High speed linking module

[0002] This application claims the benefit of Provisional No. 60/108,866 filed Nov. 17, 1998.

[0003] The present invention relates, in general, to an apparatus and method for linking data processors and peripheral devices and, more specifically, to an apparatus and method for linking devices over high speed fiber optic links.

[0004] Data centers are typically linked together so that data may be shared by multiple customers, who are remotely located from each other. The customers, such as banks and credit card issuers, need high speed connectivity between their server systems and mainframe systems to provide quality service and maximize their investment in information management. Applications that require such high speed connectivity include transaction co-processing, massive file transfers for decision support, archival databases for disaster recovery and transaction reporting requirements.

[0005] Referring to FIG. 1, there is provided a data processing interconnection system 10 of the prior art. An example of system 10 is an IBM data processing interconnection system known as Enterprise Systems Connection (“ESCON”) (Trade Mark). “ESCON” is an interconnection system using fiber optic technology. Fiber optic links, such as links 14, 18, 22 and 26 create a local area network extending for kilometers among numerous systems, such as host processor 12, control units 24 and 28.

[0006] In “ESCON” architecture multiple systems may communicate with each other via channel-to-channel communications. For example, multiple mainframe systems may communicate channel-to-channel or gain access to multiple devices or communication control units. Referring to FIG. 1, channel directors 16 and 20 are capable of employing any-to-any, point-to-point switching and may make numerous physical connections between each other and peripheral devices. As shown, channel director 20 connects four fiber optic links 18 from channel director 16 with two fiber optic links 22 and two fiber optic links 26, each respectively branching to control units 24 and 28.

[0007] Although not shown in FIG. 1, channel director 20 or 22 may have as many as 256 optical ports to support as many as 128 “ESCON” connections simultaneously and without contention. Each channel director includes a set of quad port adapters (QPAs). Each QPA handles the “ESCON” input or output data with four individual ports. The ports include either multi-mode optical transceivers or single-mode optical transceivers. Two QPA modules are shown in FIG. 1, namely QPA module 17 and QPA module 19.

[0008] A more detailed arrangement of QPAs is shown in FIG. 2. As illustrated, multiple QPAs 17a-17nare coupled to switch matrix 15 within channel director 16. Each QPA has four output/input ports (27a-27n) for providing up to four connections to remote devices. Switch matrix 15 provides the switching fabric to connect any one ESCON port to any other ESCON port. For example, two ports (27a) in QPA (1) are connected to two ports (27b) in QPA (2). In this manner, channel director 16 provides multiple interfaces and channel-to-channel switching among multiple devices.

[0009] A physical link between two points may consist of two fibers, one for transmitting and one for receiving. Information on the link is transmitted in a special 10-bit code, giving an instantaneous link rate of 20 megabytes per second or 200 megabits per second. After deducting for control (e.g., pacing bytes) and data encoding overhead, a channel data rate for real application data of 17 megabytes per second is achieved.

[0010] Data are transmitted in the form of packets of characters called frames. Each character contains 10 bits when use is made of the 8 of 10 code to provide a dc balanced code. The frames can vary in size from 12 bytes to 1036 bytes. Each frame includes both the frame source address and its destination address. The addresses are used to route frames through the network. A switch matrix controller (not shown) within the channel director examines the destination address and dynamically connects the port receiving the frame to the destination port.

[0011] “ESCON” technology permits a maximum link rate of 200 megabits/sec between channel directors. The physical links are one-to-one and one port is required at each channel director to support both sides of the link. This one-to-one arrangement can become expensive, because valuable ports and fiber are consumed to support communications between channel directors. Typically, a user must lease one fiber optic link for every port in a control unit. More detail of “ESCON” architecture is provided by S. A. Calta, et al. in “Enterprise Systems Connection (ESCON) Architecture-System Overview”, July 1992, (IBM Journal Res. Development, Vol. 36, No. 4) and is incorporated herein by reference.

[0012] A need still exists for an apparatus and method for communicating between channel directors that does not require a one-to-one physical link per port. A need also exists for an interface device that may simultaneously support connectivity from multiple “ESCON” ports onto a single fiber link to reduce the cost of leasing fiber links.

[0013] To meet this and other needs, and in view of its purposes, the present invention provides an apparatus and a method for communicating between two devices over a single link, wherein each device has a plurality of ports. The method includes:

[0014] (a) converting a plurality of data streams from respective ports of a first of the two devices into one serial data stream, including the steps of:

[0015] (i) multiplexing at least two of the plurality of data streams into a first intermediate output stream;

[0016] (ii) multiplexing at least two other of the plurality of data streams into a second intermediate output stream;

[0017] (iii) serializing the first and second intermediate output streams into the serial data stream,

[0018] (iv) synchronizing the serialization of the first and second intermediate output streams,

[0019] (v) tagging output data in the serial data stream as corresponding to data from each of the respective ports of the first device, and

[0020] (vi) transmitting the serial data stream on the single link to a second of the two devices;

[0021] (b) converting the serial data stream in the second device to at least four input data streams, and

[0022] (c) sending each of the input data streams to a port of the second device.

[0023] Step (b) includes the steps of:

[0024] (i) de-serializing the serial data stream into first and second intermediate parallel data streams,

[0025] (ii) demultiplexing the first intermediate parallel data stream into two parallel data streams on two respective data lines;

[0026] (iii) demultiplexing the second intermediate parallel data stream into two other parallel data streams on two other respective data lines;

[0027] (iv) synchronizing the de-serialization of the first and second intermediate output streams and

[0028] (v) separating data in the serial data stream to data for each of the respective data lines.

[0029] In one embodiment, the single link is a fiber optic link. Also included is a signal having a cycle with a first phase and a second phase, and tagging two of the respective ports with the first phase and tagging two other respective ports with the second phase.

[0030] It is understood that the foregoing general description and the following detailed description are exemplary, but are not restrictive, of the invention.

[0031] The invention is best understood from the following detailed description when read in connection with the accompanying drawing. Included in the drawing are the following figures:

[0032] FIG. 1 is a block diagram of a conventional channel-to-channel communications system;

[0033] FIG. 2 illustrates a conventional switch matrix for connecting optical links among devices in the communications system of FIG. 1 by using QPAs, each having four I/O ports;

[0034] FIG. 3 is a block diagram of a channel-to-channel communications system using an embodiment of the present invention;

[0035] FIG. 4 illustrates several director multiple interface facility (DMIF) modules of the present invention having replaced the QPA modules of FIG. 2;

[0036] FIG. 5 is a block diagram of a multiplexer of the present invention incorporated in the DMIF module of FIG. 4;

[0037] FIG. 6 is a block diagram of a demultiplexer of the present invention incorporated in the DMIF module of FIG. 4;

[0038] FIG. 7 illustrates an initialization sequence between two remote devices using DMIF modules and a control unit which does not have a DMIF module;

[0039] FIG. 8 is a block diagram of another embodiment of a multiplexer of the present invention further illustrating in-band message generation;

[0040] FIG. 9 is a block diagram of still another embodiment of a demultiplexer of the present invention further illustrating in-band message decoding;

[0041] FIG. 10 illustrates a data frame being transmitted or received between two devices of the present invention; and

[0042] FIG. 11 is a timing diagram illustrating multiplexing of data from ports P0, P1, P2 and P3 of the DMIF module of the present invention.

[0043] Referring to FIG. 3, an embodiment of the invention will now be described. Data processing interconnection system 30 includes host processor 12; links 14, 18, 22 and 26; and control units 24 and 28. These elements are similar to elements shown in FIG. 1. Also included are channel directors 32 and 38, respectively containing modules 34 and 36. The modules are each labeled Director Multiple Interface Facility (DMIF). In the embodiment shown in FIG. 3, DMIF 34 and DMIF 36 are connected by a single fiber optic link 18, thereby providing an advantage over the prior art system 10 which requires four fiber optic links 18 (FIG. 1).

[0044] As will be explained in detail, the DMIF module replaces, for example, a QPA module within each channel director, and acts and behaves like a QPA module to the rest of the system. The DMIF module, however, is different in that it includes one port instead of four ports. The DMIF module may multiplex four “ESCON” ports, each passing data at a rate of 200 megabits/sec, into one high speed port passing data at a rate of 960 megabits/sec. The DMIF module may also demultiplex the 960 megabits/sec into four individual data streams running at 200 megabits/sec. The DMIF module advantageously allows a user to transmit/receive data on four “ESCON” links with only one fiber link. This saves the user the expense of leasing three extra fiber cables.

[0045] Although only one fiber optic link 18 is shown in FIG. 3, it will be appreciated that there may be multiple fiber optic links between channel director 32 and channel director 38. There may also be multiple fiber optic links from one channel director to several channel directors (not shown). Thus, the savings to the user multiply for each DMIF module that replaces a QPA module. The savings multiply up to a maximum of ½ of the total number of QPA modules that may be inserted in a channel director. For example, in a channel director that contains 64 QPA modules (256 ports), a maximum of 32 DMIF modules may be inserted to provide a mix of 32 DMIF modules and 32 QPA modules. This maximum is due to the need for four individual ports to be contained within the channel director for each DMIF module.

[0046] The mix of modules is shown in FIG. 4. As shown, channel director 32 contains multiple DMIF modules 34aand 34band multiple QPA modules 17n−1 and 17n.Thus, QPA (1) and QPA (2) in FIG. 2 have been replaced with DMIF modules 34aand 34b.Since the DMIF modules each require one output port connected, for example, as shown to optic links 18 and 37, the complexity and expense of six ports is eliminated from the system shown in FIG. 2.

[0047] One embodiment of the DMIF module will now be explained by reference to FIGS. 5 and 6. FIG. 5 illustrates 4 to 1 multiplexing of data by multiplexer module 40 and FIG. 6 illustrates the inverse, namely 1 to 4 demultiplexing of data by demultiplexer module 60. It will be appreciated that each DMIF module may include both multiplexer module 40 and demultiplexer module 60. In this manner, two-way optical data may be transmitted and received on a single optic link 18. Optic link 18 may be, for example, a single mode (SM) fiber ({fraction (8/125)} microns cable) transmitting at 1300 nm wavelength at a rate up to 960 megabits per second over distances from 2 km to 10 km. A single mode (XDF), dual SC-type connector may be used. Furthermore, in the preferred embodiment there is one laser type fiber optic transceiver per module (shown separately as optical transmitter 54 in FIG. 5 and as optical receiver 62 in FIG. 6).

[0048] FIG. 5 illustrates 4 to 1 multiplexing of data, for example, from four “ESCON” ports by multiplexer module 40. “ESCON” port P0, P1, P2 or P3 provides 9 bit data bytes at 20 megabytes/sec rate. Multiplexer 42 multiplexes the data from “ESCON” ports P0 and P1 onto one data path for input to parity generator 48. Parity generator 48 adds parity to make a 10 bit word. Identical processing is performed on the data from “ESCON” ports P2 and P3 by way of multiplexer 44 and parity generator 50. Two 10 bit data streams are presented to serializer/deserializer (ser/des) transmitter 52 at 40 megabytes/sec rate. The result of multiplexing is that data from the four “ESCON” ports are sent as two data frames to ser/des transmitter 52. Time tag generator 46 provides a time tag which is toggled to identify which “ESCON” port's data resides on the 10 bit stream. This time tag is provided as an input to transmitter 52. For example, the time tag may be transmitted as a “0”, when the data frame contains data from ports P0 and P2; the time tag may be transmitted as a “1”, when the data frame contains data from ports P1 and P3. The state of the time tag is encoded into a control field, as will be explained later, and transmitted, as part of the data stream, onto optical link 18 to be received by another DMIF module, for example DMIF module 36. The time tag is then used by a demultiplexer, at the other end of the link, to demultiplex the data.

[0049] Ser/des transmitter 52 performs two to one multiplexing of data. As shown in FIG. 5, transmitter 52 accepts a data frame consisting of 20 bits (10 bits on parallel bus 41 and 10 bits on parallel bus 43) which contain data from two “ESCON” ports (P0 and P2 or P1 and P3). Transmitter 52 also performs a parallel to serial conversion of the data and outputs the data onto serial bus 45. The data are then transmitted onto link 18 by optical transmitter 54.

[0050] It will be appreciated that transmitter 52 may be, for example, part of a transceiver chip set from Hewlett Packard, the HDMP-1022 transmitter and the HDMP-1024 receiver. This chip set is also referred to herein as the GLINK chip set. As known to those familiar with the chip set, encoding, multiplexing, clock extraction, demultiplexing and decoding are handled by the chip set. The chip set operates within a frequency range of 150 megabits/sec to 1.5 gigabits/sec. Using “ESCON” data rate of 200 megabits/sec, for example, the chip set operates at 960 megabits/sec, which includes four control bits inserted within the data. The control bits delineate whether the frame is a data, control, or fill frame. These frames will be described later.

[0051] As shown, a strobe-in clock is provided on line 47 to transmitter 52. Transmitter 52 uses the strobe-in clock to latch the parallel input data and, by multiplying the clock by 24, produces a serial clock to output the serial data at the higher rate.

[0052] It will also be appreciated that transmitter 52 may maintain a DC balance of the data on the link by determining the cumulative sign of the data and control bits of the frame. Based on the sign of the previous and present frames, transmitter 52 may invert the present frame to keep the cumulative sign of all of the frames equal to 0. By maintaining the DC balance of the serial data, there are no restrictions on the type of data that may be sent.

[0053] In another embodiment, FIG. 6 shows demultiplexer module 60 of DMIF module 36. As shown, optical receiver 62 receives data from link 18 and provides the data serially to serializer/deserializer receiver 64. Receiver 64 indirectly demultiplexes the data by performing a serial to parallel conversion. The parallel data on busses 61 and 63 together contain a 20 bit wide word that is output at 40 MHz, a rate twice the normal ESCON parallel rate. A time tag is provided from receiver 64 to further demultiplex the data into four ESCON paths. It will be understood that receiver 64, for example, may be part of a transceiver chip set from Hewlett Packard, the HDMP-1024, described above.

[0054] Receiver 64 may provide the first half of the 20 bit wide word to parity checker 66 and the second half of the 20 bit wide word to parity checker 68. Each parity checker verifies the parity and provides the remaining 9 bit wide word to demultiplexer 70 and demultiplexer 72, respectively. The time tag, shown in FIG. 6, may be used to demultiplex the data into four outputs, P0, P1, P2 and P3. When the time tag is 0, for example, the data frame may contain data for ports P0 and P2; when the time tag is 1, for example, the data frame may contain data for ports P1 and P3.

[0055] It will be appreciated that receiver 64 performs the inverse function performed by transmitter 52 (FIG. 5). Receiver 64 determines the frequency of the link and generates a recovered serial clock during initialization by locking onto the transition in the fill frames (described later). The receiver obtains the 24 bit serial frame and locks the recovered serial clock to the data. The receiver may utilize the transition within the control bit portion of the frame to phase align the recovered clock with the data. The receiver performs serial to parallel conversion of the frame to produce the 20 bit wide word. The 4 bit control field is decoded to generate status indicating whether a data, control, or fill frame is received (described later). The receiver may also generate a parallel clock (STRBOUT) for clocking the data to external logic.

[0056] In one embodiment of the invention, each data frame contains 24 bits. Twenty of the bits contain information. The other four bits determine whether a data frame, a control frame, or fill frame is present.

[0057] A data frame, for example, is used to send normal “ESCON” data. Each frame contains data that is 20 bits wide, and includes four control bits to indicate that the frame is a data frame. Table 1 illustrates an example of the contents of a data frame. The D-Field contains the 20 bit data word and the C-Field contains the 4 bit control field. The Data Status column defines the data when it is unaltered (True) or complemented (Inverted) and depends on the value of the cumulative sign of past and present data. The ‘*’ after the data bits in the D-Field indicates that the bits have been inverted. In the exemplary embodiment, the D-Field is transmitted first (D₀first) and the C-Field is transmitted last.

[0058] Data characters are used, for example, to transfer information between “ESCON” hosts and peripherals and may be data placed on link 18 (FIG. 3). Each character contains 10 bits. The most significant bit is an odd parity bit and forces a data character to contain an odd number of 1s. The next most significant bit indicates whether the character is an “ESCON” data character or a control character. For data characters, the bit is a ‘0’. The next 8 bits contain information passed between the host and peripherals and may be any one of 256 combinations (2⁸).

[0059] Control characters are used, for example, to send control information between the “ESCON” hosts and peripherals any may be data placed on link 18 (FIG. 3). These characters delineate the start and end of a frame and initiate connection recovery and offline procedures. Other functions for these control characters are outlined in “ESCON I/O Interface,” IBM, SA 22-7202-02 (June 1992) and is incorporated herein by reference.

[0060] Like a data character, a control character consists of 10 bits. The most significant bit is used to force odd parity. The ninth bit determines whether the character is data or control. For a control character, this bit is a ‘1’. The next eight bits represent the “ESCON” control characters whose format results from a chip residing in a QPA module. The chip also creates characters for certain conditions occurring on the link, such as errors and frame delimiters. These characters may be transmitted from a QPA module through the switch matrix and to a DMIF module (see FIG. 4, for example). The DMIF module passes these characters without alteration.

[0061] “ESCON” architecture defines only a few control characters. Since 8 bits are available for defining a control character, as many as 256 (2⁸) possible control characters may be defined and added by the DMIF module for transmission over the link to another DMIF module. For example, Table 2 defines various control characters that may be transmitted over a link between two DMIF modules. Some of the control characters are defined by ESCON architecture and generated by a QPA module and other control characters are generated by a DMIF module. The manner of generating control characters in the DMIF module will be described later.

[0062] Table 3 defines the control characters shown in Table 2.

[0063] In-band message handling capability allows the DMIF to transfer control commands and status to other DMIFs. This information is transferred over the same communication path as the ESCON data so that an extra path between these devices is not needed. The communication is transparent to other ESCON devices attached to a DMIF.

[0064] In one embodiment, the data interface between two DMIFs carries four ESCON paths of data that are time division multiplexed onto one high-speed link. In-band messages are passed over this link as well. The particular time slot (port) it occupies as well as the particular bit within the ESCON byte defines each message. This technique has the advantage of allowing the DMIF to pass status to the other DMIF within a particular port's slot when it is not busy, even if the other ports are too busy to allow the transmission of status in their port's slots. There are two types of in-band messages that may be transferred: commands and status.

[0065] The definition of a message being transmitted depends on the 5 most significant bits of the ESCON 9 bit word, the particular time slot that it is sent, and the specific bits within the least significant 4 bits that are set. A command message is defined as a word that takes the form of 1FX or 1DX, where X represents the least significant 4 bits. A status message is defined as a word that takes the form of 18X or 19X. Since there are four time slots and each time slot has a different meaning, a specific 9-bit pattern within one slot has a different meaning than an identical pattern within a different time slot. For example, a command message of 1F4 received within port 0's time slot has a different meaning than a command message of 1F4 received within port 1's time slot. Exemplary command and status messages are defined in Tables 4 and 5.

[0066] Protocol requires that messages be transferred within two consecutive time slots. For example, a message defined for port 0 may be sent in two consecutive time slots pertaining to port 0. The message may only be sent when specific ESCON patterns are contained within the time slot. These patterns are Loss of Signal characters, sequences, or idles. The message replaces these patterns on the multiplexed link. If these patterns do not exist at the time that the message is to be transmitted, the sender may hold the message until the time slot for the message contains one of these patterns. At this point, the message may be sent.

[0067] Messages may also be received in-band with the ESCON data. The type of message, whether it is a command or status message, is determined by the most significant 5 bits of the 9 bit word. For a command word, these 5 bits are ‘1F’ or ‘1D’ and for a status word, these 5 bits are ‘18’ or ‘19’. The message has to be received in two consecutive time slots for the port from which the message was sent. If only one time slot contains the message, it is discarded. The message words are also replaced by the data content that surrounds them before being transferred to the 200 Mbitsls links. For example, if a message is sent within a not operational sequence, the message words are replaced by the ordered pair that make up the sequence such that the ESCON device attached to the DMIF sees an uninterrupted not operational sequence. If this were not done, the ESCON devices would detect the messages as code violation errors that might force an erroneous loss of synchronization condition. Additionally, for the example just given, the ESCON device would detect that the sequence ended and began again, resulting in an additional link incidence report.

[0068] A command message, for example, may be sent from one DMIF to another DMIF. The other DMIF then may respond with a status message on the same port's time slot from which the command message is received. If a response is not received within 1 second, the message is retried once.

[0069] As described in one embodiment of the invention, messages are defined according to the port's time slot in which it exists, the value of the 5 most significant bits in the 9 bit word, and the particular bit with the least significant 4 bits that is asserted. Command messages contain a value of ‘1F’ or ‘1D’ in the 5 most significant bits and status messages contain a value of ‘18’ or ‘19’. Table 4 defines the command and status messages of one embodiment. Each definition includes the port's time slot in which the message exists and the particular bit within the least significant 4 bits that is defined for the message.

[0070] Tables 4A and 4B define the ‘1F’ command messages. In general, one bit is set for each message that forces a condition within a peripheral unit attached to a DMIF. Such peripheral unit may be, for example, a DMIF downstream unit (DDU). If future commands are received with this bit set, the condition is preserved. If a future command contains the bit negated, the condition is disabled. For example, a loop-back on port 0 exists as long as commands are received on port 0 that contain bit 2 asserted. If a command is received on this port with the bit negated, the loop-back is disabled. There may be cases where multiple bits are asserted within a command message in order to preserve conditions from previous commands. For example, if port 0 is in loop-back mode from a previous command and a Send Status command is to be sent while preserving the loop-back, the message would contain the value ‘1F5’ (Send Status and Loop-back Port 0 bits asserted).

[0071] It will be understood that in DMIF to DMIF communications, only the Request Rev command is used. The other commands shown in Table 4A are used in DMIF to DDU communications.

[0072] Tables 4C and 4D show the In-Band ‘1Dx’ command messages. One bit is set for each message that forces a condition within a DDU. If future commands are received with this bit set, the condition is preserved. If a future command contains the bit negated, the condition is disabled. For example, a loop-back on port 0 exists as long as commands are received on port 0 that contain bit 2 asserted. If a command is received on this port with the bit negated, the loop-back is disabled. There may be cases where multiple bits are asserted within a command message in order to preserve conditions from previous commands. For example, if port 0 is in loop-back mode from a previous command and a Send Status command is to be sent while preserving the loop-back, the message contains the value ‘1F5’ (Send Status and Loop-back Port 0 bits asserted).

[0073] Tables 5A and 5B define the ‘18x’ status messages. These status messages may be sent in response to a command or as unsolicited alarms. In either case, the status messages are not a command/response message pair. Individual or multiple bits may be asserted in the status messages depending on the conditions that are present within the DDU.

[0074] It will be understood that in DMIF to DMIF communications, only the REV ID Response status is used (in response to a Request Rev Command). The other responses shown in Table 5A are used in DMIF to DDU communications.

[0075] Table 5C and 5D define the ‘19x’ status messages. These status messages are returned to the DMIF on the same port's time slot in which the command was received. Individual or multiple bits may be asserted in the status messages depending on the conditions that are present within a DDU.

[0076] For “ESCON” links, sequences are used by devices at both ends of the link to advance to an activation state whereby both devices are synchronized. The DMIF modules at both ends execute a similar process whereby both transceivers (for example, ser/des transmitter in DMIF module 34 and ser/des receiver in DMIF module 36) become synchronized with each other. This process involves passing fill frames between DMIF modules. Two levels of synchronization may occur when a link is being initialized. The first level is the ser/des transmitter and ser/des receiver synchronizing by using fill frames. When this completes, data may be passed on the link. The second level is “ESCON” synchronization using “ESCON” protocol sequences. Although the link is already synchronized, this second level is needed to advance the “ESCON” Loss of Synchronization state machine to an Idle state (described later). The fill frame structure is shown in Table 6.

[0077] Fill frame 0 and fill frame 1 are defined in greater detail in Table 7.

[0078] It will be appreciated that other messages with different formats may be transmitted over the link between two DMIF modules. These messages, for example, other control and status messages, may be transmitted over the link in-band or in-line with the “ESCON” messages. In-band or in-line message handling by the DMIF module allows information to be transferred over the same communications path or link such that an extra path between two channel directors is not necessary. Communication is transparent to the “ESCON” devices attached to a channel director at each end of a link.

[0079] Initialization procedure between two DMIF modules will now be described. Link initialization may occur whenever a ser/des receiver, for example, is reset. The reset may be executed when the DMIF module is powered on, the DMIF module's optical transceiver detects no light on the link, or if all of the “ESCON” ports detect a loss of synchronization due to parity errors or code violation errors. The reset initializes the state machine in the ser/des receiver, forcing it to state ‘0’. By sending specific signals (detailed below) between the ser/des receiver and ser/des transmitter, the advancement of the state machine to the active state is transparent to the rest of the “ESCON” system.

[0080] While in state 0, the ser/des transmitter transmits the FF0 frame. Once the ser/des receiver recognizes that it is receiving the FF0 frame, it advances the state machine to state 1 which enables the ser/des transmitter to transmit the FF1 frame. After the ser/des receiver detects the FF1 frame, it advances the state machine to state 2 and generates LINK READY. While in state 2, the ser/des transmitter transmits FF1 until ESCON data is ready to be sent.

[0081] An example of an initialization sequence between remote devices containing DMIF modules is shown in FIG. 7. As shown, DMIF module 60 of channel director 32 is linked to DMIF module 62 over optical link 61. DMIF module 62 of channel director 38 also communicates through four dedicated links 63 with control unit 64. The initialization sequence between DMIF module 60 and DMIF module 62 is generally designated as 66. The initialization sequence is between DMIF module 62 and control unit 64 is generally designated as 68.

[0082] The sequences, as shown, progress from top to bottom in FIG. 7. DMIF module 60 initially transmits a fill frame (see Table 7) so that the ser/des transmitter may synchronize with the ser/des receiver in DMIF module 62. An NOP (not operational) is sent from DMIF module 62 to control unit 64 indicating that the DMIF is not operational. This is normal “ESCON” traffic and represents the flow on all four links 63. DMIF module 62 then sends a fill frame to DMIF module 60. As already described, the fill frames synchronize the two DMIF modules with each other.

[0083] As normal “ESCON” traffic is flowing with messages indicating the NOP sequence, the DMIF modules transmit in-band messages (previously described). These are shown in FIG. 7 with the in-band message identified in parenthesis. For example, NOP (REQ BOARD REV) indicates a Request BOARD REV command transmitted within the NOP sequence. Accordingly, once DMIF module 60 recognizes that link 61 is synchronized (after fill frame messages), DMIF module 60 issues a Request BOARD REV in-band command to determine the identity of the external device. DMIF module 62 also issues a Request BOARD REV in-band command. Each DMIF module then responds with a DMIF REV, indicating that each is connected to a DMIF module. At this point, no further action is necessary and both DMIF modules are now on-line. Normal “ESCON” data, designated generally as 69, may now be sent between remote devices by way of the DMIF modules. For example, FIG. 7 shows OFL (Offline Sequence), UD (Unconditional Disconnect Sequence), UDR (Unconditional Disconnect Response Sequence) and IDLE (Idle Sequence) sequentially transmitted between remote devices. It will be appreciated that commands and status messages sent between DMIF 60 and DMIF 62 are never seen by control unit 64, since they are replaced with NOP by the DMIF before passing them onto the control unit. In this manner, in-band messages are transparent to the control unit.

[0084] Another exemplary embodiment of in-band messages added to “ESCON” data is shown in FIG. 8 and an embodiment in which in-band messages are removed from “ESCON” data is shown in FIG. 9. As shown in FIG. 8, multiplexer 80 receives data from four “ESCON” ports of switch matrix 82. Data from the four ports are stored in respective buffers 84-87. Data path handlers 88-91 detect the presence of data from a respective port. Each data path handler also provides control and status interface between itself and processor 95. The interface is a thirty-two bit, interrupt driven interface that allows processor 95 to emulate the “ESCON” protocol. The emulation includes establishing or breaking connection and monitoring status. The data path handlers are implemented in four XCS30-3 Xilinx Spartan FPGAs.

[0085] When a connection is made to pass data by a data path handler, the processor determines which port needs to be connected in the switch matrix. The program for processor 95 is loaded from memory 102 which is, for example, non-volatile memory. Processor 95 also provides the in-band message to in-band message generator 94, which in turn sends the in-band message to multiplexer 96 and multiplexer 97. As shown in FIG. 8, multiplexer 96 is coupled between multiplexer 92 and parity generator 99. Similarly, multiplexer 97 is coupled between multiplexer 93 and parity generator 100. The multiplexing of data from two ports each having a 200 megabit/sec rate into a serial data output from ser/des transmitter 101 having a 960 megabit/sec rate has already been described.

[0086] It will be appreciated that the data begins as 9 bits wide at the input to each multiplexer 92 and multiplexer 93, 1 bit parity is added by parity generator 99 and parity generator 100 and 4 bits of control data are added by ser/des transmitter 101. Thus, a data frame consists of 24 bits of data. The data frame is illustrated in FIG. 10 as 9 bits of data (D0-D7 and a control/data (C/D) bit), a parity bit (P) and 4 bits for control (C0-C3). Since 4 to 1 multiplexing is performed and ser/des transmitter 101 serializes the data, an effective output rate of $960\mathrm{megabits}/\sec \mathrm{is}\mathrm{produced}\left(4\times \frac{20.0\mathrm{Mbytes}}{\sec }\times \frac{1\mathrm{frame}}{2\mathrm{bytes}}\times \frac{24\mathrm{bits}}{1\mathrm{frame}}=960\mathrm{Mbits}/\sec \right).$

[0087] FIG. 11 illustrates multiplexing of data, for example, from four ports, (P0, P1, P2 and P3) by multiplexer 92 and multiplexer 93 (FIG. 8) into P0/P1 and P2/P3 data. As described earlier, time tag generator 98 (FIG. 8) produces a time tag which is toggled to identify the ESCON port in the 10 bit stream. The time tag signal has two phases, a first phase (“0”) and a second phase (“1”). When a “0” is transmitted to ser/des transmitter 101 (FIG. 8), for example, the data frame may contain data from ports P0 and P2; when a “1” is transmitted, for example, the data frame may contain data from ports P1 and P3. In this manner, data from four ports may be multiplexed in one time tag cycle.

[0088] FIG. 9 shows the inverse process illustrated in FIG. 8. As shown, demultiplexer 120 receives serialized data from an optical transceiver (not shown) and ser/des receiver 122. Receiver 122 converts the incoming message having a data rate of 960 megabits/sec into two separate 10 bit wide parallel lines, each coupled to parity checker 123 and 124, respectively. Parity checkers 123 and 124 each verify parity and provide a 9 bit wide word to demultiplexer 125 and 127, respectively. Processor 95 and message decoder 126 decode the in-band message for control and status information.

[0089] After synchronization and recovery of the serial clock by ser/des receiver 122, a time tag signal is provided to time tag generator 135, which in turn provides the time tag to demultiplexers 128 and 129. By maintaining proper phase with the time tag signal, demultiplexers 128 and 129 demultiplex the data into four “ESCON” data words, each 9 bits wide. Data path handlers 88-91 receive the four “ESCON” data words and provide the same to buffers 130-133. Data are then sent to “ESCON” ports P0, P1, P2 and P3 by way of switch matrix 82.