RFC 2643
Cabletron's SecureFast VLAN Operational Model
Pages: 60
Informational
Informational
Part 1 of 2 – Pages 1 to 27
Network Working Group D. Ruffen Request for Comments: 2643 T. Len Category: Informational J. Yanacek Cabletron Systems Incorporated August 1999 Cabletron's SecureFast VLAN Operational Model Version 1.8 Status of this Memo This memo provides information for the Internet community. It does not specify an Internet standard of any kind. Distribution of this memo is unlimited. Copyright Notice Copyright (C) The Internet Society (1999). All Rights Reserved.Abstract
Cabletron's SecureFast VLAN (SFVLAN) product implements a distributed connection-oriented switching protocol that provides fast forwarding of data packets at the MAC layer. The product uses the concept of virtual LANs (VLANs) to determine the validity of call connection requests and to scope the broadcast of certain flooded messages.Table of Contents
1. Introduction............................................. 3 1.1 Data Conventions..................................... 3 1.2 Definitions of Commonly Used Terms................... 4 2. SFVLAN Overview.......................................... 6 2.1 Features............................................. 7 2.2 VLAN Principles...................................... 8 2.2.1 Default, Base and Inherited VLANs.............. 8 2.2.2 VLAN Configuration Modes....................... 8 2.2.2.1 Endstations............................ 8 2.2.2.2 Ports.................................. 9 2.2.2.3 Order of Precedence.................... 9 2.2.3 Ports with Multiple VLAN Membership............ 10 2.3 Tag/Length/Value Method of Addressing................ 10 2.4 Architectural Overview............................... 11 3. Base Services............................................ 13 4. Call Processing.......................................... 14 4.1 Directory Service Center............................. 14 4.1.1 Local Add Server............................... 15
4.1.2 Inverse Resolve Server......................... 15
4.1.3 Local Delete Server............................ 18
4.2 Topology Service Center.............................. 18
4.2.1 Neighbor Discovery Server...................... 18
4.2.2 Spanning Tree Server........................... 18
4.2.2.1 Creating and Maintaining
the Spanning Tree........... 19
4.2.2.2 Remote Blocking........................ 19
4.2.3 Link State Server.............................. 20
4.3 Resolve Service Center............................... 21
4.3.1 Table Server................................... 22
4.3.2 Local Server................................... 22
4.3.3 Subnet Server.................................. 22
4.3.4 Interswitch Resolve Server..................... 22
4.3.5 Unresolvable Server............................ 23
4.3.6 Block Server................................... 23
4.4 Policy Service Center................................ 24
4.4.1 Unicast Rules Server........................... 24
4.5 Connect Service Center............................... 25
4.5.1 Local Server................................... 25
4.5.2 Link State Server.............................. 25
4.5.3 Directory Server............................... 26
4.6 Filter Service Center................................ 26
4.7 Path Service Center.................................. 26
4.7.1 Link State Server.............................. 26
4.7.2 Spanning Tree Server........................... 27
4.8 Flood Service Center................................. 27
4.8.1 Tag-Based Flood Server......................... 27
5. Monitoring Call Connections.............................. 27
5.1 Definitions.......................................... 27
5.2 Tapping a Connection................................. 28
5.2.1 Types of Tap Connections....................... 28
5.2.2 Locating the Probe and Establishing
the Tap Connection.......... 29
5.2.3 Status Field................................... 30
5.3 Untapping a Connection............................... 31
6. Interswitch Message Protocol (ISMP)...................... 32
6.1 General Packet Structure............................. 32
6.1.1 Frame Header................................... 32
6.1.2 ISMP Packet Header............................. 33
6.1.2.1 Version 2.............................. 33
6.1.2.2 Version 3.............................. 34
6.1.3 ISMP Message Body.............................. 35
6.2 Interswitch BPDU Message............................. 35
6.3 Interswitch Remote Blocking Message.................. 36
6.4 Interswitch Resolve Message.......................... 37
6.4.1 Prior to Version 1.8........................... 37
6.4.2 Version 1.8.................................... 41
6.5 Interswitch New User Message......................... 46
6.6 Interswitch Tag-Based Flood Message.................. 49
6.6.1 Prior to Version 1.8........................... 49
6.6.2 Version 1.8.................................... 52
6.7 Interswitch Tap/Untap Message........................ 55
7. Security Considerations.................................. 58
8. References............................................... 58
9. Authors' Addresses....................................... 59
10. Full Copyright Statement................................ 60
1. Introduction
This memo is being distributed to members of the Internet community
in order to solicit reactions to the proposals contained herein.
While the specification discussed here may not be directly relevant
to the research problems of the Internet, it may be of interest to
researchers and implementers.
1.1 Data Conventions
The methods used in this memo to describe and picture data adhere to
the standards of Internet Protocol documentation [RFC1700]. In
particular:
The convention in the documentation of Internet Protocols is to
express numbers in decimal and to picture data in "big-endian"
order. That is, fields are described left to right, with the most
significant octet on the left and the least significant octet on
the right.
The order of transmission of the header and data described in this
document is resolved to the octet level. Whenever a diagram shows
a group of octets, the order of transmission of those octets is
the normal order in which they are read in English.
Whenever an octet represents a numeric quantity the left most bit
in the diagram is the high order or most significant bit. That
is, the bit labeled 0 is the most significant bit.
Similarly, whenever a multi-octet field represents a numeric
quantity the left most bit of the whole field is the most
significant bit. When a multi-octet quantity is transmitted the
most significant octet is transmitted first.
1.2 Definitions of Commonly Used Terms
This section contains a collection of definitions for terms that have
a specific meaning for the SFVLAN product and that are used
throughout the text.
Switch ID
A 10-octet value that uniquely identifies an SFVLAN switch within
the switch fabric. The value consists of the 6-octet base MAC
address of the switch, followed by 4 octets of zeroes.
Network link
The physical connection between two switches. A network link is
associated with a network interface (or port) of a switch.
Network port
An interface on a switch that attaches to another switch.
Access port
An interface on a switch that attaches to a user endstation.
Port ID
A 10-octet value that uniquely identifies an interface of a
switch. The value consists of the 6-octet base MAC address of the
switch, followed by the 4-octet local port number of the
interface.
Neighboring switches
Two switches attached to a common (network) link.
Call connection
A mapping of user traffic through a switch that correlates the
source and destination address pair specified within the packet to
an inport and outport pair on the switch.
Call connection path
A set of 0 to 7 network links over which user traffic travels
between the source and destination endstations. Call connection
paths are selected from a list of alternate equal cost paths
calculated by the VLS protocol [IDvlsp], and are chosen to load
balance traffic across the fabric.
Ingress switch
The owner switch of the source endstation of a call connection.
That is, the source endstation is attached to one of the local
access ports of the switch.
Egress switch
The owner switch of the destination endstation of a call
connection. That is, the destination endstation is attached to
one of the local access ports of the switch.
Intermediate switches
Any switch along the call connection path on which user traffic
enters and leaves over network links. Note that the following
types of connections have no intermediate switches:
- Call connections between source and destination endstations
that are attached to the same switch -- that is, the ingress
switch is the same as the egress switch. Note also that the
path for this type of connection consists of 0 network links.
- Call connections where the ingress and egress switches are
physical neighbors connected by a single network link. The
path for this type of connection consists of a single network
link.
InterSwitch Message protocol (ISMP)
The protocol used for interswitch communication between SFVLAN
switches.
Undirected messages
Messages that are (potentially) sent to all SFVLAN switches in the
switch fabric -- that is, they are not directed to any particular
switch. ISMP messages with a message type of 5, 7 or 8 are
undirected messages.
Switch flood path
The path used to send undirected messages throughout the switch
fabric. The switch flood path is formed using a spanning tree
algorithm that provides a single path through the switch fabric
that guarantees loop-free delivery to every other SFVLAN switch in
the fabric.
Upstream Neighbor
That switch attached to the inport of the switch flood path --
that is, the switch from which undirected messages are received.
Note that each switch receiving an undirected message has, at
most, one upstream neighbor, and the originator of any undirected
ISMP message has no upstream neighbors.
Downstream Neighbors
Those switches attached to all outports of the switch flood path
except the port on which the undirected message was received.
Note that for each undirected message some number of switches have
no downstream neighbors.
Virtual LAN (VLAN) identifier
A VLAN is a logical grouping of ports and endstations such that
all ports and endstations in the VLAN appear to be on the same
physical (or extended) LAN segment even though they may be
geographically separated.
A VLAN identifier consists of a variable-length string of octets.
The first octet in the string contains the number of octets in the
remainder of the string -- the actual VLAN identifier value. A
VLAN identifier can be from 1 to 16 octets long.
VLAN policy
Each VLAN has an assigned policy value used to determine whether a
particular call connection can be established. SFVLAN recognizes
two policy values: Open and Secure.
2. SFVLAN Overview
Cabletron's SecureFast VLAN (SFVLAN) product implements a distributed
connection-oriented switching protocol that provides fast forwarding
of data packets at the MAC layer.
2.1 Features
Within a connection-oriented switching network, user traffic is routed through the switch fabric based on the source and destination address (SA/DA) pair found in the arriving packet. For each SA/DA pair encountered by a switch, a "connection" is programmed into the switch hardware. This connection maps the SA/DA pair and the port on which the packet was received to a specific outport over which the packet is to be forwarded. Thus, once a connection has been established, all packets with a particular SA/DA pair arriving on a particular inport are automatically forwarded by the switch hardware out the specified outport. A distributed switching environment requires that each switch be capable of processing all aspects of the call processing and switching functionality. Thus, each switch must synchronize its various databases with all other switches in the fabric or be capable of querying other switches for information it does not have locally. SFVLAN accomplishes the above objectives by providing the following features: - A virtual directory of the entire switch fabric. - Call processing for IP, IPX and MAC protocols. - Automatic call connection, based on VLAN policy. - Automatic call rerouting around failed switches and links. In addition, SFVLAN optimizes traffic flow across the switch fabric by providing the following features: - Broadcast interception and address resolution at the ingress port. - Broadcast scoping, restricting the flooding of broadcast packets to only those ports that belong to the same VLAN as the packet source. - A single loop-free path (spanning tree) used for the flooding of undirected interswitch control messages. Only switches running the SFVLAN switching protocol are included in this spanning tree calculation -- that is, traditional bridges or routers configured for bridging are not included. - Interception of both service and route advertisements with readvertisement sourced from the MAC address of the original advertiser.
2.2 VLAN Principles
Each SFVLAN switch port, along with its attached endstations, belongs to one or more virtual LANs (VLANs). A VLAN is a logical grouping of ports and endstations such that all ports and endstations in the VLAN appear to be on the same physical (or extended) LAN segment even though they may be geographically separated. VLAN assignments are used to determine the validity of call connection requests and to scope the broadcast of certain flooded messages.2.2.1 Default, Base and Inherited VLANs
Each port is explicitly assigned to a default VLAN. At start-up, the default VLAN to which all ports are assigned is the base VLAN -- a permanent, non-deletable VLAN to which all ports belong at all times. The network administrator can change the default VLAN of a port from the base VLAN to any other unique VLAN by using a management application known here as the VLAN Manager. A port's default VLAN is persistent -- that is, it is preserved across a switch reset. When an endstation attaches to a port for the first time, it inherits the default VLAN of the port. Using the VLAN Manager, the network administrator can reassign an endstation to another VLAN. Note: When all ports and all endstations belong to the base VLAN, the switch fabric behaves like an 802.1D bridging system.2.2.2 VLAN Configuration Modes
For both ports and endstations, there are a variety of VLAN configuration types, or modes.2.2.2.1 Endstations
For endstations, there are two VLAN configuration modes: inherited and static. - Inherited An inherited endstation becomes a member of its port's default VLAN.
- Static
A static port becomes a member of the VLAN to which it has been
assigned by the VLAN Manager.
The default configuration mode for an endstation is inherited.
2.2.2.2 Ports
For ports, there are two VLAN configuration modes: normal and
locked.
- Normal
All inherited endstations on a normal port become members of the
port's default VLAN. All static endstations are members of the
VLAN to which they were mapped by the VLAN Manager.
If the VLAN Manager reassigns the default VLAN of a normal port,
the VLAN(s) for the attached endstations may or may not change,
depending on the VLAN configuration mode of each endstation. All
inherited endstations will become members of the new default VLAN.
All others will retain membership in their previously mapped
VLANs.
- Locked
All endstations attached to a locked port can be members only of
the port's default VLAN.
If the VLAN Manager reconfigures a normal port to be a locked
port, all endstations attached to the port become members of the
port's default VLAN, regardless of any previous VLAN membership.
The default configuration mode for ports is normal.
2.2.2.3 Order of Precedence
On a normal port, static VLAN membership prevails over inherited
membership.
On a locked port, default VLAN membership prevails over any static
VLAN membership.
If a statically assigned endstation moves from a locked port back to
a normal port, the endstation's static VLAN membership must be
preserved.
2.2.3 Ports with Multiple VLAN Membership
A port can belong to multiple VLANs, based on the VLAN membership of its attached endstations. For example, consider a port with three endstations, a default VLAN of "blue" and the following endstation VLAN assignments: - One of the endstations is statically assigned to VLAN "red." - Another endstation is statically assigned to VLAN "green." - The third endstation inherits the default VLAN of "blue." In this instance, the port is explicitly a member of VLAN "blue." But note that it is also implicitly a member of VLAN "red" and VLAN "green." Any tag-based flooding (Section 4.8) directed to any one of the three VLANs ("red," "green," or "blue") will be forwarded out the port.2.3 Tag/Length/Value Method of Addressing
Within most computer networks, the concept of "address" is somewhat elusive because different protocols can (and do) use different addressing schemes and formats. For example, Ethernet (physical layer) addresses are six octets long, while IP (network layer) addresses are only four octets long. To distinguish between the various protocol-specific forms of addressing, many software modules within the SFVLAN product specify addresses in a format known as Tag/Length/Value (TLV). This format uses a variable-length construct as shown below: 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Tag | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Value length | | +-+-+-+-+-+-+-+-+ + | Address value | : : | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ Tag This 4-octet field specifies the type of address contained in the structure. The following address types are currently supported:
Tag name Value Address type
aoMacDx 1 DX ethernet dst/src/type
aoIpxSap 2 Sap
aoIpxRIP 3 RIP
aoInstYP 4 YP (YP name and version)
aoInstUDP 5 UDP (Port #)
aoIpxIpx 6 Ipx
aoInetIP 7 IP (Net address)
aoInetRPC 8 RPC (Program #)
aoInetRIP 9 INET RIP
aoMacDXMcast 10 Multicast unknown type
aoAtDDP 11 AppleTalk DDP
aoEmpty 12 (no address type specified)
aoVlan 13 VLAN identifier
aoHostName 14 Host name
aoNetBiosName 15 NetBIOS name
aoNBT 16 NetBIOS on TCP name
aoInetIPMask 17 IP Subnet Mask
aoIpxSap8022 18 Sap 8022 type service
aoIpxSapSnap 19 Sap Snap type service
aoIpxSapEnet 20 Sap Enet type service
aoDHCPXID 21 DHCP Transaction ID
aoIpMcastRx 22 IP class D receiver
aoIpMcastTx 23 IP class D sender
aoIpxRip8022 24 Ipx Rip 8022 type service
aoIpxRipSnap 25 Ipx Rip type service
aoIpxRipEnet 26 Ipx Rip Enet service
aoATM 27 ATM
aoATMELAN 28 ATM LAN Emulation Name
Value length
This 1-octet field contains the length of the value of the
address. The value here depends on the address type and actual
value.
Address value
This variable-length field contains the value of the address. The
length of this field is stored in the Value length field.
2.4 Architectural Overview
The SFVLAN software executes in the switch CPU and consists of the
following elements as shown in Figure 1:
- The SFVLAN base services that handles traffic intercepted by the
switch hardware. The base services are described in Section 3.
+------------------------------------------------------+
| +-----+ |
| +------------+ | I | |
| | CALL TAP <--(8)--> N | |
| +------------+ | T | |
| | E | |
| +-----------+ +------------+ | R | |
| | PATH | | TOPOLOGY | | S | |
| | | | | | W | |
| | Lnk state <------> Lnk state <--(3)--> I | | Flood path
| | | | | | T <----(5,7,8)-->
| | Span tree <------> Span tree <--(4)--> C | |
| +--^--------+ | | | H | |
| | | Discovery <--(2)--> | |
| | +------------+ | M | |
| | | E | |
| +------^--+ +--------+ | S | |
| | CONNECT >---------+--> FILTER | | S | |
| +--^------+ | +--------+ | A | | specific
| | | | G | | netwrk lnks
| | +--------^-+ +-------+ | E <----(2,3,4)-->
| +-------< POLICY | | FLOOD >--(7)--> | |
| +------^---+ +-^-----+ | P | |
| | | | R | |
| +-----------+ +-^-----------V-+ | O | |
| | DIRECTORY <----> RESOLVE <------(5)--> T | |
| +-----^-----+ +---^-----------+ | O | |
| | | | C | |
| | +---------^-----------+ | O | |
| +----< Base Services | | L | |
| +-----^---------------+ +-----+ |
+------------------|-----------------------------------+
Switch CPU |
| Host control port
+-----O----------------+
| ^ no cnx |
Layer 2 | | |
---------->O-----+--------------->O----------->
SA/DA pr | known cnx |
+----------------------+
Switch hardware
Figure 1: SFVLAN Architectural Overview
- Eight call processing service centers that provide the essential
services required to process call connections. The call
processing service centers are described in Section 4.
- A Call Tap module that supports the monitoring of call
connections. The Call Tap module is described in Section 5.
- The InterSwitch Message Protocol (ISMP) that provides a consistent
method of encapsulating and transmitting control messages
exchanged between SFVLAN switches. (Note that ISMP is not a
discrete software module. Instead, its functionality is
distributed among those service centers and software modules that
need to communicate with other switches in the fabric.) The
Interswitch Message Protocol and the formats of the individual
interswitch messages are described in Section 6.
3. Base Services
The SFVLAN base services act as the interface between the switch
hardware and the SFVLAN service centers running on the switch CPU.
This relationship is shown in Figure 2. This figure is a replication
of the bottom portion of Figure 1.
| Directory Resolve |
| ^ ^ |
| | | |
| | +---------^-----------+ |
| +----< Base Services | |
| +-----^---------------+ |
+-------------------|--------------------------+
Switch CPU |
| Host control port
+-----O----------------+
| ^ no cnx |
Layer 2 | | |
---------->O-----+--------------->O----------->
SA/DA pr | known cnx |
+----------------------+
Switch hardware
Figure 2: Base Services
During normal operation of the switch, data packets arriving at any one of the local switch ports are examined in the switch hardware. If the packet's source and destination address (SA/DA) pair match a known connection, the hardware simply forwards the packet out the outport specified by the connection. If the SA/DA pair do not match any known connection, the hardware diverts the packet to the host control port where it is picked up by the SFVLAN base services. The base services generate a structure known as a state box that tracks the progress of the call connection request as the request moves through the call processing service centers. After creating the call's state box, the base services check to determine if the call is a duplicate of a call already being processed. If not, a request is issued to the Directory Service Center (Section 4.1) to add the call's source address to the local Node and Alias Tables. The base services then hand the call off to the Resolve Service Center (Section 4.3) for further processing.4. Call Processing
Call connection processing is handled by a set of eight service centers, each with one or more servers. The servers within a service center are called in a particular sequence. Each server records the results of its processing in the call connection request state box and passes the state box to the next server in the sequence. In the sections that follow, servers are listed in the order in which they are called.4.1 Directory Service Center
The Directory Service Center is responsible for cataloging the MAC addresses and alias information for both local and remote endstations. The information is stored in two tables -- the Node Table and the Alias Table. - The Node Table contains the MAC addresses of endstations attached to the local switch. It also contains a cache of remote endstations detected by the Resolve Service Center (Section 4.3). Every entry in the Node Table has one or more corresponding entries in the Alias Table.
- The Alias Table contains protocol alias information for each
endstation. An endstation alias can be a network address (such
as an IP or IPX address), a VLAN identifier, or any other
protocol identifier. Since every endstation is a member of at
least one VLAN (the default VLAN for the port), there is always
at least one entry in the Alias Table for each entry in the
Node Table.
Note:
The Node and Alias Tables must remain synchronized.
That is, when an endstation's final alias is removed
from the Alias Table, the endstation entry is removed
from the Node Table.
Note that the total collection of all Node Tables and Alias Tables
across all switches is known as the "virtual" directory of the
switch fabric. The virtual directory contains address mappings of
all known endstations in the fabric.
4.1.1 Local Add Server
The Directory Local Add server adds entries to the local Node or
Alias Tables. It is called by the base services (Section 3) to
add a local endstation and by the Interswitch Resolve (Section
4.3.4) server to add an endstation discovered on a remote switch.
4.1.2 Inverse Resolve Server
The Directory Inverse Resolve server is invoked when a new
endstation has been discovered on the local switch (that is, when
the Local Add server was successful in adding the endstation).
The server provides two functions:
- It populates the Node and Alias Tables with local entries
during switch initialization.
- It processes a new endstation discovered after the fabric
topology has converged to a stable state.
In both instances, the processing is identical.
When a new endstation is detected on one of the switch's local ports, the Inverse Resolve server sends an Interswitch New User request message (Section 6.5) over the switch flood path to all other switches in the fabric. The purpose of the Interswitch New User request is two-fold: - It informs the other switches of the new endstation address. Any entries for that endstation in the local databases of other switches should be dealt with appropriately. - It requests information about any static VLAN(s) to which the endstation has been assigned. When a switch receives an Interswitch New User request message from one of its upstream neighbors, it first forwards the message to all its downstream neighbors. No actual processing or VLAN resolution is attempted until the message reaches the end of the switch flood path and begins its trip back along the return path. This ensures that all switches in the fabric receive notification of the new user and have synchronized their databases. If a switch receives an Interswitch New User request message but has no downstream neighbors, it does the following: - If the endstation was previously connected to one of the switch's local ports, the switch formulates an Interswitch New User Response message by loading the VLAN identifier(s) of the static VLAN(s) to which the endstation was assigned, along with its own MAC address. (VLAN identifiers are stored in Tag/Length/Value (TLV) format. See Section 2.3.) The switch then sets the message status field to NewUserAck, and returns the message to its upstream (requesting) neighbor. Otherwise, the switch sets the status field to NewUserUnknown and returns the message to its upstream neighbor. - The switch then deletes the endstation from its local database, as well as any entries associated with the endstation in its connection table. When a switch forwards an Interswitch New User request message to its downstream neighbors, it keeps track of the number of requests it has sent out and does not respond back to its upstream neighbor until all requests have been responded to.
- As each response is received, the switch checks the status
field of the message. If the status is NewUserAck, the switch
retains the information in that response. When all requests
have been responded to, the switch returns the NewUserAck
response to its upstream neighbor.
- If all the Interswitch New User Request messages have been
responded to with a status of NewUserUnknown, the switch checks
to see if the endstation was previously connected to one of its
local ports. If so, the switch formulates an Interswitch New
User Response message by loading the VLAN identifier(s) of the
static VLAN(s) to which the endstation was assigned, along with
its own MAC address. The switch then sets the message status
field to NewUserAck, and returns the message to its upstream
(requesting) neighbor.
Otherwise, the switch sets the status field to NewUserUnknown
and returns the message to its upstream neighbor.
- The switch then deletes the endstation from its local database,
as well as any entries associated with the endstation in its
connection table.
When the originating switch has received responses to all the
Interswitch New User Request messages it has sent, it does the
following:
- If it has received a response message with a status of
NewUserAck, it loads the new VLAN information into its local
database.
- If all responses have been received with a status of
NewUserUnknown, the originating switch assumes that the
endstation was not previously connected anywhere in the network
and assigns it to a VLAN according to the VLAN membership rules
and order of precedence.
If any Interswitch New User Request message has not been responded
to within a certain predetermined time (currently 5 seconds), the
originating switch recalculates the switch flood path and resends
the Interswitch New User Request message.
4.1.3 Local Delete Server
The Directory Local Delete server removes entries (both local and remote) from the local Node and Alias Tables. It is invoked when an endstation, previously known to be attached to one switch, has been moved and discovered on another switch. Note also that remote entries are cached and are purged from the tables on a first-in/first-out basis as space is needed in the cache.4.2 Topology Service Center
The Topology Service Center is responsible for maintaining three databases relating to the topology of the switch fabric: - The topology table of SFVLAN switches that are physical neighbors to the local switch. - The spanning tree that defines the loop-free switch flood path used for transmitting undirected interswitch messages. - The directed graph that is used to calculate the best path(s) for call connections.4.2.1 Neighbor Discovery Server
The Topology Neighbor Discovery server uses Interswitch Keepalive messages to detect the switch's neighbors and establish the topology of the switching fabric. Interswitch Keepalive messages are exchanged in accordance with Cabletron's VlanHello protocol, described in detail in [IDhello].4.2.2 Spanning Tree Server
The Topology Spanning Tree server is invoked by the Topology Neighbor Discovery server when a neighboring SFVLAN switch is either discovered or lost -- that is, when the operational status of a network link changes. The Spanning Tree server exchanges interswitch messages with neighboring SFVLAN switches to calculate the switch flood path over which undirected interswitch messages are sent. There are two parts to this process: - Creating and maintaining the spanning tree - Remote blocking
4.2.2.1 Creating and Maintaining the Spanning Tree
In a network with redundant network links, a packet traveling between switches can potentially be caught in an infinite loop -- an intolerable situation in a networking environment. However, it is possible to reduce a network topology to a single configuration (known as a spanning tree) such that there is, at most, one path between any two switches. Within the SFVLAN product, the spanning tree is created and maintained using the Spanning Tree Algorithm defined by the IEEE 802.1d standard. Note: A detailed discussion of this algorithm is beyond the scope of this document. See [IEEE] for more information. To implement the Spanning Tree Algorithm, SFVLAN switches exchange Interswitch BPDU messages (Section 6.2) containing encapsulated IEEE-compliant 802.2 Bridge Protocol Data Units (BPDUs). There are two types of BPDUs: - Configuration (CFG) BPDUs are exchanged during the switch discovery process, following the receipt of an Interswitch Keepalive message. They are used to create the initial the spanning tree. - Topology Change Notification (TCN) BPDUs are exchanged when changes in the network topology are detected. They are used to redefine the spanning tree to reflect the current topology. See [IEEE] for detailed descriptions of these BPDUs.4.2.2.2 Remote Blocking
After the spanning tree has been computed, each network port on an SFVLAN switch will be in one of two states: - Forwarding. A port in the Forwarding state will be used to transmit all ISMP messages.
- Blocking. A port in the Blocking state will not be used to
forward undirected ISMP messages. Blocking the rebroadcast of
these messages on selected ports prevents message duplication
arising from multiple paths that exist in the network topology.
Note that all other types of ISMP message will be transmitted.
Note:
The IEEE 802.1d standard specifies other port states used
during the initial creation of the spanning tree. These states
are not relevant to the discussion here.
Note that although a port in the Blocking state will not forward
undirected ISMP messages, it may still receive them. Any such
message received will ultimately be discarded, but at the cost of CPU
time necessary to process the packet.
To prevent the transmission of undirected messages to a port, the
port's owner switch can set remote blocking on the link by sending an
Interswitch Remote Blocking message (Section 6.3) out over the port.
This notifies the switch on the other end of the link that undirected
messages should not be sent over the link, regardless of the state of
the sending port.
Each SFVLAN switch sends an Interswitch Remote Blocking message out
over all its blocked network ports every 5 seconds. A flag within
the message indicates whether remote blocking should be turned on or
off over the link.
4.2.3 Link State Server
The Topology Link State server is invoked by any process that detects
a change in the state of the network links of the local switch.
These changes include (but are not limited to) changes in operational
or administrative status of the link, path "cost" or bandwidth.
The Link State server runs Cabletron's Virtual LAN Link State (VLS)
protocol which exchanges interswitch messages with neighboring SFVLAN
switches to calculate the set of best paths between the local switch
and all other switches in the fabric. (The VLS protocol is described
in detail in [IDvlsp].)
The Link State server also notifies the Connect Service Center
(Section 4.5) of any remote links that have failed, thereby
necessitating potential tear-down of current connections.
4.3 Resolve Service Center
The Resolve Service Center is responsible for resolving the destination address of broadcast data packets (such as an IP ARP packet) to a unicast MAC address to be used in mapping the call connection. To do this, the Resolve Service Center attempts to resolve such broadcast packets directly at the access port of the ingress switch. Address resolution is accomplished as follows: 1) First, an attempt is made to resolve the address from the switch's local databases by calling the following servers: - The Table server attempts to resolve the address from the Resolve Table (Section 4.3.1). - Next, the Local server attempts to resolve the address from the Node and Alias Tables (Section 4.3.2). - If the address is not found in these tables but is an IP address, the Resolve Subnet server (Section 4.3.3) is also called. 2) If the address cannot be resolved locally, the Interswitch Resolve server (Section 4.3.4) is called to access the "virtual directory" by sending an Interswitch Resolve request message out over the switch flood path. 3) If the address cannot be resolved either locally or via an Interswitch Resolve message -- that is, the destination endstation is unknown to any switch, perhaps because it has never transmitted a packet to its switch -- the following steps are taken: - The Unresolvable server (Section 4.3.5) is called to record the unresolved packet. - The Block server (Section 4.3.6) is called to determine whether the address should be added to the Block Table. - The Flood Service Center (Section 4.8) is called to broadcast the packet to other SFVLAN switches using a tag-based flooding mechanism.
4.3.1 Table Server
The Resolve Table server maintains the Resolve Table which contains a collection of addresses that might not be resolvable in the normal fashion. This table typically contains such things as the addresses of "quiet" devices that do not send data packets or special mappings of IP addresses behind a router. Entries can be added to or deleted from the Resolve Table via an external management application.4.3.2 Local Server
The Resolve Local server checks the Node and Alias Tables maintained by the Directory Service Center (Section 4.1) to determine if it can resolve the address.4.3.3 Subnet Server
If the address to be resolved is an IP address but cannot be resolved via the standard processing described above, the Resolve Subnet server applies the subnet mask to the IP address and then does a lookup in the Resolve Table.4.3.4 Interswitch Resolve Server
If the address cannot be resolved locally, the Interswitch Resolve server accesses the "virtual directory" by sending an Interswitch Resolve request message (Section 6.4) out over the switch flood path. The Interswitch Resolve request message contains the destination address as it was received within the packet, along with a list of requested addressing information. When a switch receives an Interswitch Resolve request message from one of its upstream neighbors, it checks to see if the destination endstation is connected to one of its local access ports. If so, it formulates an Interswitch Resolve response message by filling in the requested address information, along with its own MAC address. It then sets the message status field to ResolveAck, and returns the message to its upstream (requesting) neighbor. If the receiving switch cannot resolve the address, it forwards the Interswitch Resolve request message to its downstream neighbors. If the switch has no downstream neighbors, it sets the message status field to Unknown, and returns the message to its upstream (requesting) neighbor.
When a switch forwards an Interswitch Resolve request message to its downstream neighbors, it keeps track of the number of requests it has sent out and received back. It will only respond back to its upstream (requesting) neighbor when one of the following conditions occurs: - It receives any response with a status of ResolveAck - All downstream neighbors have responded with a status of Unknown Any Interswitch Resolve request message that is not responded to within a certain predetermined time (currently 5 seconds) is assumed to have a response status of Unknown. When the Interswitch Resolve server receives a successful Interswitch Resolve response message, it records the resolved address information in the remote cache of its local directory for use in resolving later packets for the same endstation. Note that this process results in each switch building its own unique copy of the virtual directory containing only the endstation addresses in which it is interested.4.3.5 Unresolvable Server
The Unresolvable server is called when a packet destination address cannot be resolved. The server records the packet in a table that can then be examined to determine which endstations are generating unresolvable traffic. Also, if a particular destination is repeatedly seen to be unresolvable, the server calls the Block server (Section 4.3.6) to determine whether the address should be blocked.4.3.6 Block Server
The Resolve Block server is called when a particular destination has been repeatedly seen to be unresolvable. This typically happens when, unknown to the packet source, the destination endstation is either not currently available or no longer exists. If the Block server determines that the unresolved address has exceeded a configurable request threshold, the address is added to the server's Block Table. Interswitch Resolve request messages for addresses listed in the Block Table are sent less frequently, thereby reducing the amount of Interswitch Resolve traffic throughout the fabric.
If an address listed in the Block Table is later successfully resolved by and Interswitch Resolve request message, the address is removed from the table.4.4 Policy Service Center
Once the destination address of the call packet has been resolved, the Policy Service Center is called to determine the validity of the requested call connection based on the VLAN policy of the source and destination VLANs.4.4.1 Unicast Rules Server
The Policy Unicast Rules server recognizes two VLAN policy values: Open or Secure. The default policy for all VLANs is Open. The policy value is used as follows when determining the validity of a requested call connection: - If the VLAN policy of either the source or destination cannot be determined, the Filter Service Center is called to establish a filter (i.e., blocked) for the SA/DA pair. - If the source and destination endstations belong to the same VLAN, then the connection is permitted regardless of the VLAN policy. - If the source and destination endstations belong to different VLANs, but both VLANs are running with an Open policy, then the connection is permitted, providing cut-through switching between different VLAN(s). - If the source and destination endstations belong to different VLANs and one or both of the VLANs are running with a Secure policy, then the Flood Service Center (Section 4.8) is called to broadcast the packet to other SFVLAN switches having ports or endstations that belong to the same VLAN as the packet source. Note that if any of the VLANs to which the source or destination belong has a Secure policy, then the policy used in the above algorithm is Secure.
4.5 Connect Service Center
Once the Policy Service Center (Section 4.4) has determined that a requested call connection is valid, the Connect Service Center is called to set up the connection. Note that connectivity between two endstations within the fabric is established on a switch-by-switch basis as the call progresses through the fabric toward its destination. No synchronization is needed between switches to establish an end-to-end connection. The Connect Service Center maintains a Connection Table containing information for all connections currently active on the switch's local ports. Connections are removed from the Connection Table when one of the endstations is moved to a new switch (Section 4.1.2) or when the Topology Link State server (Section 4.2.3) notifies the Connect Service Center that a network link has failed. Otherwise, connections are not automatically aged out or removed from the Connection Table until a certain percentage threshold (HiMark) of table capacity is reached and resources are needed. At that point, some number of connections (typically 100) are aged out and removed at one time.4.5.1 Local Server
If the destination endstation resides on the local switch, the Connect Local server establishes a connection between the source and destination ports. Note that if the source and destination both reside on the same physical port, a filter connection is established by calling the Filter Service Center (Section 4.6).4.5.2 Link State Server
The Connect Link State server is called if the destination endstation of the proposed connection does not reside on the local switch. The server executes a call to the Path Link State server (Section 4.7.1) which returns up to three "best" paths of equal cost from the local switch to the destination switch. If more than one path is returned, the server chooses a path that provides the best load balancing of user traffic across the fabric.
4.5.3 Directory Server
The Connect Directory server is called if the Connect Link State server is unable to provide a path for some reason. The server examines the local directory to determine on which switch the destination endstation resides. If the port of access to the destination switch is known, then a connection is established using that port as the outport of the connection.4.6 Filter Service Center
The Filter Service Center is responsible for establishing filtered connections. This service center is called by the Connect Local server (Section 4.5.1) if the source and destination endstations reside on the same physical port, and by the Policy Service Center (Section 4.4) if the VLAN of either the source or destination is indeterminate. A filter connection is programmed in the switch hardware with no specified outport. That is, the connection is programmed to discard any traffic for that SA/DA pair.4.7 Path Service Center
The Path Service Center is responsible for determining the path from a source to a destination.4.7.1 Link State Server
The Path Link State server is called by the Connect Link State server (Section 4.5.2) to return up to three best paths of equal cost between a source and destination pair of endstations. These best paths are calculated by the Topology Link State server (Section 4.2.3). The Path Link State server is also called by the Connect Service Center to return a complete source-to-destination path consisting of a list of individual switch port names. A switch port name consists of the switch base MAC address and a port instance relative to the switch.
4.7.2 Spanning Tree Server
The Path Spanning Tree server is called by any server needing to forward an undirected message out over the switch flood path. The server returns a port mask indicating which local ports are currently enabled as outports of the switch flood path. The switch flood path is calculated by the Topology Spanning Tree server (Section 4.2.2).4.8 Flood Service Center
If the Resolve Service Center (Section 4.3) is unable to resolve the destination address of a packet, it invokes the Flood Service Center to broadcast the unresolved packet.4.8.1 Tag-Based Flood Server
The Tag-Based Flood server encapsulates the unresolved packet into an Interswitch Tag-Based Flood message (Section 6.6), along with a list of Virtual LAN identifiers specifying those VLANs to which the source endstation belongs. The message is then sent out over the switch flood path to all other switches in the fabric. When a switch receives an Interswitch Tag-Based Flood message, it examines the encapsulated header to determine the VLAN(s) to which the packet should be sent. If any of the switch's local access ports belong to one or more of the specified VLANs, the switch strips off the tag-based header and forwards the original packet out the appropriate access port(s). The switch also forwards the entire encapsulated packet along the switch flood path to its downstream neighboring switches, if any.
(page 27 continued on part 2)