IN].res)
      treeIncoming.res = refresh.bw
   // previous hop thinks more than I
   if (refresh.actualBW > treeIncoming.res)
      error situation
end if

13.4 ERROR processing
// information related to messages
error.failedBW
new_error.failedBW

// ERROR processing
if(this hop is destination)
   advise intradomain requester
else
// prepare and forward ERROR message to previous hop
// update route hop
new_error.failedBW = error.failedBW





S. Salsano (ed.)        Expires November 2002                       16


Html markup produced by rfcmarkup 1.123, available from https://tools.ietf.org/tools/rfcmarkup/