Method and apparatus for converting addresses of a backing store having addressable data storage devices for accessing a cache attached to the backing store
Multiprocessor multisystem communications network
Virtual memory address translation mechanism with combined hash address table and inverted page table
Methods and apparatus for decontaminating hash tables
Integrated CPU and DMA with shared executing unit
System with a plurality of hash tables each using different adaptive hashing functions
Method of reducing the amount of information included in topology database update messages in a data communications network
Packet routing switch
Computer network switching system
DescriptionFIELD OF THE INVENTION
The present invention relates to network switches generally and to switching Ethernet controllers in particular.
BACKGROUND OF THE INVENTION
A network switch creates a network among a plurality of end nodes, such as workstations, and other network switches connected thereto. Each end node is connected to one port of the network. The ports also serve to connect network switchestogether.
Each end node sends packets of data to the network switch which the switch then routes either to another of the end nodes connected thereto or to a network switch to which the destination end node is connected. In the latter case, the receivingnetwork switch routes the packet to the destination end node.
Each network switch has to temporarily store the packets of data which it receives from the units (end node or network switch) connected to it while the switch determines how, when and through which port to retransmit the packets. Each packetcan be transmitted to only one destination address (a "unicast" packet) or to more than one unit (a "multicast" or "broadcast" packet). For multicast and broadcast packets, the switch typically stores the packet only once and transmits multiple copiesof the packet to some (multicast) or all (broadcast) of its ports. Once the packet has been transmitted to all of its destinations, it can be removed from the memory or written over.
Switching Ethernet controllers are network switches that implement the Ethernet switching protocol. According to the protocol, the Ethernet network (cabling and Ethernet ports) operates at 10 Megabits per second. However, most switches do notoperate at that speed, since they require longer than the 10 Mbps to process the incoming packets. Thus, their throughput is less than 10 Mbps. Switches which do operate at the desired speed are known as providing "full-wire" throughput.
SUMMARY OF THE PRESENT INVENTION
It is an object of the present invention to provide an improved switching Ethernet controller (SEC) which provides full-wire throughput.
The SEC of the present invention achieves the high-speed operation by utilizing a plurality of elements whose operations are faster than those of the prior art.
For example, in accordance with a preferred embodiment of the present invention, the communication between SECs attempts to utilize the bus as little as possible so that the bus will be available as soon as an SEC wants to utilize it. Inaccordance with the present invention, each SEC includes a write-only bus communication unit which transfers the packets out of the SEC by utilizing the bus only for write operations. Thus, packets enter each SEC by having been written therein fromother SECs and not by reading them in, since read operations utilize the bus for significant amounts of time compared to write operations. Having the bus available generally whenever a SEC needs it helps to provide the full-wire throughput.
In addition, the address table controller operates with a hash table storing addresses of the ports within the Ethernet network. The controller hashes the address of a packet to an initial hash table location value and then accesses that tablelocation. If the address stored at the table location matches that of the input address, the port information is retrieved. However, if the address stored at the table location is other than that of the input address, rather than reading a pointer tothe next location where values corresponding to the same hashed address can be found (as in the prior art), the present invention changes the hash table location values by a fixed jump amount and reads the address stored at the next table address. Dueto the fixed jump amount, the hash table controller of the present invention always knows what the next possible table location is.
A further speed increase is found in the accessing of the temporarily stored packets. In the present invention, the packets are stored in a storage buffer including a multiplicity of contiguous buffers. Associated with the buffers is an emptylist including a multiplicity of single bit buffers. A packet storage manager associates the state of the bit of a single bit buffer with the empty or full state of an associated contiguous buffer and generates the address of a contiguous buffer througha simple function of the address or number of its associated single bit buffer. The simple function is typically a multiplication operation.
The present invention also incorporates a network of SECs interconnected with PCI busses.
Finally, there is provided, in accordance with a preferred embodiment of the present invention, an Ethernet network including a) at least two groups of network switches, b) at least two PCI switch busses, wherein each group of network switchesis connected to one of the PCI busses, c) at least two PCI-to-PCI bridges, wherein each PCI-to-PCI bridge is connected to one of the PCI switch busses and d) at least one interconnection PCI bus to which the PCI-to-PCI bridges are connected.
BRIEF DESCRIPTION OF THE DRAWINGS
The present invention will be understood and appreciated more fully from the following detailed description taken in conjunction with the drawings in which:
FIG. 1A is a schematic illustration of a network of switching Ethernet controllers;
FIG. 1B is a schematic illustration of a network of switching Ethernet controllers interconnected by PCI busses;
FIG. 2 is a block diagram illustration of a generally full-wire throughput, switching Ethernet controller, constructed and operative in accordance with a preferred embodiment of the present invention;
FIG. 3 is a schematic illustration of an empty list block unit forming part of the switching Ethernet controller of FIG. 2;
FIG. 4 is a flow chart illustration of a bit clearing mechanism forming part of the empty list block unit of FIG. 3;
FIG. 5 is a schematic illustration of a hash table address recognition unit, constructed and operative in accordance with a preferred embodiment of the present invention;
FIG. 6 is a block diagram illustration of the logic elements of the address recognition unit of FIG. 3;
FIG. 7 is a schematic diagram of a hash function, useful in the address recognition unit of FIG. 3;
FIG. 8 is a schematic illustration of two network switches performing a write-only bus transfer protocol;
FIG. 9A is a flow chart illustration of the operations performed by the two switches of FIG. 8 during the data transfer operation of the present invention; and
FIG. 9B is a timing diagram illustration of the activity of the bus during the operation of FIG. 9A.
DETAILED DESCRIPTION OF THE PRESENT INVENTION
Reference is now made to FIGS. 1A, 1B and 2 which illustrate, in general terms, the generally full-wire throughput, switching Ethernet controller (SEC) 10 of the present invention and its connection within a network, wherein each SEC 10 formspart of a network switch 12.
FIG. 1A illustrates a plurality of network switches 12 connected as a peripheral component interface (PCI) bus, thereby to form a network. Optionally, a processor 16 and its associated memory unit 18 can also be connected to the bus 14.
FIG. 1A illustrates one network switch 12 in some detail. As shown, the switch 12 comprises a memory unit 20, such as a dynamic random access memory (DRAM) array, and a plurality of twisted pair drivers 22 for filtering data from an Ethernetunit 19 which implements a plurality of Ethernet ports. There typically is one twisted pair driver 22 per Ethernet port. The SECs 10 of each network switch typically provide the switching operations, switching data from port to port and from port tonetwork switch, all in accordance with the switching information found in the headers of each data packet. The processor 16 can also be involved in the switching, as described in more detail hereinbelow.
FIG. 1B illustrates the interconnection of network switches 12 to create a large network or to enlarge an existing network. A plurality of network switches 12 are connected to PCI busses 14A and 14B. In FIG. 1B, PCI busses 14A and 14B areconnected together via PCI bus 14C to which they are connected through PCI-to-PCI bridges 28A and 28B, respectively. Thus, two bus networks can be connected together through the addition of another PCI bus and two PCI-to-PCI bridges.
FIG. 2 details the elements of one SEC 10. It comprises an Ethernet interface unit 30, a frame control unit 32, a switching unit 34, an inter SEC control unit 36 and a bus interface unit 38. The Ethernet interface unit 30 performs the Ethernetprotocol through which unit 30 communicates the packets to and from the other elements of the SEC 10. The frame control unit 32 directs the packet into and out of the DRAM memory unit 20, as per instructions of the switching unit 34, and provides theEthernet header data to the switching unit 34. The switching unit 34 determines where to send each packet, either out one of its ports or out the bus 14 to another of the network switches. The inter SEC control unit 36 controls the communication withthe bus 14. The bus interface unit 38 physically transfers packets to and from the bus 14. The two interface units 30 and 38 perform standard protocols (Ethernet and PCI bus, respectively) and therefore, will not be described hereinbelow in any detail.
The frame control unit 32 typically includes input and output multiple first-in, first-out (FIFO) buffers 40 and 42, respectively, a direct memory access (DMA) unit 44 and a descriptor control 46. The FIFO buffers 40 and 42 each have one FIFObuffer per port defined by the Ethernet interface unit 30. The descriptor control 46 controls 9 circular (or ring) transmit queues which are stored in the DRAM 20. Each queue lists the packets to be transmitted through one of the eight ports or throughthe PCI bus 14. The descriptor control 46 maintains read and write pointers for each queue so as to know which packets are still waiting to be transmitted.
For incoming packets, input FIFO buffer 40 receives and buffers packets from the ports. The DMA unit 44 transfers the currently available packet provided by the input FIFO buffer 40 to the DRAM 20 in accordance with the instructions from theswitching unit 34. After the packet has been properly received, the switching unit 34 indicates to the descriptor control 46 through which port to transfer the packet. The descriptor control 46 places information about the packet into the relevanttransmit queue and, when the packet rises to the top of the transmit queue, the descriptor control 46 indicates to the DMA 44 to transfer the packet from the DRAM 20 to the buffer in output FIFO 42 for the appropriate port.
The switching unit 34 typically includes an empty list block 50, a hash table address control unit 52, an arbiter 54 and a DRAM interface 56. The empty list block 50 manages the organization of the DRAM 20, noting which buffers of the DRAM 20are available for storing newly arrived packets and which buffers contain packets to be transferred out. As will be described in more detail hereinbelow, the empty list block 50 associates an empty list of single bit buffers with the buffers of the DRAM20. In addition, the empty list block 50 associates the state of the bit of a single bit buffer with the empty or full state of an associated DRAM buffer and generates the address of a DRAM buffer through a simple function of the address or number ofits associated single bit buffer. The simple function is typically a multiplication operation. Thus, when a buffer request is received, the empty list block 50 relatively quickly can determine the address of the next available buffer.
When the empty list block 50 receives buffer assignment requests from the DMA 44 or from the inter SEC control unit 36, the empty list block 50 assigns the currently available buffer based on the state of the single bits of the empty list. Similarly, on output, when the empty list block 50 receives notification from the descriptor control 46 of the buffers which have successfully been either placed into the output FIFO 42 or transferred to another SEC (via the inter SEC control wait 36),the empty list block 50 then updates the state of the associated single bit buffer.
The hash table address control unit 52 receives the source and destination address information of the packet header from the Ethernet interface unit 30. As will be described in more detail hereinbelow, control unit 52 operates in conjunctionwith a hash table (physically found in DRAM 20) of the possible addresses of the entire network. The control unit 52 hashes the address of a packet to an initial hash table location value and then accesses that table location. If the address stored atthe table location matches that of the input address, the port information is retrieved. However, if the address stored at the table location is other than that of the input address, the present invention changes the hash table location values by afixed jump amount and reads the address stored at the next table address. Due to the fixed jump amount, the hash table controller of the present invention always knows what the next possible table location is for the current hash value and thus, cangenerally quickly move through the hash table to match the input address and to produce the associated port number.
Arbiter 54 controls the access to the DRAM 20 and DRAM interface 56 accesses the DRAM 20 for each piece of data (a packet or an address in the hash table) being stored or removed. Arbiter 54 receives DRAM access requests from the hash tablecontrol unit 52, the DMA unit 44, the descriptor control unit 46 and the inter SEC control unit 36.
The hash table control unit 52 provides the port associated with the destination address of the incoming packet to the descriptor control 46. Similarly, the empty list block 50 provides the descriptor control 46 with the buffer number in whichthe incoming packet is stored. When both values are received and the packet has been properly received (that is, without any corrupted data), the descriptor control 46 places the received buffer information in the transmit queue for the received buffernumber and, at the appropriate moment, initiates the transfer of the packet from the DRAM 20 into the queue of output FIFO 42 for the appropriate port. A slightly different operation occurs for the PCI transmit queue, as will be described hereinbelow.
The inter SEC control unit 36 typically includes a PCI DMA 60, a write-only transfer manager 62 and three interrupt registers, buffer request register 64, start of packet register 66 and end of packet register 68. The transfer manager 62supervises the transfer protocol which, in accordance with a preferred embodiment of the present invention, is performed with only write operations. As discussed hereinabove, write operations utilize the bus 14 for relatively short periods of time only.
The descriptor control 46 activates the write only transfer manager 62 whenever there is buffer information in the PCI transmit queue for a packet which has not been transmitted. The descriptor control 46 provides the transfer manager 62 withthe buffer address of the packet to be transferred and the port number of the destination SEC 10 to which the destination end node is attached.
To begin the transfer, the transfer manager 62 first prepares a "buffer request" message and writes the message into the buffer request register 64 of the destination SEC 10. Typically the buffer request includes at least the address of thebuffer storing the packet to be transferred and the port number of the destination SEC 10 to which the destination end node is attached.
The presence of a message in register 64 causes the transfer manager 62 of the destination SEC 10 to request that the empty list block 50 allocate a buffer in the DRAM 20 for the packet to be transferred. The empty list block 50 reviews itsempty list (without reading anything from the DRAM 20) and allocates the next available buffer (by changing the state of the bit associated with the buffer) to the packet to be transferred. The empty list block 50 provides the address of the allocatedbuffer to the transfer manager 62 which prepares a "start of packet" message with the address of the allocated buffer. The transfer manager 62 of the destination SEC 10 then writes the "start of packet" message to the start of packet register 66 of thesource SEC 10. Typically, the "start of packet" message includes at least the address of the allocated buffer (in the destination SEC 10), the address of the buffer (in the source SEC 10) storing the packet to be transferred and the port number of thedestination end node.
The presence of a message in the start of packet register 66 causes the transfer manager 62, of the source SEC 10, to activate the PCI DMA 60 to write the contents of the buffer storing the packet to be transferred in the allocated buffer in thedestination SEC 10. The PCI DMA 60 of the source SEC 10 actually writes the packet to the PCI DMA 60 of the destination SEC 10 which, in turn, writes the transferred packet to the allocated buffer of its DRAM 20 after receiving permission from itsarbiter 54. The transfer manager 62 also prepares an "end of packet" message and then writes the message into the end of packet register 68 once the packet to be transferred has been successfully transferred. Finally, the transfer manager 62 indicatesto the empty list block 50 to clear the bit of the empty list which is associated with the transferred packet. The "end of packet" message includes at least the destination port number.
The transfer manager 62 of the destination SEC 10 responds to the "end of packet" message by providing its descriptor control 46 with the port and buffer numbers of the transferred packet. The descriptor control 46 then adds the bufferinformation to the transmit queue for the indicated port. The packet is then transferred to the port as described hereinabove.
The following describe the empty list block 50, the hash table control unit 52 and the write-only transfer protocol in more detail.
Empty List Block 50
Reference is now made to FIG. 3 which schematically illustrates the empty list block 50 and its operation with the other elements of the SEC 10. Block 50 comprises an empty list 110 and its associated multiple buffer 112 (stored in DRAM 20), anempty list controller 114 and a bit clearing mechanism 121. FIG. 3 also shows the ports 120 (of Ethernet unit 30) to and from which the packets of data pass, DMA 44 and hash table address control 52.
In accordance with the present invention, the buffer 112 comprises a multiplicity of contiguous buffers 122, each of M bits and large enough to store, for example, at least one packet of 1518 bytes. For example, M might be 1.5K or 1536 bytes. Alternatively, each buffer 122 might hold many packets.
Furthermore, in accordance with a preferred embodiment of the present invention, the empty list 110 is a buffer of single (0 or 1) bits 124, each associated with one of the buffers 122. FIG. 3 shows 12 of each of buffers 122 and single bitbuffers 124; typically, there will be 1024 or more of each of buffers 122 and single bit buffers 124.
Buffers 124 store the value of 1 when their associated buffer 122 stores a not-yet retransmitted packet and a 0 when their associated buffer 122 is free to be written into. The buffers 122 and bits 124 are associated as follows: the address ofthe beginning of a buffer 122 is M times the address (or number) of the single bit buffer 124 associated therewith. In other words, for M=1.5K, the buffer 122 labeled 3 begins at address 4.5K and the buffer 122 labeled 0 begins at address 0. Alternatively, the first buffer 122 can begin at an offset K and thus, the address of the beginning of a buffer i is M times the address of the single bit buffer 124 associated therewith plus the offset K.
The empty list block 50 operates as follows: when a port 120 provides a packet, the DMA 44 requests the number of the next available buffer 122 from the empty list controller 114. Empty list controller 114 reviews the empty list 110 for thenext available single bit buffer 124 whose bit has a 0 value. Empty list controller 114 then changes the bit value to 1, multiplies the address of next available buffer 124 by M (and adds an offset K if there is one) and provides the resultant address,which is the start location of the corresponding buffer 122, to DMA 44.
It will be appreciated that the empty list block 50 provides a very simple mechanism by which to determine and store the address of the next available buffer 122. The mechanism only requires one multiplication operation to determine the addressand the address value is stored as a single bit (the value of buffer 124), rather than as a multiple bit address.
DMA 44 then enters the data from the incoming packet into the selected buffer 122. Once DMA 44 has finished entering the data, it indicates such to the hash table address control unit 52 which in the meantime, has received the destination andsource end node addresses from the Ethernet unit 30. Unit 52 determines through which port to retransmit the packet. Empty list controller 114 provides unit 52 with the number of the buffer 122 in which the packet is stored.
When a packet is to be retransmitted, the empty list controller 114 provides the DMA 44 with the buffer address for the packet and the hash table address control 52 provides the DMA 44 with the port number. DMA 44 reads the data from the buffer122 and provides the packet to the FIFO buffer for the relevant port 120.
For unicast packets, once the DMA 44 has finished transmitting the data of the selected buffer 122, DMA 44 indicates such to empty list controller 114 and includes in the indication the beginning address of the selected buffer 122. Empty listcontroller 114 then determines the buffer number of the selected buffer 122 and changes the bit value of the associated single bit buffer 124 to 0, thereby indicating that the selected buffer 122 is now available.
Buffers 122 are larger by at least N bits than the maximum amount of data to be stored therein. N is the number of ports connected to the switch plus the number of switches connected to the current switch. For example, N might be 46. Theextra bits, labeled 132, are utilized, for multicast packets, to indicate the multiple ports through which the packet has to be transmitted.
When the multicast packet enters the switch, DMA 44 sets all of the bits 132 (since multicast packets are to be sent to everyone). After the DMA 44 has transmitted a packet, whose port number it receives from the address control 52, the DMA 44indicates such to the empty list controller 114. If the packet is a multicast packet, the address control unit 52 indicates to the empty list controller 114 to read the N bits 132 to determine if any of them are set. If they are, empty list controller114 indicates to DMA 44 to reset the bit associated with the port 120 through which the packet was sent. When the DMA 44 indicates that it has finished resetting the bit, the empty list controller 114 does not change the associated single bit buffer124.
If the empty list controller 114 reads that only one bit is still set (i.e. the previous transmission was the last time the packet had to be transmitted), when the DMA 44 indicates that it has finished resetting the bit, the empty listcontroller 114 changes the bit value of the associated single bit buffer 124 to 0, thereby indicating that the associated buffer 122 is now available.
In the empty list 110, bits typically change as data is received and transmitted. However, it is possible for data not to be transmitted if there are some errors in the network, such as a port being broken or a switch being removed from thenetwork. In any of these cases, the bits in the empty list 110 associated with those ports must be cleared or else the associated buffers 122 will never be rewritten.
Therefore, the present invention includes bit clearing mechanism 121 which reviews the activity of the bits in the single bit buffers 124 and clears any set bits (i.e. of value 1) which have not changed during a predetermined period T. Theperiod T is typically set to be small enough to avoid wasting storage space for too long but large enough to avoid clearing a buffer before its turn for transmission has occurred.
Bit clearing mechanism 121 comprises a multiplexer 140 and a state reviewer 142. The multiplexer 140 connects, at one time, to a group of single bit buffers 124 and switches between groups of buffers every period T. State reviewer 142 reviewsthe state of the group of single bit buffers 124 to determine if all of the single bit buffers 124 changed from 1 to 0 at least once during the period T. If, at the end of period T, one or more bits in buffers 124 have remained in the set state (i.e.with value 1), the state reviewer 142 clears them to 0. Multiplexer 140 then connects to the next group of single bit buffers 124.
The operations of the bit clearing mechanism 121 are detailed in FIG. 4. Specifically, at each clock tick ti, the state reviewer 142 checks (step 150) each bit. If the bit has changed to 0, the bit is marked (step 152) as "changed". Otherwise, nothing occurs. The process is repeated until the period T has ended (step 154).
At the end of the period T, the state reviewer 142 clears (step 156) any unchanged bits and the multiplexer 140 changes (step 158) the group. The process is repeated for the next time period T.
Hash Table Control Unit 52
Reference is now made to FIGS. 5 and 6 which illustrate the hash table control unit 52 of the present invention. FIG. 5 illustrates the hash table control unit 52 and its operation and FIG. 6 details the elements of unit 52. The term "address"will be used herein to refer to MAC addresses and the term "location" will be utilized to refer to addresses within the hash table 212.
Hash table control unit 52 comprises a hash table 212 and a hash table location generator 214. Hash table 212 is shown with only 18 locations; it will be appreciated that this is for the purposes of clarity only. Typically, hash table 212 willhave 32K locations therein and, in accordance with the present invention, stores only the MAC address and the port associated therewith.
Location generator 214 receives the MAC address, whether of the source end node or of the destination end node, and transforms that address, via a hash function, to a table location. The hash function can be any suitable hash function; onesuitable function is provided hereinbelow with respect to FIG. 7.
In accordance with the present invention, if the generated table location stores an address which is not the same as the input MAC address, the location generator 214 generates a second location which is X locations further down in the hashtable 212. The hash table does not store any pointers to the next location. In accordance with the present invention, X is a prime number such that, if it is necessary to move through the entire hash table 212, each location will be visited only onceduring the review.
For example, and as shown in FIG. 5, X is 5 and the first table location is the location labeled 1. If the MAC address of location 2 does not match that of the input MAC address, the location generator 214 "jumps" to location 6 (as indicated byarrow 220), and then to location 11 (arrow 222), and then to location 16 of the hash table 212 (arrow 224). Since there are only 18 locations in the hash table 212 of FIG. 5, location generator 214 then jumps to location 3 (arrow 226) which is (16+5)mod 18. If location 4 is also full, location generator 214 will generate locations until all of the locations of table 212 have been visited.
It will be appreciated that the hash table control unit 52 does not need to have pointers in table 212 pointing to the "next" location in the table. As a result, unit 52 knows, a priori, which locations in the table are next and can,accordingly, generate a group of locations upon receiving the MAC address. If desired, the data in the group of locations can be read at once and readily compared to the input MAC address.
FIG. 6 illustrates the elements of the location generator 214 and its operation in conjunction with the table 212. Location generator 214 comprises a hash function generator 230, DRAM interface 56 (since the hash table 212 is typicallyimplemented in DRAM 20), a latch 234 and a comparator 236.
The hash function generator 230 converts the MAC address MA, of 48 bits, to the table location TL0, of 15 bits. The DRAM interface 56 generates the group of next table locations TL0, TL1 and TL2, where TL1=TL.sub.0+Xand TL2=TL.sub.0+2X, It will be appreciated that FIG. 6 illustrates only three table locations but many more or many less can be generated at once, as desired.
DRAM interface 56 accesses the table 212 to read the addresses, A0, A1 and A2, and their associated data d0, d1 and d2, stored in table locations TL0, TL1 and TL2, respectively. The data diinclude the necessary information about the address, such as the switch identification number and any other desired information. The read operation can be performed at once or successively.
The output of each table location is latched by latch 234. Comparator 236 then compares the address information Ai with that of MAC address MA. If the two addresses match (i.e. a "hit"), then comparator 236 indicates to latch 234 tooutput the associated data di stored therein. Otherwise, comparator 236 indicates to DRAM interface 56 to read the address Ai and associated data di stored in the next table location.
If many table locations are to be read at once, the location generator 214 can include a multiplicity of latches 234, one for each location to be read at once.
If one of the table locations is empty, as indicated by a valid bit of the data di, all locations after it will also be empty. Thus, the input MAC address has no corresponding stored address and therefore, the input MAC address istypically input into the empty table location. The valid bit in the associated data di is then set to `not empty`.
FIG. 7, to which reference is now made, illustrates an exemplary hash function, for typical MAC addresses, which can be performed by hash function generator 230. In this embodiment, generator 230 considers only the 33 lowest significant bits(LSBS) of the MAC address. The 33 LSBs are divided into four bytes, labeled A, B, C and D. Byte A consists of bits 0:5, byte B consists of bits 6:14, byte C consists of bits 15:23 and byte D consists of bits 24:32. Thus, byte A is 6 bits and theremaining bytes are 9 bits.
Hash function generator 230 comprises two XOR units 240A and 240B, a concatenator 242 and a swap unit 244. The XOR unit 240A performs an exclusive OR between bytes C and D and XOR unit 240B performs an exclusive OR between the output of XORunit 240A and byte B. Concatenator 242 concatenates the output of XOR unit 240B with byte A, thereby producing variable T of 15 bits. Swap unit 244 swaps the bits of variable T to produce the output table location TL. Thus, the value of TLreceives the value of T, the value of TL receives that of T, etc. It will be appreciated that any hash function can be utilized. However, the desired hash functions are those which provide a uniform distribution of tablelocations for the expected MAC addresses. It is noted that the above hash function is easily implemented in hardware since XOR units and concatenators are simple to implement.
Write-Only Transfer Manager 62
Reference is now made to FIG. 8 which illustrates the network configuration of the present invention and to FIGS. 9A and 9B which illustrate the data transfer operation of the present invention. Elements of FIG. 8 which are similar to those ofFIG. 2 have the same reference numerals. It is noted that bus 14 has at least two lines, a data line 340 and an address line 342.
In accordance with the write-only protocol of the present invention, packets of data are not transferred until a buffer location 319 is allocated for them in the DRAM 20 of the destination network switch 12B. Furthermore, since the transferoperation is a DMA transfer, a packet is directly written into the location allocated therefor.
In accordance with a preferred embodiment of the present invention, when a packet of data is to be transferred, the source network switch 12A initially writes (step 350, FIG. 9A) a "buffer request" message to the buffer request register 64b ofthe destination network switch 12B. The buffer request message asks that the destination network switch allocate a buffer for the data to be transferred.
In the DMA transfer embodiment of the present invention, the source network switch 12A provides, on address line 342, the address of the "buffer request" register, the address of destination network switch 12B and its "return" address. Sourcenetwork switch 12A provides, on data line 340, the size (or byte count) of the packet to be transferred and the buffer location 319A in which it is stored. The data of the data line is then written directly into the buffer request register.
In response to the buffer request message, the destination network switch 12B determines (step 352) the buffer location 319B in which the packet can be stored. It then writes (step 354) a "start of packet" message to the start of packetregister 66a of the source network switch 12A which includes at least the location of the allocated buffer and the port numbers of the source and destination network switches. It can also include the byte count.
For example, in the DMA transfer embodiment of the present invention described hereinabove, the destination network switch 12B provides, on address line 342, the address of the "start of packet" register and the address of source network switch12A. Destination network switch 12B provides, on data line 340, at least the following: the byte count of the packet to be transferred, the address 319B of the allocated buffer, the port number of the destination network switch 12B, and, foridentification, the buffer location 319A in which the data is stored in the source network switch 12A and the port number of the source network switch 12A. As before, the data of the data line is then directly written into the start of packet register.
In response to receipt of the start of packet message in the start of packet register, the source network switch 12A writes (step 356) the packet of data to the allocated buffer location, followed by an "end of packet" message. Once the sourcenetwork switch 12A has finished writing the end of packet message, it is free to send the next packet, beginning at step 350.
In the above described embodiment, the writing of the packet of data involves providing the address of the destination network switch 12B and the buffer location 319B on the address line 342 and the packet to be transferred on the data line 340. The transferred packet is then directly written into the allocated buffer location 319B. The end of packet message is written in a similar manner to the other messages, but to end of packet register 68b. The address information includes the address ofthe end of packet register and the address of the destination network switch 12B. The data includes the port number of the destination network switch 12B, the buffer location 319B and the byte count.
When the packet arrives at the destination network switch 12B it directly writes (step 360) the packet into the allocated buffer location 319B, as per the address on the address line 342, until it receives the end of packet message for thatallocated buffer location. The destination network switch 12B is now free to perform other operations until it receives a next buffer allocation request.
FIG. 9B illustrates the timing of the packet transfer described in FIG. 9A. The initial source write operation of the buffer request message (step 350) is typically relatively short since write operations take relatively little time and sincethe message to be transferred is small. Some time later, there is a destination write (DW) operation of the start of packet message (step 354). The destination write operation takes approximately the same length of time as the first source writeoperation. Some time later, there is a further source write operation (step 356) of the packet transfer and end of packet message. Since, for this operation, there is more data to be transferred, this source write operation is shown to take a longertime than the other two write operations. The source and network switches are free to perform other operations after they finish their writing operations.
It is also noted that, in the present invention, the source network switch 12A is free to operate on other packets once it has finished writing its packet, and its associated end of packet message, to the bus. The source network switch 12A doesnot need to ensure that the destination network switch 12B has successfully received the racket since, in the present invention, the address for the data (in the destination network switch) is known and is fully allocated prior to sending the packet; thepacket would not be sent if there was no buffer location available for it. In the present invention, the time it takes for the destination network switch 12B to process the packet is not relevant to the operation of the source network switch 12A.
It will be appreciated by persons skilled in the art that the present invention is not limited to what has been particularly shown and described hereinabove. Rather the scope of the present invention is defined by the claims which follow: