Device for datastream decoding

Abstract

A device for data stream analyzing that are able to recognize different data streams and then start processors or functionalities to store or check data in a data stream, comprising processor means including a program memory, making it possible to parse a data stream in a way that is controlled by an interchangeable program. There will be no need for changing the hardware. This could save time and money for companies responsible for providing, maintaining and updating network switches. The device also includes a multiplexable data stream delayline for receiving said data streams, and multiplexing means for connecting different parts of the data stream to said processor.

Description

FIELD OF THE INVENTION

[0001] The present invention relates to packet switching, more specifically to data stream decoding and data stream analysing.

RELATED ART

[0002] In the field of data and computer communications there is an increasing need for high speed/high bandwidth products. Documents relating to packet switching and more specifically to data stream decoding and pertinent to the present invention includes:

[0003] U.S. Pat. No. 5509006,

[0004] JP 6/276198,

[0005] EP 767565,

[0006] EP 953897,

[0007] U.S. Pat. No. 5594869

[0008] The problem of extracting address information in a switch from a packet in a data stream is in the prior art solved by applying masks on the content of one or more delay lines to filter out the required information. One disadvantage with this approach is the difficulty to adjust the switch to new communication protocols, because the masks are hardware implemented. Another disadvantage with the prior art is that the data in the delay line is only accessible at a certain position or certain positions, instead of being available all the time they reside in the delay line.

[0009] Accordingly, it is an object of the present invention to provide a device for improved programmable datastream analysis in the context of packet switching. In the context of this document a datastream can be any type of data stream, e.g. a bytewise Ethernet datastream in a computer network, including an Ethernet packet with different combinations of contents.

SUMMARY OF THE INVENTION

[0010] The invention relates to a device for data stream analysing. Said device is able to recognise different data streams and then start other processors or functionalities to store or check data in a data stream. Special features are: a compare processor, a compare instruction memory, a data stream pipeline, a multiplexer and a multiplexer control unit, making it possible to test packet data under program control using several instructions and under several clock cycles even though said data is moving forward in the pipeline and even though other bytes of data is entering the device.

BRIEF DESCRIPTION OF THE DRAWINGS

[0011] The invention will be described in detail below with reference to the accompanying drawings in which

[0012] FIG. 1 is a block diagram of the invention

[0013] FIG. 2 is a block diagram of the multiplexer control unit

[0014] FIG. 3 is an interface overview of the invention

DETAILED DESCRIPTION OF PREFERRED EMBODIMENTS

[0015] The invention is preferably implemented as an integrated circuit (IC) having an electrical interface to the outside. The invention comprises a number of physical or logical units including;

[0016] a delayline 1

[0017] a multiplexer 2

[0018] a multplexer control unit 3

[0019] a compare processor 5

[0020] a compare instruction memory 4

[0021] a save engine 6

[0022] a bit save unit 7

[0023] a save instruction memory 8

[0024] a stream save unit 9

[0025] an address bus

[0026] When a data stream enters the device of the invention, it is passed through a delay-line 1, preferably a 23 shifts deep and 1 byte wide shift register. As long as a byte resides in the first 16 positions it can be accessed by the compare processor 5, which basically will act as a packet parser. The compare processor 5 is responsible for decoding the packets. It is also connected to a compare instruction memory 4 which inherits the parsing code.

[0027] One characteristic property of the invention is that every incoming byte in the data stream is numbered with a tag. When the compare processor 5 asks for a specific tag the multiplexer control unit 3 delivers the byte located at the right position.

[0028] When the compare processor 5 have come to some kind of conclusion it might want to report something to a result field or an option field, see below. This is done by starting up a save sequence. A start address for a save sequence will be sent from the compare processor 5 to the save engine 6. Said save engine 6 examine the incoming address and decides if it is a save regarding the result field or the option field. According to this decision the address is placed in either a bit save fifo register 61 or a stream save fifo register 62 respectively.

[0029] The bit save unit 7 has three functions; it can set bits in the result field, perform checksum control and length control.

[0030] The stream save unit 9 executes the instruction that saves the option field. Said stream save unit 9 also inserts the result field into the stream and regulates a number of control signals.

[0031] The delayline 1 preferably comprises a 23 shifts deep, 1 byte wide shift register. The 16 first positions of the shift register is reachable from the compare processor 5 through a multiplexer 2. The two last positions are connected to the two save units 7, 9 (bit save and stream save). The stream save unit 9 is actually only using the very last position, and only the bit save unit 7 needs the last two positions because the checksum control works with 16 bits at a time. There are five positions that are prevented from being accessed by the parsing function of the compare procesor 5 and by the save units 7, 9 (bit save and stream save). The reason for a delay before the byte stream arrives to the save units 7, 9 is that all start addresses sent from the compare processor 5 to the save units 7, 9 are queued in a fifo register. Depending on how many save sequences in the queue and how long they are, this might in some extreme situations generate an error. This is because vital data already have passed through the delayline before a save sequence is started. The actual delay needed to secure that no such error occur is 4*64=192 clock cycles. 64 is the maximum length of a save sequence and 4 is the maximum of start addresses waiting to be executed. However, calculus have showed that five delay cycles is enough, since all save sequences normally written are very short.

[0032] A characteristic function of the invention is that it automatically keeps track of where a specific byte has its location in the delayline. The programmer only need to specify which tag, i.e. which number the byte has, where the first byte in a packet is number zero, the second is number 1 and so on. This is why every byte arriving to the delayline 1 should be tagged (numbered). The tagging operating could easily be done by just adding an extra field in every shift in the delayline 1 inheriting the byte's tag. But this is disadvantageous in two aspects. First, much silicon would be used to implement the extra field in the delayline 1. Second, when the parser wants to look at a specific tag it would take a lot of time if every shift had to be searched to find the wanted tag.

[0033] Instead, the present invention has solved the above problem by making a part of the delayline multiplexable; said multiplexable part of the delayline comprises preferably the 16 latest incoming bytes. Worst case for the length of a packet is 1 byte (erroneous), but since the first 12 bytes always contain the OSI Media Access Control address (MAC-address), no useful information can be extracted if the packet is shorter than 13 bytes. These packets will force the compare processor 5 to begin with the next packet at once and their DV (data valid) signal will be unset so the rest of the device or a switch will never see it. With a limit of at least two clock cycles (bytes) between different packets it is possible to guarantee that never more than two packets exist at the same time in the delayline 1.

[0034] According to the ethernet standard the IFG (Inter Frame Gap), which means the distance between packets, is at least 20 cycles, but a smaller distance is always desirable. E.g. a minimum distance of 6 cycles makes it possible to easy extend the device to be able to take care of SONET frames (An alternative ISO-OSI Layer 2 frame instead of ethernet).

[0035] The multiplex control unit 3 uses two identical Tag Units 32, 33 (TU), one for each possible packet, a Controlling Statemachine 31 (CS) to control the TU:s 32, 33 and a TU muliplexer 34 to choose which one of the TU:s 32, 33 that the compare processor 5 is interested in. One TU includes a tagfield register 321, and a lastfield register 322, some adders and a simple statemachine 323. The other TU 33 is identical. When a packet arrives, the tagfield register 321 starts to increment for every byte. When the DV signal becomes false again the tagfield register 321 stops counting and the lastfield register 322 starts to increment. The TU 32 sends an ‘end_of_packet’ signal when the lastfield register 322 reaches the number of shifts in the delayline 1. If the packet was shorter than 13 bytes a ‘too_short’ signal will be generated.

[0036] The position of a requested byte is located according to the expression

p=tagfield+lastfield−wanted — tag

[0037] In the above expression “p” is the position of the wanted byte in the delayline; “tagfield” is the value of the tagfield register (321 or 331); “lastfield” is the value of the lastfield register (322 or 332) and “wanted_tag” is the position of the wanted byte relative to the beginning of the packet.

[0038] A TU 32, 33 also generates a ‘tag_error’ signal if the requested tag never will be available or ‘tag_soon’ if the requested tag has not arrived to the delayline yet.

[0039] The controlling statemachine (CS) 31 is responsible for selecting a free TU (32 or 33) for an arriving packet and to pause the compare processor 5 when no new packets are available. The CS 31 will unselect a TU (32, 33) when the TU generates an ‘end_of_packet’ signal. An unselected TU (32, 33) will be reset to prepare it to receive the next incoming packet. The CS 31 is also controlling the TU multiplexer 34 to change its state every time the compare processor is asking for a new packet.

[0040] A feature of the device according to the present invention is that the compare processor 5 and the compare instruction memory 4 together act as a programmable parser. The description of the full instruction set of said parser is not part of this document, but some instruction types are mentioned below. The parser uses four registers 51, 52, 54, 55 to fulfil it's tasks.

[0041] One PC register 55 that holds the value of the program counter.

[0042] One general register 52. It can be used with instructions for arithmetic operations and for ‘IF_THEN_ELSE’ operations.

[0043] One base register 54. When the parser searches a tag, the value in the base register 54 is added to the searched tag value. This is used to be able to reuse instruction code for e.g. OSI Layer 3 frames, even if they are encapsulated in different OSI Layer 2 frames.

[0044] One stack address register 51 used to store addresses when subroutines are called with ‘JUMP_SUBROUTINE’ type instructions. Accordingly, ‘RETURN’ type instructions copy the stack back to the PC 55.

[0045] All instructions are executed in one clockcycle, except in two cases. This is possible because the compare processor unit receives two instructions every clockcycle from the compare instruction memory 4 which is of the double ported memory type. This features decreases the total amount of clock cycles needed for the compare processor 5 to parse a packet, thereby decreasing the needed size of the delayline. Some instructions are able to start save sequences. Said instructions have a field that tells what address in the save instruction memory 8 that shall start the execution. Save address 0×00 will not generate a start of a save sequence.

[0046] The compare processor 5 must know when a new parsing is started so the registers 51, 52, 54, 55 can be reset. Therefore, when parsing of a packet is done, there shall be a ‘jump_and_save’ instruction with jump adress 0×7f(=last compare instruction memory 4 address). When this is detected it resets and starts looking for a new packet. If the compare processor 5 gets the signal ‘too_short’ it is reset. Further, a ‘tag_soon’ signal pauses the processor 5 and a ‘tag_error’ signal forces it to begin with the next packet.

[0047] The save engine 6 takes the address sent from the compare processor 5 and determine if it is the start address of a bit save sequence or a byte stream sequence. After this the address together with the current value of the base register 54 is put in the specific fifo 61, 62. The value of the base register 54 is needed for all save instructions that is using tag numbers. When the device according to the invention is programmed, a constant is written to the save engine 6 to tell where bit save sequences ends in the save instruction memory 8. This feature exist because it is hard to tell how many instructions are needed to the different parts and it is more expensive to map two memories than one twice as big.

[0048] The bit save unit 7 writes to the result field 76. The result field 76 preferably consists of 24 bits or 3 bytes. It is controlled by the save instruction memory 8 and orders other units to execute the instructions. The executing units are:

[0049] Checksum

[0050] The checksum unit 73 executes the a checksum control instruction which performs a 16-bit one complement addition. The unit needs to know what tag to start the execution from (Tag) and how many bytes the checksum should cover (Length). If there are checksum errors (i.e. the sum differs from 0×FFFF) the unit writes to the result field 76. Further, this block need the value of the base register 54 as it was when the compare processor 5 sent the start address of the current save sequence.

[0051] Bit

[0052] The bit unit 74 executes bit save command which bitwise ‘xor’-ise one selected byte in the result field 76 with the data field. In other words, all bits which are set in the data field will invert the corresponding bit in the result field 76. It is only possible to invert one specific bit one time per packet, this is because e.g. an OSI Layer 3 error could be found in many ways, but if the bit which indicate a Layer 3 error is set an even number of ties, this would look like a correct Layer 3 packet in the result field 76. The address field tells to which one of the total three bytes in the result field 76 to write to.

[0053] Length error

[0054] The length error unit 75 is the most complex unit and investigates lengths in a packet and is used with one or more length control instructions. In a network there might occur packets that has been cut of. This causes many sorts of errors, e.g. if layer 4 is shorter than two bytes the result field 76 should indicate Layer 3 error but not Layer 2 error. The length error unit unit 75 consists of two identical checkboxes and one controller. A checkbox needs to know at which tag to start the measurement from, what kind of comparison it is supposed to perform (more, less, equal or not equal) and what length to match this comparison to. If a checkbox detects a length error, a field which is part of the instruction tells to which one of four possible bits in the results field 76 to write to. As with the checksum unit 73, this unit 75 also needs the value from the base register 54 as it was when the compare processor 5 sent the start address of the current save sequence.

[0055] The stream save executing unit 9 has only one save instruction to handle, a byte stream save instruction. Said instruction is used to save to the option field and includes a start tag number, a length and an six bit wide address to tell were in the option field the selected bytes are to be written. Besides of this it also inserts the result field 76 as soon as all bit save instructions are executed.

[0056] Interface

[0057] The electrical interface of a preferred embodiment of the invention to the outside world is described in conjunction with FIG. 3. It includes an input interface and an output interface. The input interface of the invention includes nine input terminals for a synchronous, eight bit wide, serial data stream, and a data valid (DV) signal, both used by the data that should be decoded. The input interface also includes a programming interface that comprises an 8-bit address bus, an 18-bit data bus, a chip select and a write enable signal for programming the two instruction. These 28 input terminals are used to program the invention after power on.

[0058] The output interface includes output terminals for a serial byte stream together with some control signals. The control signals include a data valid (1 bit), an option field address (6 bits), a store and a halt signal (1 bit each). The store signal tells if the current byte is to be stored in an option field, the halt signal together with the store signal tells if the stream out is the inserted result field. The address bus allows addressing in the option field.

[0059] A typical application for the present invention is for packet switching in a computer network together with a packet switch by extracting information, especially addresses, from the packet headers, because it is possible to test data using several instructions and under several clock cycles even though said data is moving forward in the delayline (1) and even though other bytes of data is entering the device. One of the features of the invention is that the decoding of the protocol is programmable. This is a major advantage because new or different types of protocols can be handled by just reprogramming the device. There will be no need for changing the hardware. This could save time and money for companies responsible for providing, maintaining and updating network switches.

Claims

1. A device for data stream analysing, comprising processor means including a program memory making it possible to parse a data stream in a way that is controlled by an interchangeable program.

2. A device according to claim 1, also including a multiplexable data stream delayline for receiving said data stream, and multiplexing means for connecting different parts of the data stream to said processor means

3. A device according to claim 2, where the multiplexing means include multiplexing control means for automatically keeping track of where specific data is located in the delayline, making it possible to write programs for controlling the device that can start executing at any time after the data have arrived to the device, and without the need for starting execution at a specific time relative to when a data stream was entering the device.

4. A device according to claim 2, where said delayline comprises a 23 shift deep, 1 byte wide shift register.

5. A device according to claim 2, where the multiplexing control means automatically keeps track of where specific data is located in the delayline by the use of position registers (named tagfield and lastfield) that changes according to certain rules when a packet is forwarded in the delay line

6. A device according to claim 4, where the value of the position registers are changed in the following way; when a packet arrives, the tagfield register starts to increment for every byte; when the packet has come to its end, i.e., the packets DV (data valid) signal becomes false again, the tagfield register stops counting and the lastfield register starts to increment.

7. A device according to claim 5, which automatically keeps track of where specific data is located in the delayline, by the use of said dedicated position registers together with the use of the formula

8. A device according to claim 1, including registers for making logical and/or arithmetic operations on data-stream data, before an actual comparison of the data with other data is executed.

9. A device according to claim 1, including stack memory means which enables the writing of programs with subroutines for reducing the need of large program memories.

10. A device according to claim 1, which includes a base address register for the compare processor to make it possible to reuse code to recognise a given pattern even if it start at different positions in a data stream.

11. A device according to claim 1, where the program memory is of double ported type.