RFCs in HTML Format


RFC 1795

             Data Link Switching: Switch-to-Switch Protocol
       AIW DLSw RIG: DLSw Closed Pages, DLSw Standard Version 1.0


Wells & Bartky                                                  [Page 1]

RFC 1795 Data Link Switching April 1995 1. Introduction Data Link Switching (DLSw) is a forwarding mechanism for the IBM SNA (Systems Network Architecture) and IBM NetBIOS (Network Basic Input Output Services) protocols. This memo documents the Switch-to-Switch Protocol (SSP) that is used between Data Link Switches. This protocol does not provide full routing, but instead provides switching at the SNA Data Link layer (i.e., layer 2 in the SNA architecture) and encapsulation in TCP/IP for transport over the Internet. This RFC documents the frame formats and protocols for multiplexing data between Data Link Switches. The initial implementation of SSP uses TCP as the reliable transport between Data Link Switches. However, other transport connections such as OSI TP4 could be used in the future. A Data Link Switch (abbreviated also as DLSw in this document) can support SNA (Physical Unit (PU) 2, PU 2.1 and PU 4) systems and optionally NetBIOS systems attached to IEEE 802.2 compliant Local Area Networks, as well as SNA (PU 2 (primary or secondary) and PU2.1) systems attached to IBM Synchronous Data Link Control (SDLC) links. For the latter case, the SDLC attached systems are provided with a LAN appearance within the Data Link Switch (each SDLC PU is presented to the SSP protocol as a unique MAC/SAP address pair). For the Token-Ring LAN attached systems, the Data Link Switch appears as a source-routing bridge. Token-Ring Remote systems that are accessed through the Data Link Switch appear as systems attached to an adjacent ring. This ring is a virtual ring that is manifested within each Data Link Switch. 1.1 Backwards Compatibility with RFC 1434 This document defines significant changes to RFC 1434 and does not state details on how to interoperate with RFC 1434 or "enhanced" implementations (e.g., those that added enter and exit busy flow control). It is up to the implementer to refer to RFC 1434 and/or any other vendor's documentation in order to interoperate with a given vendor's implementation, if interoperability with pre-AIW DLSw RIG standards is desired. Wells & Bartky [Page 2]
RFC 1795 Data Link Switching April 1995 2. Overview Data Link Switching was developed to provide support for SNA and NetBIOS in multi-protocol routers. Since SNA and NetBIOS are basically connection oriented protocols, the Data Link Control procedure that they use on the LAN is IEEE 802.2 Logical Link Control (LLC) Type 2. Data Link Switching also accommodates SNA protocols over WAN (Wide Area Network) links via the SDLC protocol. IEEE 802.2 LLC Type 2 was designed with the assumption that the network transit delay would be predictable (i.e., a local LAN). Therefore the LLC Type 2 elements of procedure use a fixed timer for detecting lost frames. When remote bridging is used over wide area lines (especially at lower speeds), the network delay is larger and it can vary greatly based upon congestion. When the delay exceeds the time-out value LLC Type 2 attempts to retransmit. If the frame is not actually lost, only delayed, it is possible for the LLC Type 2 procedures to become confused. And as a result, the link may be eventually taken down if the delay exceeds the T1 timer times N2 retry count. Given the use of LLC Type 2 services, Data Link Switching addresses the following bridging problems: DLC Time-outs DLC Acknowledgments over the WAN Flow and Congestion Control Broadcast Control of Search Packets Source-Route Bridging Hop Count Limits NetBIOS also makes extensive use of datagram services that use connectionless LLC Type 1 service. In this case, Data Link Switching addresses the last two problems in the above list. The principal difference between Data Link Switching and bridging is that for connection-oriented data DLSw terminates the Data Link Control whereas bridging does not. The following figure illustrates this difference based upon two end systems operating with LLC Type 2 services. Wells & Bartky [Page 3]
RFC 1795 Data Link Switching April 1995 Bridging -------- Bridge Bridge +------+ +----+ +----+ +------+ | End | +-----+ | +-----/ | | +-----+ | End | |System+-+ LAN +-+ | /------+ +-+ LAN +-+System| | | +-----+ | | TCP/IP | | +-----+ | | +------+ +----+ +----+ +------+ Info-----------------------------------------------> <-----------------------------------------------RR Data Link Switching ------------------- +------+ +----+ +----+ +------+ | End | +-----+ | +-----/ | | +-----+ | End | |System+-+ LAN +-+DLSw| /------+DLSw+-+ LAN +-+System| | | +-----+ | | TCP/IP | | +-----+ | | +------+ +----+ +----+ +------+ Info---------------> -------------> Info <---------------RR ------------> <------------RR In traditional bridging, the Data Link Control is end-to-end. Data Link Switching terminates the LLC Type 2 connection at the switch. This means that the LLC Type 2 connections do not cross the wide area network. The DLSw multiplexes LLC connections onto a TCP connection to another DLSw. Therefore, the LLC connections at each end are totally independent of each other. It is the responsibility of the Data Link Switch to deliver frames that it has received from a LLC connection to the other end. TCP is used between the Data Link Switches to guarantee delivery of frames. As a result of this design, LLC time-outs are limited to the local LAN (i.e., they do not traverse the wide area). Also, the LLC Type 2 acknowledgments (RR's) do not traverse the WAN, thereby reducing traffic across the wide area links. For SDLC links, polling and poll response occurs locally, not over the WAN. Broadcast of search frames is controlled by the Data Link Switches once the location of a target system is discovered. Finally, the switches can now apply back pressure to the end systems to provide flow and congestion control. Only one copy of an Link Protocol Data Unit (LPDU) is sent between Data Link Switches in SSP messages (XIDFRAME and INFOFRAME). Retries of the LPDU are absorbed by Data Link Switch that receives it. The Wells & Bartky [Page 4]
RFC 1795 Data Link Switching April 1995 Data Link Switch that transmits the LPDU received in an SSP message to a local DLC, will perform retries in a manner appropriate for the local DLC. This may involve running a reply timer and maintaining a poll retry count. The length of the timer and the number of retries is an implementation choice based on user configuration parameters and the DLC type. Data Link Switching uses LAN addressing to set up connections between SNA systems. SDLC attached devices are defined with MAC and SAP addresses to enable them to communicate with LAN attached devices. For NetBIOS systems, Data Link Switching uses the NetBIOS name to forward datagrams and to set up connections for NetBIOS sessions. For LLC type 2 connection establishment, SNA systems send TEST (or in some cases, XID) frames to the null (0x00) SAP. NetBIOS systems have an address resolution procedure, based upon the Name Query and Name Recognized frames, that is used to establish an end-to-end circuit. Since Data Link Switching may be implemented in multi-protocol routers, there may be situations where both bridging and switching are enabled. SNA frames can be identified by their link SAP. Typical SAP values for SNA are 0x04, 0x08, and 0x0C. NetBIOS always uses a link SAP value of 0xF0. Wells & Bartky [Page 5]
RFC 1795 Data Link Switching April 1995 3. Transport Connection Data Link Switches can be in used in pairs or by themselves. A Single DLSw internally switches one data link to another without using TCP (DLC(1) to DLC(2) in the figure below). This RFC does not go into details on how to implement this feature and it is not a requirement to support this RFC. A paired DLSw multiplexes data links over a reliable transport using a Switch-to-Switch Protocol (SSP). +-------------------------------------------+Switch-to-Switch | DLC Interfaces | Protocol (SSP) |+-----------+ DLC Request +-----------+ | || Data |<---------------| | |Send SSP Frame || Link | DLC Indication | | |--------------> || Control 1 |--------------->| | | |+-----------+ | Data Link | | |+-----------+ DLC Request | Switch | | || Data |<-------------- | | |Rec. SSP Frame || Link | DLC Indication | | |<------------- || Control 2 | -------------->| | | |+-----------+ +-----------+ | | Multi-Protocol Router | +-------------------------------------------+ Before Data Link Switching can occur between two routers, they must establish two TCP connections between them. Each Data Link Switch will maintain a list of DLSw capable routers and their status (active/inactive). After the TCP connection is established, SSP messages are exchanged to establish the capabilities of the two Data Link Switches. Once the exchange is complete, the DLSw will employ SSP control messages to establish end-to-end circuits over the transport connection. Within the transport connection, DLSw SSP messages are exchanged. The message formats and types for these SSP messages are documented in the following sections. The default parameters associated with the TCP connections between Data Link Switches are as follows: Socket Family AF_INET (Internet protocols) Socket Type SOCK_STREAM (stream socket) Read Port Number 2065 Write Port Number 2067 Wells & Bartky [Page 6]
RFC 1795 Data Link Switching April 1995 Two or more Data Link Switches may be attached to the same LAN, consisting of a number of token-ring segments interconnected by source-routing bridges. In this case, a TCP connection is not defined between bridges attached to the same LAN. This will allow using systems to select one of the possible Data Link Switches in a similar manner to the selection of a bridge path through a source- routed bridged network. The virtual ring segment in each Data Link Switch attached to a common LAN must be configured with the same ring number. This will prevent LAN frames sent by one Data Link Switch from being propagated through the other Data Link Switches. Wells & Bartky [Page 7]
RFC 1795 Data Link Switching April 1995 3.1 SSP Frame Formats The following diagrams show the two message header formats exchanged between Data Link Switches, Control and Information. The Control message header is used for all messages except Information Frames (INFOFRAME) and Independent Flow Control Messages (IFCM), which are sent in Information header format. The INFOFRAME, KEEPALIVE and IFCM message headers are 16 bytes long, and the control message header is 72 bytes long. The fields in the first sixteen bytes of all message headers are the same. CONTROL MESSAGES (72 Bytes) (zero based offsets below shown in decimal (xx) ) +-----------------------------+-----------------------------+ | (00) Version Number | (01) Header Length (= 72) | +-----------------------------+-----------------------------+ | (02) Message Length | +-----------------------------+-----------------------------+ | (04) Remote Data Link Correlator | +- - - - - - - - - - - - - - -+- - - - - - - - - - - - - - -+ | | +-----------------------------+-----------------------------+ | (08) Remote DLC Port ID | +- - - - - - - - - - - - - - -+- - - - - - - - - - - - - - -+ | | +-----------------------------+-----------------------------+ | (12) Reserved Field | +-----------------------------+-----------------------------+ | (14) Message Type | (15) Flow Control Byte | +-----------------------------+-----------------------------+ | (16) Protocol ID | (17) Header Number | +-----------------------------+-----------------------------+ | (18) Reserved | +-----------------------------+-----------------------------+ | (20) Largest Frame Size | (21) SSP Flags | +-----------------------------+-----------------------------+ | (22) Circuit Priority | (23) Message Type (see note)| +-----------------------------+-----------------------------+ | (24) Target MAC Address (non-canonical format) | +- - - - - - - - - - - - - - -+- - - - - - - - - - - - - - -| | | +- - - - - - - - - - - - - - -+- - - - - - - - - - - - - - -+ | | +-----------------------------+-----------------------------+ | (30) Origin MAC Address (non-canonical format) | +- - - - - - - - - - - - - - -+- - - - - - - - - - - - - - -| Wells & Bartky [Page 8]
RFC 1795 Data Link Switching April 1995 | | +- - - - - - - - - - - - - - -+- - - - - - - - - - - - - - -+ | . . | +-----------------------------+-----------------------------+ | (36) Origin Link SAP | (37) Target Link SAP | +-----------------------------+-----------------------------+ | (38) Frame Direction | (39) Reserved | +-----------------------------+-----------------------------+ | (40) Reserved | +-----------------------------+-----------------------------+ | (42) DLC Header Length | +-----------------------------+-----------------------------+ | (44) Origin DLC Port ID | +- - - - - - - - - - - - - - -+- - - - - - - - - - - - - - -+ | | +-----------------------------+-----------------------------+ | (48) Origin Data Link Correlator | +- - - - - - - - - - - - - - -+- - - - - - - - - - - - - - -+ | | +-----------------------------+-----------------------------+ | (52) Origin Transport ID | +- - - - - - - - - - - - - - -+- - - - - - - - - - - - - - -+ | | +-----------------------------+-----------------------------+ | (56) Target DLC Port ID | +- - - - - - - - - - - - - - -+- - - - - - - - - - - - - - -+ | | +-----------------------------+-----------------------------+ | (60) Target Data Link Correlator | +- - - - - - - - - - - - - - -+- - - - - - - - - - - - - - -+ | | +-----------------------------+-----------------------------+ | (64) Target Transport ID | +- - - - - - - - - - - - - - -+- - - - - - - - - - - - - - -+ | | +-----------------------------+-----------------------------+ | (68) Reserved Field | +-----------------------------+-----------------------------+ | (70) Reserved Field | +-----------------------------+-----------------------------+ (Even Byte) (Odd Byte) Wells & Bartky [Page 9]
RFC 1795 Data Link Switching April 1995 INFORMATION MESSAGE (16 Bytes) +-----------------------------+-----------------------------+ | (00) Version Number | (01) Header Length (= 16) | +-----------------------------+-----------------------------+ | (02) Message Length | +-----------------------------+-----------------------------+ | (04) Remote Data Link Correlator | +- - - - - - - - - - - - - - -+- - - - - - - - - - - - - - -+ | | +-----------------------------+-----------------------------+ | (08) Remote DLC Port ID | +- - - - - - - - - - - - - - -+- - - - - - - - - - - - - - -+ | | +-----------------------------+-----------------------------+ | (12) Reserved Field | +-----------------------------+-----------------------------+ | (14) Message Type | (15) Flow Control Byte | +-----------------------------+-----------------------------+ (Even Byte) (Odd Byte) The first sixteen bytes of control and information message headers contain identical fields. A brief description of some of the fields in an SSP message are shown below (if not defined below, the fields and/or their values are described in subsequent sections). The Version Number field (offset 0) is set to 0x31 (ASCII '1'), indicating a decimal value of 49. This is used to indicate DLSw version 1. The Header Length field (offset 1) is 0x48 for control messages, indicating a decimal value of 72 bytes, and 0x10 for information and Independent Flow Control messages, indicating a decimal value of 16 bytes. The Message Length field (offset 2) defines the number of bytes within the data field following the header. The Flow Control Byte field (offset 15) is described in section 8. The Header Number field (offset 17) is 0x01, indicating a value of one. The Circuit Priority field (offset 22) is described in section 4. The Frame Direction field (offset 38) is set to 0x01 for frames sent from the origin DLSw to the target DLSw, and is set to 0x02 for frames sent from the target DLSw to the origin DLSw. Wells & Bartky [Page 10]
RFC 1795 Data Link Switching April 1995 Note: The Remote Data Link Correlator and Remote DLC Port ID are set equal to the Target Data Link Correlator and Target DLC Port ID if the Frame Direction field is set to 0x01, and are set equal to the Origin Data Link Correlator and Origin DLC Port ID if the Direction Field is set to 0x02. The Protocol ID field is set to 0x42, indicating a decimal value of 66 The DLC Header Length is set to zero for SNA and is set to 0x23 for NetBIOS datagrams, indicating a length of 35 bytes. This includes the Access Control (AC) field, the Frame Control (FC) field, Destination MAC Address (DA), the Source MAC Address (SA), the Routing Information (RI) field (padded to 18 bytes), the Destination link SAP (DSAP), the Source link SAP (SSAP), and the LLC control field (UI). NOTE: The values for the Message Type field are defined in section 3.5. Note that this value is specified in two different fields (offset 14 and 23 decimal) of the control message header. Only the first field is to be used when parsing a received SSP message. The second field is to be ignored by new implementations on reception. The second field was left in for backwards compatibility with RFC 1434 implementations and this field may be used in future versions if needed. The SSP Flags field contains additional information related to the SSP message. The flags are defined as follows (bit 7 being the most significant bit and bit 0 the least significant bit of the octet): Bit(s) 76543210 Name Meaning --------- ----- ------- x....... SSPex 1 = explorer message (CANUREACH and ICANREACH) Reserved fields are set to zero upon transmission and should be ignored upon receipt. 3.2 Address Parameters A data link is defined as a logical association between the two end stations using Data Link Switching. It is identified by a Data Link ID (14 bytes) consisting of the pair of attachment addresses associated with each end system. Each attachment address is represented by the concatenation of the MAC address (6 bytes) and the LLC address (1 byte). Each attachment address is classified as either "Target" in the context of the Destination MAC/SAP addresses of an explorer frame sent in the first frame used to establish a Wells & Bartky [Page 11]
RFC 1795 Data Link Switching April 1995 circuit, or "Origin" in the context of the Source MAC/SAP addresses. All MAC addresses are expressed in non-canonical (Token-Ring) format. DATA LINK ID (14 Bytes @ Control message offset 24 decimal) +-----------------------------+-----------------------------+ | Target MAC Address | +- - - - - - - - - - - - - - -+- - - - - - - - - - - - - - -+ | | +- - - - - - - - - - - - - - -+- - - - - - - - - - - - - - -+ | | +-----------------------------+-----------------------------+ | Origin MAC Address | +- - - - - - - - - - - - - - -+- - - - - - - - - - - - - - -+ | | +- - - - - - - - - - - - - - -+- - - - - - - - - - - - - - -+ | | +-----------------------------+-----------------------------+ | Origin Link SAP | Target Link SAP | +-----------------------------+-----------------------------+ An end-to-end circuit is identified by a pair of Circuit ID's. A Circuit ID is a 64 bit number that identifies the DLC circuit within a single DLSw. It consists of a DLC Port ID (4 bytes), and a Data Link Correlator (4 bytes). The Circuit ID must be unique in a single DLSw and is assigned locally. The pair of Circuit ID's along with the Data Link IDs, uniquely identify a single end-to-end circuit. Each DLSw must keep a table of these Circuit ID pairs, one for the local end of the circuit and the other for the remote end of the circuit. In order to identify which Data Link Switch originated the establishment of a circuit, the terms, "Origin" DLSw and "Target" DLSw, will be employed in this document. CIRCUIT ID (8 Bytes) +-----------------------------+-----------------------------+ | DLC Port ID | +- - - - - - - - - - - - - - -+- - - - - - - - - - - - - - -+ | | +-----------------------------+-----------------------------+ | Data Link Correlator | +- - - - - - - - - - - - - - -+- - - - - - - - - - - - - - -+ | | +-----------------------------+-----------------------------+ The Origin Transport ID and the Target Transport ID fields in the message header are used to identify the individual TCP/IP port on a Data Link Switch. The values have only local significance. However, each Data Link Switch is required to reflect the values contained in Wells & Bartky [Page 12]
RFC 1795 Data Link Switching April 1995 these two fields, along with the associated values for DLC Port ID and the Data Link Correlator, when returning a message to the other Data Link Switch. The following figure shows the use of the addressing parameters during the establishment of an end-to-end connection. The CANUREACH, ICANREACH, and REACH_ACK message types all carry the Data Link ID, consisting of the MAC and Link SAP addresses associated with the two end stations. The CANUREACH and ICANREACH messages are qualified by the SSPex flag into CANUREACH_ex, ICANREACH_ex (explorer messages) and CANUREACH_cs, ICANREACH_cs (circuit start). The CANUREACH_ex is used to find a remote MAC and Link SAP address without establishing an SSP circuit. Upon receipt of a CANUREACH_cs message, the target DLSw starts a data link for each port, thereby obtaining a Data Link Correlator. If the target station can be reached, an ICANREACH_cs message is returned to the origin DLSw containing the Target Circuit ID parameter. Upon receipt, the origin DLSw starts a data link and returns the Origin Circuit ID to the target DLSw within the REACH_ACK message. (Note for a full list of message types, see section 3.5.) +------------+ +------------+ |Disconnected| |Disconnected| +------------+ CANUREACH_cs (Data Link ID) +------------+ -------------------------------------------------> ICANREACH_cs (Data Link ID, Target Circuit ID) <------------------------------------------------ REACH_ACK (Data Link ID, Origin Cir ID, Target Cir ID) -------------------------------------------------> +------------+ +------------+ |Circuit Est.| |Circuit Est.| +------------+ +------------+ XIDFRAME (Data Link ID, Origin Cir ID, Target Cir ID) <------------------------------------------------> CONTACT (Data Link ID, Origin Cir ID, Target Cir ID) -------------------------------------------------> CONTACTED (Data Link ID, Origin Cir ID, Target Cir ID) <------------------------------------------------- +------------+ +------------+ | Connected | | Connected | +------------+ +------------+ INFOFRAME (Remote Circuit ID = Target Circuit ID) -------------------------------------------------> INFOFRAME (Remote Circuit ID = Origin Circuit ID) <------------------------------------------------- During the exchange of the XIDFRAME, CONTACT, and CONTACTED messages, the pair of Circuit ID parameters is included in the message format along with the DATA LINK ID parameter. Once the connection has been Wells & Bartky [Page 13]
RFC 1795 Data Link Switching April 1995 established, the INFOFRAME messages are exchanged with the shorter header. This header contains only the Circuit ID associated with the remote DLSw. The Remote Data Link Correlator and the Remote DLC Port ID are set equal to the Data Link Correlator and the DLC Port ID that are associated with the origin or target Data Link Switch, dependent upon the direction of the packet. 3.3 Correlators The local use, and contents of the Data Link Correlator, Port ID and Transport ID fields in SSP messages is an implementation choice. These fields have local significance only. The values received from a partner DLSw must not be interpreted by the DLSw that receives them and should be echoed "as is" to a partner DLSw in subsequent messages. All implementations must obey the following rules in this section (3.3) on the assignment and fixing of these correlator fields for each transport connection or circuit: The Transport ID fields are learned from the first SSP message exchanged with a DLSw partner (the Capabilities exchange). This field should not be varied by a DLSw after the capabilities exchange and must be reflected to the partner DLSw in every SSP control message. The Target Data Link Correlator, Target Port ID and Target Transport ID must remain the same once the Target DLSw has sent the ICANREACH_cs for a given circuit. The Origin DLSw must store the values specified in the ICANREACH_cs and use these on all subsequent SSP messages for this circuit. The Origin DLSw must allow these fields to vary until the ICANREACH_cs is received. Each SSP message issued for a circuit must reflect the values specified by the Target DLSw in the last SSP message for this circuit received by the Origin DLSw. Binary zero should be used if no such message has yet been received for a given circuit (apart from the Target Transport ID which will have been learnt as specified above). The Origin Data Link Correlator, Origin Port ID and Origin Transport ID must remain the same once the Origin DLSw has issued the REACH_ACK for a given circuit. The Target DLSw must store the values specified in the REACH_ACK and use these on all subsequent SSP messages for this circuit. The Target DLSw must allow these fields to vary until the REACH_ACK is received. Each SSP message issued for a circuit must reflect the values specified by the Origin DLSw in the last SSP message for this circuit received by the Target DLSw. Binary zero should be used if Wells & Bartky [Page 14]
RFC 1795 Data Link Switching April 1995 no such message has yet been received for a given circuit (apart from the Origin Transport ID which will have been learnt as specified above). For the purposes of correlator exchange, explorer messages form a separate circuit. Both DLSw partners must reflect the last received correlator values as specified above. However correlators learned on explorer messages need not be carried over to a subsequent circuit setup attempt. In particular, the Origin DLSw may elect to use the same values for the Origin Data Link Correlator and Origin Port ID when it issues a CANUREACH_cs after receiving an ICANREACH_ex or NETBIOS_NR_ex. However the Target DLSw must not assume that the CANUREACH_cs will specify any of the Target Data Link Correlator or Target Port ID that were exchanged on the explorer messages. Received SSP messages that require a valid Remote Circuit ID but cannot be associated with an existing circuit should be rejected with a HALT_DL_NOACK message. This is done to prevent a situation where one DLSw partner has a circuit defined while the other partner does not. The exception would be a HALT_DL_NOACK message with an invalid Remote Circuit ID. The HALT_DL_NOACK message is typically used in error situations where a response is not appropriate. The SSP messages requiring a valid Remote Circuit ID are all messages except the following: CANUREACH_ex, CANUREACH_cs, ICANREACH_ex, ICANREACH_cs, NETBIOS_NQ_cs, NETBIOS_NR_cs, DATAFRAME, NETBIOS_ANQ, NETBIOS_ANR, KEEPALIVE and CAP_EXCHANGE. 3.4 Largest Frame Size Field The Largest Frame Size (LF Size) field in the SSP Control Header is used to carry the LF Size bits across the DLSw connection. This should be used to ensure that the two end-stations always negotiate a frame size to be used on a circuit that does not require the Origin and Target DLSw partners to re-segment frames. This field is valid on CANUREACH_ex, CANUREACH_cs, ICANREACH_ex, ICANREACH_cs, NETBIOS_NQ_ex and NETBIOS_NR_ex messages only. The contents of this field should be ignored on all other frames. Every DLSw forwarding a SSP frame to its DLSw partner must ensure that the contents of this frame reflect the minimum capability of the route to its local end-station or any limit imposed by the DLSw itself. The bit-wise definition of this field is as follows (bit 7 is the most significant bit, bit 0 is the least significant bit): Wells & Bartky [Page 15]
RFC 1795 Data Link Switching April 1995 7 6 5 4 3 2 1 0 +-------------------------------+ | c | r | b | b | b | e | e | e | +-------------------------------+ c . . . . . . . LF Size Control flag (significant on messages from Origin to Target DLSw only) 0=fail circuit if route obtained requires a smaller LF size 1=don't fail the circuit but return the LF size obtained even if it is smaller . r . . . . . . Reserved . . b . . . . . Largest Frame Bit Base . . . b . . . . Largest Frame Bit Base . . . . b . . . Largest Frame Bit Base . . . . . e . . Largest Frame Bit Extended . . . . . . e . Largest Frame Bit Extended . . . . . . . e Largest Frame Bit Extended <----- LF Bits -----> Refer to IEEE 802.1D Standard, Annex C for encoding of Largest Frame base and extended bit values. The Origin DLSw "Size Control" flag informs a Target DLSw that chooses to reply to *_cs messages on the basis of cached information that it may safely return a smaller LF Size on the ICANREACH_cs frame if it has had to choose an alternative route on which to initialize the circuit. If this bit is set to 1, the Origin DLSw takes responsibility for ensuring that the end-stations negotiate a suitable frame size for the circuit. If this bit is set to 0, the Target DLSw must not reply to the CANUREACH_cs if it cannot obtain a route to the Target end station that support an LF Size at least as large as that specified in the CANUREACH_cs frame. 3.5 Message Types The following table lists the protocol data units that are exchanged between Data Link Switches. All values not listed are reserved for potential use in follow-on releases. Wells & Bartky [Page 16]
RFC 1795 Data Link Switching April 1995 Command Description Type flags/notes ------- -------- ------ ----------- CANUREACH_ex Can U Reach Station-explorer 0x03 SSPex CANUREACH_cs Can U Reach Station-circuit start 0x03 ICANREACH_ex I Can Reach Station-explorer 0x04 SSPex ICANREACH_cs I Can Reach Station-circuit start 0x04 REACH_ACK Reach Acknowledgment 0x05 DGRMFRAME Datagram Frame 0x06 (note 1) XIDFRAME XID Frame 0x07 CONTACT Contact Remote Station 0x08 CONTACTED Remote Station Contacted 0x09 RESTART_DL Restart Data Link 0x10 DL_RESTARTED Data Link Restarted 0x11 ENTER_BUSY Enter Busy 0x0C (note 2) EXIT_BUSY Exit Busy 0x0D (note 2) INFOFRAME Information (I) Frame 0x0A HALT_DL Halt Data Link 0x0E DL_HALTED Data Link Halted 0x0F NETBIOS_NQ_ex NETBIOS Name Query-explorer 0x12 SSPex NETBIOS_NQ_cs NETBIOS Name Query-circuit setup 0x12 (note 3) NETBIOS_NR_ex NETBIOS Name Recognized-explorer 0x13 SSPex NETBIOS_NR_cs NETBIOS Name Recog-circuit setup 0x13 (note 3) DATAFRAME Data Frame 0x14 (note 1) HALT_DL_NOACK Halt Data Link with no Ack 0x19 NETBIOS_ANQ NETBIOS Add Name Query 0x1A NETBIOS_ANR NETBIOS Add Name Response 0x1B KEEPALIVE Transport Keepalive Message 0x1D (note 4) CAP_EXCHANGE Capabilities Exchange 0x20 IFCM Independent Flow Control Message 0x21 TEST_CIRCUIT_REQ Test Circuit Request 0x7A TEST_CIRCUIT_RSP Test Circuit Response 0x7B Note 1: Both the DGRMFRAME and DATAFRAME messages are used to carry information received by the DLC entity within UI frames. The DGRMFRAME message is addressed according to a pair of Circuit IDs, while the DATAFRAME message is addressed according to a Data Link ID, being composed of a pair of MAC addresses and a pair of link SAP addresses. The latter is employed prior to the establishment of an end-to-end circuit when Circuit IDs have yet to be established or during circuit restart when Data Links are reset. Note 2: These messages are not used for the DLSw Standard but may be used by older DLSw implementations. They are listed here for informational purposes. These messages were added after publication of RFC 1434 and were deleted in this standard (adaptive pacing is now used instead). Wells & Bartky [Page 17]
RFC 1795 Data Link Switching April 1995 Note 3: These messages are not normally issued by a Standard DLSw, which uses the NB_*_ex messages as shown in section 5.4. However if a Standard DLSw attempts to interoperate with older DLSw implementations, these messages correspond to the NETBIOS_NQ and NETBIOS_NR messages used in RFC1434 both to locate the resource and to setup a circuit. This document does not attempt to provide a complete specification of the use of these messages. Note 4: A KEEPALIVE message may be sent by a DLSw to a partner DLSw in order to verify the TCP connection (or other future SSP carrying protocol) is still functioning. If received by a DLSw, this message is discarded and ignored. Use of this message is optional. For the exchange of NetBIOS control messages, the entire DLC header is carried as part of the message unit. This includes the MAC header, with the routing information field padded to 18 bytes, and the LLC header. The following message types are affected: NETBIOS_NQ, NETBIOS_NR, NETBIOS_ANQ, NETBIOS_ANR, and DATAFRAME when being used by NetBIOS systems. The routing information in the DLC header is not used by the remote Data Link Switch upon receiving the above five messages. Any SSP message types not defined above if received by a DLSw are to be ignored (i.e., no error action is to be performed). A Data Link Switch should quietly drop any SSP message with a Message Type that is not recognized or not supported. Receipt of such a message should not cause the termination of the transport connection to the message sender. 4. Circuit Priority At circuit start time, each circuit end point will provide priority information to its circuit partner. The initiator of the circuit will choose which circuit priority will be effective for the life of the circuit. If Priority is not implemented by the Data Link Switch, then "Unsupported" priority is used. 4.1 Frame format Circuit priority will be valid in the CANUREACH_cs, ICANREACH_cs, and REACH_ACK frames only. The relevant header field is shown below. The Circuit Priority value is a byte value at offset 22 in an SSP Control Message. Wells & Bartky [Page 18]
RFC 1795 Data Link Switching April 1995 The following describes the format of the Circuit Priority byte. 7 6 5 4 3 2 1 0 +-------------------+-----------+ | reserved | CP | +-------------------+-----------+ CP: Circuit Priority bits 000 - Unsupported (note 1) 001 - Low Priority 010 - Medium Priority 011 - High Priority 100 - Highest Priority 101 to 111 are reserved for future use Note 1: Unsupported means that the Data Link Switch that originates the circuit does not implement priority. Actions taken on Unsupported priority are vendor specific. 4.2 Circuit Startup The sender of a CANUREACH_cs is responsible for setting the CP bits to reflect the priority it would like to use for the circuit being requested. The mechanism for choosing an appropriate value is implementation dependent. The sender of an ICANREACH_cs frame will set the CP bits to reflect the priority it would like to use for the circuit being requested, with the mechanism for choosing the appropriate value being implementation dependent. The receiver of the ICANREACH_cs will select from the priorities in the CANUREACH_cs and ICANREACH_cs frames, and will set the value in the CP field of the REACH_ACK frame that follows to the value to be used for this circuit. This priority will be used for the life of the circuit. A CANUREACH_cs or ICANREACH_cs with the circuit priority value set to Unsupported (CP=000) indicates that the sender does not support the circuit priority function. Wells & Bartky [Page 19]
RFC 1795 Data Link Switching April 1995 Flow: DLSw A DLSw B CANUREACH_cs (CP=011) -----> Circuit initiator requests high Priority. <--------- ICANREACH_cs (CP=010) Circuit target requests medium priority. REACH_ACK (CP=010) --------> Circuit initiator sets the priority for this circuit to medium. The circuit initiator could choose either high or medium in this example. 5. DLSw State Machine The following state tables describe the states for a single circuit through the Data Link Switch. State information is kept for each connection. The initial state for a connection is DISCONNECTED. The steady state is either CIRCUIT_ESTABLISHED or CONNECTED. In the former state, an end-to-end circuit has been established allowing the support of Type 1 LLC between the end systems. The latter state exists when an end-to-end connection has been established for the support of Type 2 LLC services between the end systems. For SNA, LLC type 2 connection establishment is via the use of IEEE 802.2 Test or XID frames. SNA devices send these frames to the null SAP in order to determine the source route information in support of bridging. Normally SNA devices use SAP 0x04, 0x08, or 0x0C (most SNA LLC2 devices that have a single PU per MAC address use a default of 0x04). Typically the SAP would be used to determine if the Test frames should be sent to the DLSw code in the router. If both bridging and DLSw are enabled, this allows the product to ensure that SNA frames are not both bridged and switched. Note that although typically SNA uses a DSAP and SSAP of 0x04, it allows for other SAPs to be configured and supports unequal SAPs. This allows multiple PUs to share connections between two given MAC addresses (each PU to PU session uses one LLC2 connection). For NetBIOS, LLC type 2 connection establishment is via the Name Query and Name Recognized frames. These frames are used for both address resolution and source route determination. NetBIOS devices use SAP 0xF0. Wells & Bartky [Page 20]
RFC 1795 Data Link Switching April 1995 5.1 Data Link Switch States The Switch-to-Switch Protocol is formally defined through the state machines described in this chapter. The following table lists the thirteen possible states for the main circuit FSM. A separate state machine instance is employed for each end-to-end circuit that is maintained by the Data Link Switch. State Name Description ---------- ----------- CIRCUIT_ESTABLISHED The end-to-end circuit has been established. At this time LLC Type 1 services are available from end-to-end. CIRCUIT_PENDING The target DLSw is awaiting a REACH_ACK response to an ICANREACH_cs message. CIRCUIT_RESTART The DLSw that originated the reset is awaiting the restart of the data link and the DL_RESTARTED response to a RESTART_DL message. CIRCUIT_START The origin DLSw is awaiting a ICANREACH_cs in response to a CANUREACH_cs message. CONNECTED The end-to-end connection has been established thereby allowing LLC Type 2 services from end-to-end in addition to LLC Type 1 services. CONNECT_PENDING The origin DLSw is awaiting the CONTACTED response to a CONTACT message. CONTACT_PENDING The target DLSw is awaiting the DLC_CONTACTED confirmation to a DLC_CONTACT signal (i.e., DLC is waiting for a UA response to an SABME command). DISCONNECTED The initial state with no circuit or connection established, the DLSw is awaiting either a CANUREACH_cs, or an ICANREACH_cs. DISCONNECT_PENDING The DLSw that originated the disconnect is awaiting the DL_HALTED Wells & Bartky [Page 21]
RFC 1795 Data Link Switching April 1995 response to a HALT_DL message. HALT_PENDING The remote DLSw is awaiting the DLC_DL_HALTED indication following the DLC_HALT_DL request (i.e., DLC is waiting for a UA response to a DISC command), due to receiving a HALT_DL message. HALT_PENDING_NOACK The remote DLSw is awaiting the DLC_DL_HALTED indication following the DLC_HALT_DL request (i.e., DLC is waiting for a UA response to a DISC command), due to receiving a HALT_DL_NOACK message. RESTART_PENDING The remote DLSw is awaiting the DLC_DL_HALTED indication following the DLC_HALT_DL request (i.e., DLC is waiting for a UA response to a DISC command), and the restart of the data link. RESOLVE_PENDING The target DLSw is awaiting the DLC_DL_STARTED indication following the DLC_START_DL request (i.e., DLC is waiting for a Test response as a result of sending a Test command). The DISCONNECTED state is the initial state for a new circuit. One end station starts the connection via an XID or SABME command (i.e., DLC_XID or DLC_CONTACTED). Upon receipt, the Data Link Switches exchange a set of CANUREACH_cs, ICANREACH_cs and REACH_ACK messages. Upon completion of this three-legged exchange both Data Link Switches will be in the CIRCUIT_ESTABLISHED state. Three pending states also exist during this exchange. The CIRCUIT_START state is entered by the origin Data Link Switch after it has sent the CANUREACH_cs message. The RESOLVE_PENDING state is entered by the target Data Link Switch awaiting a Test response to a Test Command. And lastly, the CIRCUIT_PENDING state is entered by the target DLSw awaiting the REACH_ACK reply to an ICANREACH_cs message. The CIRCUIT_ESTABLISHED state allows for the exchange of LLC Type 1 frames such as the XID exchanges between SNA stations that occurs prior to the establishment of a connection. Also, datagram traffic (i.e., UI frames) may be sent and received between the end stations. These exchanges use the XIDFRAME and DGRMFRAME messages sent between Wells & Bartky [Page 22]
RFC 1795 Data Link Switching April 1995 the Data Link Switches. In the CIRCUIT_ESTABLISHED state, the receipt of a SABME command (i.e., DLC_CONTACTED) causes the origin DLSw to issue a CONTACT message, to send an RNR supervisory frame (i.e., DLC_ENTER_BUSY) to the origin station, and to enter the CONNECT_PENDING state awaiting a CONTACTED message. The target DLSw, upon the receipt of a CONTACT message, will issue a SABME command (i.e., DLC_CONTACT) and enter the Contact Pending state. Once the UA response is received (i.e., DLC_CONTACTED), the target DLSw sends a CONTACTED message and enters the CONNECTED state. When received, the origin DLSw enters the CONNECTED state and sends an RR supervisory frame (i.e., DLC_EXIT_BUSY). The CONNECTED state is the steady state for normal data flow once a connection has been established. Information frames (i.e., INFOFRAME messages) are simply sent back and forth between the end points of the connection. This is the path that should be optimized for performance. The connection is terminated upon the receipt of a DISC frame or under some other error condition detected by DLC (i.e., DLC_ERROR). Upon receipt of this indication, the DLSw will halt the local data link, send a HALT_DL message to the remote DLSw, and enter the DISCONNECT_PENDING State. When the HALT_DL frame is received by the other DLSw, the local DLC is halted for this data link, a DL_HALTED message is returned, and the DISCONNECTED state is entered. Receipt of this DL_HALTED message causes the other DLSw to also enter the DISCONNECTED state. The CIRCUIT_RESTART state is entered if one of the Data Link Switches receives a SABME command (i.e., DLC_RESET) after data transfer while in the CONNECTED state. This causes a DM command to be returned to the origin station and a RESTART_DL message to be sent to the remote Data Link Switch. This causes the remote data link to be halted and then restarted. The remote DLSw will then send a DL_RESTARTED message back to the first DLSw. The receipt of the DL_RESTARTED message causes the first DLSw to issue a new CONTACT message, assuming that the local DLC has been contacted (i.e., the origin station has resent the SABME command). This is eventually responded to by a CONTACTED message. Following this exchange, both Data Link Switches will return to the CONNECTED state. If the local DLC has not been contacted, the receipt of a DL_RESTARTED command causes the Data Link Switch to enter the CIRCUIT_ESTABLISHED state awaiting the receipt of a SABME command (i.e., DLC_CONTACTED signal). The HALT_PENDING, HALT_PENDING_NOACK and RESTART_PENDING states correspond to the cases when the Data Link Switch is awaiting Wells & Bartky [Page 23]
RFC 1795 Data Link Switching April 1995 responses from the local station on the adjacent LAN (e.g., a UA response to a DISC command). Also in the RESTART_PENDING state, the Data Link Switch will attempt to restart the data link prior to sending a DL_RESTARTED message. For some implementations, the start of a data link involves the exchange of a Test command/response on the adjacent LAN (i.e., DLC_START_DL). For other implementations, this additional exchange may not be required. 5.2 State Transition Tables This section provides a detailed representation of the Data Link Switch, as documented by a single state machine. Many of the transitions are dependent upon local signals between the Data Link Switch entity and one of the DLC entities. These signals and their definitions are given in the following tables. DLC Events: Event Name Description ---------- ----------- DLC_CONTACTED Contact Indication: DLC has received an SABME command or DLC has received a UA response as a result of sending an SABME command. DLC_DGRM Datagram Indication: DLC has received a UI frame. DLC_ERROR Error condition indicated by DLC: Such a condition occurs when a DISC command is received or when DLC experiences an unrecoverable error. DLC_INFO Information Indication: DLC has received an Information (I) frame. DLC_DL_HALTED Data Link Halted Indication: DLC has received a UA response to a DISC command. DLC_DL_STARTED Data Link Started Indication: DLC has received a Test response from the null SAP. DLC_RESET Reset Indication: DLC has received an SABME command during the time a connection is currently active and has responded with DM. DLC_RESOLVE_C Resolve Command Indication: DLC has received a Test command addressed to the null SAP, or an XID command addressed to the null SAP. Wells & Bartky [Page 24]
RFC 1795 Data Link Switching April 1995 DLC_RESOLVED Resolve request: DLC has received a TEST response frame (or equivalent for non-LAN DLCs) but has not reserved the resources required for a circuit yet. DLC_XID XID Indication: DLC has received an XID command or response to a non-null SAP. Other Events: Event Name Description ---------- ----------- XPORT_FAILURE Failure of the transport connection used by the circuit. CS_TIMER_EXP The CIRCUIT_START timer (started when the circuit went into CIRCUIT_START state) has expired. DLC Actions: Action Name Description ----------- ----------- DLC_CONTACT Contact Station Request: DLC will send a SABME command or a UA response to an outstanding SABME command. DLC_DGRM Datagram Request: DLC will send a UI frame. DLC_ENTER_BUSY Enter Link Station Busy: DLC will send an RNR supervisory frame. DLC_EXIT_BUSY Exit Link Station Busy: DLC will send an RR supervisory frame. DLC_HALT_DL Halt Data Link Request: DLC will send a DISC command. DLC_INFO Information Request: DLC will send an I frame. DLC_RESOLVE Resolve request: DLC should issue a TEST (or appropriate equivalent for non-LAN DLCs) but need not reserve the resources required for a circuit yet. DLC_RESOLVE_R Resolve Response Request: DLC will send a Test response or XID response from the null SAP. DLC_START_DL Start Data Link Request: DLC will send a Test command to the null SAP. Wells & Bartky [Page 25]
RFC 1795 Data Link Switching April 1995 DLC_XID XID Request: DLC will send an XID command or an XID response. Other Actions: Action Name Description ---------- ----------- START_CS_TIMER Start the CIRCUIT_START timer. DLC_RESOLVE_R and DLC_START_DL actions require the DLC to reserve the resources necessary for a link station as they are used only when a circuit is about to be started. The DLC_RESOLVE action is used for topology explorer traffic and does not require such resources to be reserved, though a DLC implementation may choose not to distinguish this from the DLC_START_DL action. See section 5.4 for details of the actions and events for explorer frames. The Data Link Switch is described by a state transition table as documented in the following sections. Each of the states is described below in terms of the events, actions, and next state for each transition. If a particular event is not listed for a given state, no action and no state transition should occur for that event. Any significant comments concerning the transitions within a given state are given immediately following the table representing the state. A separate state machine instance is maintained by the Data Link Switch for each end-to-end circuit. The number of circuits that may be supported by each Data Link Switch is a local implementation option. The CANUREACH_ex, ICANREACH_ex, NETBIOS_NQ_ex, and NETBIOS_NR_ex are SSP messages that are not associated with a particular circuit. The processing of these messages is covered in section 5.4. Wells & Bartky [Page 26]
RFC 1795 Data Link Switching April 1995
RFC 1795 Data Link Switching April 1995 000 - Repeat Window Operator 001 - Increment Window Operator 010 - Decrement Window Operator 011 - Reset Window Operator 100 - Halve Window Operator 101 - Reserved 110 - Reserved 111 - Reserved A frame with the FCI bit set is referred to as a Flow Control Indication (FCIND). An FCIND is used to manage the flow in the opposite direction of the frame which bears it. A frame with the FCA bit set is referred to as a Flow Control Acknowledgment (FCACK). An FCACK is used to manage the flow in the same direction of the frame which bears it. NOTE: A frame may be both a FCIND and an FCACK. A frame bearing an FCIND or FCACK may also contain data for the flow in the direction it is traveling. In such a frame, the FCIND or FCACK are said to be piggy-backed. A non-piggy-backed FCIND is called an Independent Flow Control Indication (IFCIND) and a non- piggy-backed FCACK is called an Independent Flow Control Acknowledgment (IFCACK). IFCIND and IFCACK messages are sent in a Independent Flow Control SSP message (type 0x21). NOTE: A frame may be both an IFCIND and an IFCACK. It is desirable to carry information in control messages so as to reduce the need to send a flow control only message. The diagram below shows the messages that may carry valid flow control information: ====== ___ ====== | | --------- __/ \__ --------- | | | | __| _|_ |__ / IP \ __| _|_ |__ | | ====== | | | < Network > | | | ====== /______\ --------- \__ __/ --------- /______\ Origin Origin DLSw \___/ Target DLSw Target Station partner partner Station May have valid FCI/FCA/FCO Data carrying N N CANUREACH_cs -----------> Y* N ICANREACH_cs Wells & Bartky [Page 78]
RFC 1795 Data Link Switching April 1995 <----------- Y N REACH_ACK -----------> Y Y XIDFRAMEs <------------> Y Y DGRMFRAMEs <------------> Y N CONTACT -----------> Y N CONTACTED <----------- Y Y INFOFRAMEs <------------> Y N RESTART_DL -----------> Y N DL_RESTARTED <----------- Y N CONTACT -----------> Y N CONTACTED <----------- N N HALT_DL -----------> N N DL_HALTED <----------- *Note: ICANREACH_cs cannot carry FCA, as there could not be an outstanding FCI. 8.3 Granting Permission to Send Data A receiver grants a sender permission to send units of data by sending FCIND. Each FCIND is further qualified by a flow control operator, which is encoded in the FCO bits of the FCIND header. With one exception (the Reset Window operator) all operators may be either piggy-backed or carried in a IFCIND. The five flow control operators are outlined below: 8.3.1 Repeat Window Operator This operator is processed as follows: (CurrentWindow unchanged) GrantedUnits += CurrentWindow Wells & Bartky [Page 79]
RFC 1795 Data Link Switching April 1995 8.3.2 Increment Window Operator This operator is processed as follows: CurrentWindow++ GrantedUnits += CurrentWindow 8.3.3 Decrement Window Operator This operator is processed as follows: CurrentWindow-- GrantedUnits += CurrentWindow NOTE: This operator may only be sent if CurrentWindow is greater than one. 8.3.4 Reset Window Operator This operator is processed as follows: CurrentWindow = 0; GrantedUnits = 0; NOTE: This operator may only flow on an independent pacing indication (may NOT be piggy-backed). NOTE: After sending this operator, the only legal subsequent operator is Increment Window. 8.3.5 Halve Window Operator This operator shall be processed as follows: IF CurrentWindow > 1 THEN CurrentWindow = CurrentWindow / 2 ENDIF GrantedUnits += CurrentWindow Note: The divide by two operation is an unsigned integer divide (round down) or bit shift right operation. 8.4 Acknowledging a Flow Control Operator Each sender must acknowledge each FCIND with an FCACK which is piggy-backed on the next frame in the opposite direction in all cases except the Reset Window Operator. Wells & Bartky [Page 80]
RFC 1795 Data Link Switching April 1995 The receiver may have no more than one unacknowledged FCIND outstanding at any time with one exception: A Reset Window Operator may be sent while another FCIND is pending acknowledgment. NOTE: The FCI and FCO bits of the FCACK are used independently by the flow in the opposite direction 8.4.1 Acknowledging a Reset Window Operator Since this operator revokes all previously granted units, the sender must acknowledge this FCIND using an IFCACK (Independent Flow Control Acknowledgment). This is the only case where IFCACK is used. Should a sender receive a non-reset FCIND followed by a Reset Window FCIND before acknowledging the first, it only acknowledges the Reset Window. NOTE: The FCI and FCO bits on these frames are used independently by the flow in the opposite direction. 8.5 Capabilities Exchange Initial Window Size When two nodes establish a transport connection, they engage in a capabilities exchange (this is a requirement). Refer to the Capabilities Exchange section 7 for further details. The two nodes are required to exchange the following parameter: InitialWindowSize - This indicates to the partner what the sending flow entity initializes its CurrentWindow value to for each multiplexed circuit subsequently established on that transport connection. This value must be non-zero. 8.6 Circuit Startup Process as follows: CurrentWindow = InitialWindowSize GrantedUnits = 0 NOTE: The InitialWindow Size variable has a scope of one per DLSw transport connection, while CurrentWindow and Granted units are maintained on a per circuit basis. At circuit startup, a sender may not send data units until the receiver grants explicit permission with an FCIND message. This grant may be an independent FCIND message or the FCIND may be piggy-backed on any of the message types Wells & Bartky [Page 81]
RFC 1795 Data Link Switching April 1995 listed in section 8.2. 8.7 Example Receiving Implementations The following two examples illustrate receiving implementations of varying degrees of complexity. These are not meant to be complete implementations but rather serve to illustrate the protocol. NOTE: The examples are independent of the buffering model ( buffers may be deterministicly or statistically committed) NOTE: The examples assume a process model where each event processes to completion without being preempted by another event. 8.7.1 Fixed Pacing Example Consider the following variables, in addition to InitialWindowSize and CurrentWindow and FCACKOwed: GrantDelayed - Boolean GrantedUnits - Outstanding Units The following section describes how various events are processed in this example implementation: 8.7.1.1 Circuit Startup CurrentWindow = InitialWindowSize FCACKOwed = FALSE GrantDelayed = FALSE GrantedUnits = 0 Repeat Window Operator 8.7.1.2 Check Buffers Available Can my implementation afford to grant CurrentWindow just now? 8.7.1.3 Buffers Become Available IF Check Buffers Available THEN Send FCIND( Repeat Window) GrantDelayed = FALSE ELSE Wait on buffers to become available (LIFO) ENDIF Wells & Bartky [Page 82]
RFC 1795 Data Link Switching April 1995 8.7.1.4 Repeat Window Operator IF Check Buffers Available THEN Send FCIND( Repeat Window) ELSE GrantDelayed = TRUE Wait on buffers to become available (FIFO) ENDIF 8.7.1.5 Send FCIND( operator) GrantedUnits += CurrentWindow FCACKOwed = TRUE Encode and Transmit FCIND piggybacked or as IFCIND 8.7.1.6 A Frame Arrives from Sender GrantedUnits--; IF frame is FCACK THEN IF FCACKOwed THEN FCACKOwed = FALSE ELSE Protocol Violation ENDIF ENDIF IF NOT GrantDelayed THEN IF GrantedUnits <= CurrentWindow THEN IF FCACKOwed THEN Protocol Violation ELSE Repeat Window Operator ENDIF ENDIF ENDIF 8.7.2 Adaptive Pacing Example The following example illustrates a receiving implementation that adjusts the window size and granted units based on buffer availability and transport utilization. NOTE: This example ignores other factors which might compel the receiving implementation to adjust the window size (i.e., Outbound queue length, traffic priority, ...) Consider the following variables, in addition to InitialWindowSize, CurrentWindow and FCACKOwed: Wells & Bartky [Page 83]
RFC 1795 Data Link Switching April 1995 GrantDelayed - Boolean GrantedUnits - Outstanding Units 8.7.2.1 Circuit Startup CurrentWindow = InitialWindowSize FCACK = FALSE GrantDelayed = FALSE GrantedUnits = 0 Repeat Window Operator 8.7.2.2 Check Buffers Available ( X) Can my implementation afford to grant X units just now? 8.7.2.3 Buffers Become Available IF Check Buffers Available THEN CurrentWindow--; Send FCIND( Decrement Window) GrantDelayed = FALSE ELSE Wait on buffers to become available (LIFO) ENDIF 8.7.2.4 Repeat Window Operator IF Check Buffers Available (CurrentWindow) THEN Send FCIND( Repeat Window) ELSE GrantDelayed = TRUE Wait on buffers to become available (FIFO) ENDIF 8.7.2.5 Increment Window Operator IF Check Buffers Available ( CurrentWindow + 1) THEN CurrentWindow++ Send FCIND( Increment Window) ELSE Repeat Window Operator ENDIF 8.7.2.6 Send FCIND( operator) FCACKOwed = TRUE GrantedUnits += CurrentWindow Encode and Transmit FCIND piggybacked or as IFCIND Wells & Bartky [Page 84]
RFC 1795 Data Link Switching April 1995 8.7.2.7 An FCACK Arrives from Sender GrantedUnits--; IF NOT FCACKOwed THEN Protocol Violation ENDIF FCACKOwed = FALSE; IF NOT GrantDelayed THEN IF GrantedUnits < CurrentWindow THEN Increment Window Operator ELSE IF GrantedUnits == CurrentWindow THEN Repeat Window Operator END ENDIF 8.7.2.8 A Non-FCACK Frame Arrives from Sender GrantedUnits--; IF NOT GrantDelayed THEN IF FCACKOwed THEN IF GrantedUnits < CurrentWindow THEN Protocol Violation END ELSE IF GrantedUnits <= CurrentWindow THEN Repeat Window Operator ENDIF ENDIF ENDIF Wells & Bartky [Page 85]
RFC 1795 Data Link Switching April 1995 8.8 Adaptive Pacing Example Flow Diagrams 8.8.1 Example Flows from the Above Implementation The following diagram illustrates the use of adaptive pacing (use of Halve Window, and Reset operation are shown in subsequent diagrams). -----SENDER----- ----RECEIVER---- Granted Window Window Granted 0 2 circuit established 2 0 2 2 <-------- FCIND(Rpt) 2 2 1 2 FCACK--------------> 2 1 4 3 <-------- FCIND(Inc) 3 4 3 3 FCACK--------------> 3 3 +- FCIND(Rpt) 3 6 2 3 DATA---|-----------> 3 5 1 3 DATA---|-----------> 3 4 4 3 <------+ 3 3 FCACK--------------> 3 3 6 3 <-------- FCIND(Rpt) 3 6 5 3 FCACK--------------> 3 5 4 3 DATA---------------> 3 4 3 3 DATA---------------> 3 3 +- FCIND(Rpt) 3 6 2 3 DATA---|-----------> 3 5 1 3 DATA---|-----------> 3 4 0 3 DATA---|-----------> 3 3 3 3 <------+ 2 3 FCACK--------------> 3 2 6 4 <-------- FCIND(Inc) 4 6 5 4 FCACK--------------> 4 5 4 4 DATA---------------> 4 4 Waiting on Buffer +- FCIND(Dec) 3 7 3 4 DATA---|-----------> 3 6 2 4 DATA---|-----------> 3 5 1 4 DATA---|-----------> 3 4 0 4 DATA---|-----------> 3 3 3 3 <------+ 2 3 FCACK--------------> 3 2 Waiting on Buffer +- FCIND(Dec) 2 4 1 3 DATA---|-----------> 2 3 0 3 DATA---|-----------> 2 2 2 2 <------+ 1 2 FCACK--------------> 2 1 4 3 <-------- FCIND(Inc) 3 4 3 3 FCACK--------------> 3 3 Wells & Bartky [Page 86]
RFC 1795 Data Link Switching April 1995 6 3 <-------- FCIND(Rpt) 3 6 5 3 FCACK--------------> 3 5 4 3 DATA---------------> 3 4 3 3 DATA---------------> 3 3 6 3 <-------- FCIND(Rpt) 3 6 8.8.2 Example Halve Window Flow The following flow illustrates the use of the Halve Window Operator: -----SENDER----- ----RECEIVER---- Granted Window Window Granted 0 2 circuit established 2 0 2 2 <-------- FCIND(Rpt) 2 2 1 2 FCACK--------------> 2 1 4 3 <-------- FCIND(Inc) 3 4 3 3 FCACK--------------> 3 3 Resource Shortage 2 3 DATA---------------> 1 2 1 3 DATA---------------> 1 1 0 3 DATA---------------> 1 0 1 1 <-------- FCIND(Hlv) 1 1 0 1 FCACK--------------> 1 0 NOTE: The Halve Window Operator could have been sent before the granted units fell to zero. The implementer may make a choice based on the severity of the condition. 8.8.3 Example Reset Window Flows The following flow diagram illustrates the ResetWindow operation if the receiver has no FCIND outstanding. -----SENDER----- ----RECEIVER---- Granted Window Window Granted 0 2 circuit established 2 0 2 2 <-------- FCIND(Rpt) 2 2 1 2 FCACK--------------> 2 1 4 3 <-------- FCIND(Inc) 3 4 3 3 FCACK--------------> 3 3 +- FCIND(Rpt) 3 6 2 3 DATA---|-----------> 3 5 1 3 DATA---|-----------> 3 4 4 3 <------+ 3 3 FCACK--------------> 3 3 6 3 <-------- FCIND(Rpt) 3 6 5 3 FCACK--------------> 3 5 Resource shortage! Wells & Bartky [Page 87]
RFC 1795 Data Link Switching April 1995 0 0 <-------- FCIND(Rst) 0 5 (note still committed) 0 0 IFCACK-------------> 0 0 Condition eases 1 1 <-------- FCIND(Inc) 1 1 0 1 FCACK--------------> 1 0 2 2 <-------- FCIND(Inc) 2 2 1 2 FCACK--------------> 3 4 The next two flows illustrate the Reset Window operation if the receiver has an outstanding FCIND. -----SENDER----- ----RECEIVER---- Granted Window Window Granted 0 2 circuit established 2 0 2 2 <-------- FCIND(Rpt) 2 2 1 2 FCACK--------------> 2 1 4 3 <-------- FCIND(Inc) 3 4 3 3 FCACK--------------> 3 3 +- FCIND(Rpt) 3 6 2 3 DATA---|-----------> 3 5 | Resource shortage! |+-FCIND(Rst) 0 5 1 3 DATA---||----------> 0 4 4 3 <------+| 3 3 FCACK---+----------> 0 3 (Not IFCACK!) 2 3 DATA----|----------> 0 2 0 0 <-------+ 0 0 IFCACK-------------> 0 0 Condition eases 1 1 <-------- FCIND(Inc) 1 1 0 1 FCACK--------------> 1 0 2 2 <-------- FCIND(Inc) 2 2 1 2 FCACK--------------> 3 4 -----SENDER----- ----RECEIVER---- Granted Window Window Granted 0 2 circuit established 2 0 2 2 <-------- FCIND(Rpt) 2 2 1 2 FCACK--------------> 2 1 4 3 <-------- FCIND(Inc) 3 4 3 3 FCACK--------------> 3 3 +- FCIND(Rpt) 3 6 2 3 DATA---|-----------> 3 5 | Resource shortage! |+-FCIND(Rst) 0 5 1 3 DATA---||----------> 0 4 4 3 <------+| Wells & Bartky [Page 88]
RFC 1795 Data Link Switching April 1995 0 0 <-------+ 0 0 IFCACK-------------> 0 0 Condition eases 1 1 <-------- FCIND(Inc) 1 1 0 1 FCACK--------------> 1 0 2 2 <-------- FCIND(Inc) 2 2 1 2 FCACK--------------> 3 4 8.9 Other Considerations 8.9.1 Protocol Violations The following events are considered protocol violations: 1. Sender exceeds granted units or does not acknowledge FCIND on first frame after its receipt (the receiver can not discern the difference between the two). 2. Receiver does not follow a Reset Window Operator with an Increment Window Operator. 3. Receiver has two unacknowledged FCINDs ( other than Reset Window) outstanding. 4. Receiver sends Decrement Window Operator with a window size of one. 5. Receiver attempts to increment the window size beyond 0xFFFF. Actions taken in response to protocol violations are left to the implementation of the node which discovers the violation. If an implementation chooses to take down the circuit on which the violation occurred, HALT_DL is the appropriate action. Acknowledgments Original RFC 1434 Authors: Roy C. Dixon, IBM David M. Kushi, IBM Chair of APPN Implementers Workshop Data Link Switching Related Interest Group: Louise Herndon Wells, Internetworking Technology Institute Wells & Bartky [Page 89]
RFC 1795 Data Link Switching April 1995 Working Group Chairs (and significant contributors to this document): Connect/Disconnect (State Machines): Steve Klein, IBM Capabilities Exchange: Wayne Clark, Cisco Systems Flow Control (Adaptive Pacing): Shannon Nix, Metaplex Priority/Class of Service: Gene Cox, IBM Other significant contributors: Peter Gayek, IBM Paul Brittain, Data Connection Limited References 1) ISO 8802-2/IEEE Std 802.2 International Standard, Information Processing Systems, Local Area Networks, Part 2: Logical Link Control, December 31, 1989. 2) IBM LAN Technical Reference IEEE 802.2 and NETBIOS Application Program Interfaces SC30-3587-00, December 1993. 3) ISO/IEC DIS 10038 DAM 2, MAC Bridging, Source Routing Supplement, December 1991. 4) ISO 8802-2/IEEE Std 802.1D International Standard, Information Processing Systems, Local Area Networks, Part 2: MAC layer Bridging. Wells & Bartky [Page 90]
RFC 1795 Data Link Switching April 1995 Security Considerations Security issues are not discussed in this memo. Chair's Address Louise Wells Internetwork Technology Institute 2021 Stratford Dr. Milpitas, CA 95035 EMail: lhwells@cup.portal.com Editor's Address Alan K. Bartky Manager of Technology Sync Research Inc. 7 Studebaker Irvine, CA 91728-2013 Phone: 1-714-588-2070 EMail: alan@sync.com Note: Any questions or comments relative to the contents of this RFC should be sent to the following Internet address: aiw-dlsw@networking.raleigh.ibm.com. This address will be used to coordinate the handling of responses. NOTE 1: This is a widely subscribed mailing list and messages sent to this address will be sent to all members of the DLSw mailing list. For specific questions relating to subscribing to the AIW and any of it's working groups send email to: appn@vnet.ibm.com Information regarding all of the AIW working groups and the work they are producing can be obtained by copying, via anonymous ftp, the file aiwinfo.psbin or aiwinfo.txt from the Internet host networking.raleigh.ibm.com, located in directory aiw. NOTE 2: These mailing lists and addresses are subject to change.



Back to RFC index

 

Associates:

 



Sponsered-Sites:

Register domain name and transfer | Cheap webhosting service | Domain name registration

 

 

""