This application is based upon and claims the benefit of priority from the prior Japanese Patent Application No. 2005-85307 filed on Mar. 24, 2005 and the prior Japanese Patent Application No. 2005-345822 filed on Nov. 30, 2005, the entire contents of which are incorporated herein by reference.
1. Field of the Invention
The present invention relates to a communication system and communication method thereof, with PCI-Express used in an interface, and with a plurality of nodes connected through a PCI-Express.
2. Description of the Related Art
Peripheral Component Interconnect (PCI) is a standard for computer bus connecting between each part inside computers, and is currently applied to most of computers; however, specification of PCI-Express, which is faster transfer interface standards, is newly developed.
For example, as described in Patent Document 1, in the PCI-Express, in order to prevent skews between signal lines, which occur in a parallel bus, a serial link is applied, and packets are transferred asynchronously through the serial link. Also, the PCI-Express allows upgrade in units of layers, and therefore functions thereof are layered. By adding a header in a transaction layer to transmission data produced in number and a CRC are added to the head and the tail of the transaction packet, generating a data link packet. Finally, in the physical layer, it is converted to physical signals and transmitted through a transmission media.
Meanwhile, in the PCI-Express, between nodes can be connected by a switch (PCI-Express switch).
In the following description, a part of the PCI-Express relating to the present invention is further explained.
According to the PCI-Express specification, a module, called a root complex, which is located on top of the upper layer, such as a module A(1) in
Next, packet communications between modules via the PCI-Express switch (6) is explained.
There are two types of packet routing on top of the PCI-Express switch (6); address routing for routing according to an address comprised in a packet header, and ID routing for routing according to a requester ID constituted from device and action numbers, comprised in a packet header.
These two types of routing, in the PCI-Express specification, are separately used so that address routing is used for a request packet, and the ID routing is used for a response packet. Therefore, the address routing is used for a request packet for a memory write request command and a memory read request command. Note that a response by a response packet is not made to the memory write request command in this PCI-Express specification.
As shown in
For example, assume a bus number and a device number are set in each port of the PCI-Express switch (6) as described in
As shown in
In
[Patent Document 1]
Japanese Published Patent Application No. 2004-326151
Problems to be solved by the present invention is to be able to transmit a response packet to a target, which is the original request source node, even if, after issuing request from a node to another, a bus number/a device number is replaced in the PCI-Express switch before said another node makes a response to the request source node in a PCI-Express communication system, which performs communication between nodes using a PCI-Express switch.
In order to solve the problem in the present invention, a unique node ID is set to each module connected by the PCI-Express. Dedicated channels are provided for each of writing data in a module at the other end (PUT transfer), and reading out data from a module at the other end (GET transfer), and a channel ID is set to each channel. The node ID and the channel ID are set in the address of a header of a transferred transaction layer packet according to the PCI-Express, and at the same time, a packet type for identifying whether the packet is a request or a response is set.
For data transfer, only the memory write request packet, being routed according to address routing of the PCI-Express specification, is used.
When a packet is request, the response is always made by a response packet (a response-type memory write request packet).
According to the present invention, because transfer means does not employ the ID routing, even if replacement of a bus number and a device number are caused in the PCI-Express switch, a response can be securely made to a module that issued a request.
A module in the receiving side analyzes the transmitted packet header, determines whether the packet is PUT transfer or GET transfer from the channel ID and whether the packet is a request or a response from the packet type, and performs a proper operation.
It is also possible to determine whether the memory write request command is executed in a normal manner or not by adding information to a response-type memory write request packet.
In the following description, details of the preferred embodiment of the present invention are set forthwith reference to
An example in the PCI-Express communication system (100) of the present invention described in
In the present invention, only the memory write request packet routed in accordance with the address routing of the PCI-Express specification is used for data transfer. Therefore, even if it is semantically a memory read request, a memory write request packet is used.
In the present invention, a unique node ID is set for each module connected to the PCI-Express. In an example of
In addition, dedicated channels are provided for PUT transfers for writing data to a module at the other end, and for GET transfers for reading out data from a module at the other end, and a channel ID is set for each channel.
As shown in
A channel ID of the packet transfer source is set to the tag field, and when the transfer destination responds with a response packet, the tag field is used as a channel ID of the response destination, in an address field of the response packet.
In the following description, a GET transfer sequence of the present invention for executing a memory read request is explained with reference to
This sequence is the same as that of the conventional example shown in
The receiving side identifies the packet as a request packet from the packet type, and determines that the GET transfer reception channel is used from the channel ID, identifying the packet as a GET transfer. Subsequent, steps of reading out data from memory, and transmitting read out data are the same as those of the conventional example described in
The transmitting side identifies that the packet type is a response and that the packet is a response to a GET request from the GET transfer transmitting channel ID included in the address field, and terminates the process after writing the read data to memory.
Next, a PUT transfer sequence of the present invention for performing a memory write request is explained with reference to
This sequence is the same as that of the conventional example shown in
The receiving side identifies the packet as a request packet from the packet type, and determines that PUT transfer reception channel is used from the channel ID, identifying the packet as a PUT transfer. Then the receiving side writes the transmitted data to memory.
Up until this step, the sequence is the same as that of a conventional PUT transfer; however, in the present invention, a function to respond with a response packet using a memory write request packet is incorporated into the hardware, for example. Because the PCI-Express switch replies to the transmitting side with a memory write request packet of which a packet type is set as a response, it identifies the packet as a request packet, and transfers the packet to the transmitting side while routing a packet according to the address routing of the PCI-Express specification. However, because the address field of the received packet contains a PUT transfer transmitting channel ID, the transmitting side can identify the packet as a response to the memory write request made by its own node.
Next, the embodiments of the present invention are explained with reference to
In the present embodiments, a DMA chip in a central control module CM realizes the communication method of the present invention in a PCI-Express communication system (200) in which up to eight central control modules CM for controlling all operations of each subsystem perform intercommunication via a PCI-Express switch.
The system configuration of the present embodiment shown in
Node ID=0 through 7 are assigned in order from CM#0 to central control module nodes. A central control module comprises, as shown in
The front end router FRT#0 (71) and FRT#1 (72) constitutes duplicated front end routers. Each of the front end routers FRT#0(71) and FRT#1(72), and each of the central control modules CM#0-CM#7 is connected by four PCI-Express 2.5 GHz lines as shown in
As shown in
There are two channels of the message-receiving channel (331), eight channels of the message transmitting channel (332), five channels of the PUT transmitting channel (342), four channels of the GET transmitting channel (352), and one channel is provided for each of the PUT receiving channel (341) and the GET receiving channel (351).
The transfer buffer (361) is used for data transfer from the south port side to the north port side, and the transfer buffer (362) is used for data transfer from the north port side to the south port side. The transfer buffer (370) is used for loop back from the north port side.
A solid line represents a control line (REQ), which indicates a direction of a request notification, a dotted line represents a control line (CMP), which indicates a direction of a request completion notification, and a bold line represents a data line. An outlined arrow represents a PCI-header line, and lines TX and RX connected to the PCI-Express interface control units (311 and 312) represent PCI-Express transmission and reception lines.
In the example shown in
In the following description, an example of data transfer in the present embodiments is explained. Some descriptions are the same as the description in
(1) The CPU of the central control module CM#0 instructs the PUT transmitting channel of a DMA chip to start a PUT transfer. (2) The DMA chip of the central control module CM#0 converts the bit [62:85] (the node ID field) in the address field of the transmitting packet header to 7, which is the node ID of the central control module CM#7, transfers data to the south port side, and then, transmits it as a memory write request packet. (3) The PCI-Express switch analyzes a transmitting destination address, and transfers a packet to the corresponding address. (4) The DMA chip of the central control module CM#7 receives the packet from the south port side, confirms the correspondence between the node ID in the packet and the its own node ID, and informs the PUT receiving channel of the reception of the packet.
Next, with reference to
(1) The CPU creates a descriptor for a PUT transfer, and writes it to a descriptor area in the memory. Fifteen 32-byte areas are provided in the example shown in
Next, the operation of the receiving channel in the above PUT transfer example is explained with reference to
(1) The CPU activates the PUT receiving channel, and instructs the DMA chip to start the transfer. (Only performed the first time) (2) The DMA chip receives data from the south port side. (3) The DMA chip transfers the received data to the north port side and writes the data to the memory. (4) The DMA chip returns a response packet to the south port side.
(1) The CPU of the central control module CM#0 instructs the GET transmitting channel of the DMA chip to start the GET transfer. (2) The DMA chip of the central control module CM#0 converts the bit [62:85] (the node ID field) in the address field of the transmitting packet header to 7, which is the node ID of the central control module CM#7, transfers data to the south port side, and then, transmits it as a memory write request packet. (3) The PCI-Express switch analyzes a transmitting destination address, and transfers a packet to the corresponding address. (4) The DMA chip of the central control module CM#7 receives the packet from the south port side, confirms the correspondence between the node ID in the packet and its own node ID, and informs the GET receiving channel of the confirmation. (5) The DMA chip of the central control module CM#7 reads out data from the memory of its own central control module, transfers the data to the south port side, and transmits the data as a memory write request packet. (6) The PCI-Express switch analyzes the transmitting destination address, and transfers the packet to the corresponding address. (7) The GET transmitting channel in the DMA chip of the central control module CM#0 receives GET data from the south port side, and transfers the data to the north port side.
Next, the operation of the transmitting side channel in the above GET transfer example is explained with reference to
(1) The CPU creates a descriptor for a GET transfer, and writes it to a descriptor area in the memory. (2) The CPU instructs the DMA chip to start the GET transfer. (3) The DMA chip reads out descriptor data from the descriptor area in the memory. (4) The DMA chip analyzes the descriptor and makes a GET data request to the DMA chip at the other end. (5) The DMA chip receives the GET data from the south port side, and transfers the data to the north port. (6) The DMA chip updates a completion pointer of the descriptor, and writes a pointer value (a descriptor termination pointer) to a descriptor termination pointer area in the memory. (7) The DMA chip interrupts the CPU in order to notify the CPU of the transfer termination. (8) The CPU releases the interruption and clears a termination code as a fixing operation of the termination pointer.
Next, the operation of the receiving channel in the above PUT transfer example is explained with reference to
(1) The CPU activates the GET receiving channel, and instructs the DMA chip to start the transfer. (Only performed the first time) (2) The DMA chip receives a GET request from the south port side. (3) The DMA chip reads data from the memory of its own central control module CM. (4) The DMA chip transfers the GET data to the south port side.
As explained above, in the PCI-Express communication system of the present invention, only a memory write request packet routed according to the address routing of the PCI-Express specification is employed for data transfer, and therefore, even in an occurrence of ID replacement due to a route complex failure, a response can be returned in a normal manner.
Also, a response packet can be returned regardless of the operations of the PCI-Express switch or the command types, a command issue source module can securely acquire information regarding whether the data transfer terminated in a normal manner or not.
Number | Date | Country | Kind |
---|---|---|---|
2005-085307 | Mar 2005 | JP | national |
2005-345822 | Nov 2005 | JP | national |
Number | Name | Date | Kind |
---|---|---|---|
6874042 | Sauber | Mar 2005 | B2 |
6928126 | Ono et al. | Aug 2005 | B2 |
6971098 | Khare et al. | Nov 2005 | B2 |
7058738 | Stufflebeam, Jr. | Jun 2006 | B2 |
7165131 | Creta et al. | Jan 2007 | B2 |
7191255 | Wong et al. | Mar 2007 | B2 |
7293127 | Caruk | Nov 2007 | B2 |
7340555 | Ashmore et al. | Mar 2008 | B2 |
7380018 | Moll | May 2008 | B2 |
20040210754 | Barron et al. | Oct 2004 | A1 |
20050034045 | Lueck et al. | Feb 2005 | A1 |
20050289306 | Muthrasanallur et al. | Dec 2005 | A1 |
20060179195 | Sharma et al. | Aug 2006 | A1 |
20060195644 | Arndt et al. | Aug 2006 | A1 |
20060206655 | Chappell et al. | Sep 2006 | A1 |
20070130397 | Tsu | Jun 2007 | A1 |
20080055808 | Burkland et al. | Mar 2008 | A1 |
Number | Date | Country |
---|---|---|
2001-142845 | May 2001 | JP |
2004-326151 | Nov 2004 | JP |
Number | Date | Country | |
---|---|---|---|
20060218336 A1 | Sep 2006 | US |