JP3644158B2 - Data transmission / reception method in parallel computer - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a data transmission / reception method between PUs in a parallel computer in which a plurality of elemental computers (processing units, hereinafter referred to as PUs) are connected by a communication network, and more particularly to a data transmission / reception method for ensuring high speed of message passing and data safety. .
[0002]
[Prior art]
A parallel computer improves processing speed by connecting multiple PUs via a communication network and operating them simultaneously. In particular, the present invention is directed to a distributed memory type parallel computer in which each PU can access only the memory space associated therewith. A distributed memory type parallel computer cannot directly access data on the memory of another PU. It is necessary to send and receive data each time it is needed and move the data to its own PU.
[0003]
In a distributed memory parallel computer, all data exchange between PUs must be described in a program. Here, data passed between PUs is called a message. This message exchange is called message passing. In the parallel computer program, if the data required by other PUs is in the memory of the own PU, the own PU sends these data to the other PUs in advance, and the data in the other PU memory is automatically transmitted. When a PU requires it, it is necessary for the own PU to explicitly describe in advance an instruction to receive these data from other PUs in the program of each PU.
[0004]
In many parallel computer systems, a function (or subroutine) group called a message passing library is prepared in advance for the purpose of supporting the transmission and reception of messages between PUs, and communication is performed from programs such as C and FORTRAN. It can be described as a function call. Some message passing libraries are implemented on different parallel computer hardware and provide a de facto standard communication environment. Examples include PVM developed at the Oak Ridge National Laboratory in the United States and MPI, which has been standardized in recent years. There is a high possibility (portability) that a parallel program written by calling these communication libraries can be operated only by recompilation on a different parallel computer.
[0005]
To exchange messages between PUs in the communication library, call the send function of the message on the sending PU, call the receive function of the corresponding message on the receiving PU, and send and receive messages between them. Is currently in common use. If the receive function is called before the send function, the receive function will block (stop) until the arrival of data, and if the send function is called first, it will block until the start of the receive function, or the message will be Buffered in. This is called a send / receive method.
[0006]
As a method for realizing high-speed data movement between PUs, there is a distributed memory type parallel computer having a remote memory writing mechanism. In remote memory writing, each PU can directly transfer data to a specific memory area in the partner PU without the intervention of the partner PU. A specific memory area where remote memory writing can be performed is called a remote memory writing area. In the remote memory writing area, real addresses are continuous and are not swapped, so that each PU can transfer data at any time.
[0007]
FIG. 2 shows a conventional method for realizing message passing on a parallel computer having a remote memory writing mechanism. Actual data transfer between each PU is performed between remote memory write areas. First, when the transmission function is called, data is copied from the buffer (201) in the user program on the transmission side to the buffer (202) in the remote memory writing area. Next, data is transferred from there to the buffer (203) in the remote memory writing area on the receiving side using remote memory writing. Finally, the reception function is called, the message is copied from the remote memory writing area (203) to the buffer (204) in the user program, and the message delivery is completed.
[0008]
[Problems to be solved by the invention]
As described above, in remote memory writing, when transferring data without intervention of the partner PU, if the use of the remote memory writing area on the receiving side is not confirmed, the memory area required on the receiving side is overwritten with the write data. As a result, the data on the receiving side may be destroyed. Therefore, unless the remote memory writing area on the receiving side is confirmed, data transfer cannot be continued. Also, data transfer from the user program buffer to the remote memory writing area on the sending side, data transfer from the remote memory area on the sending side to the remote memory area on the receiving side, and from the remote memory writing area on the receiving side to the user program Data transfer to the buffer and a total of three data transfers. Also, it was impossible to guarantee the order between the data sent separately to the remote memory writing area.
[0009]
An object of the present invention is to perform remote memory transfer at a high speed with a limited amount of transfer memory, and to guarantee the order of data sent uniquely by the partner PU.
[0010]
[Means for Solving the Problems]
The present invention relates to a parallel computer comprising a plurality of computers and a communication path interconnecting them, and in each computer, a static reception buffer area which is an area fixed for each computer which is a communication partner, and generation of communication A step of securing a dynamic reception buffer area that is dynamically allocated at times, and a step of writing transmission data in the static buffer area of a transmission destination when the data length of transmission data is shorter than a predetermined value; When the data length of the transmission data is longer than a predetermined value, the step of receiving the address of the dynamic buffer area at the transmission destination using the static buffer area, and the step of receiving the address at the transmission destination This is achieved by providing a step of writing the transmission data in the dynamic buffer area.
[0011]
When data longer than a predetermined value is transferred, the data can be transferred at high speed with a limited amount of memory by transferring the data by pipeline processing.
[0012]
Further, the order of the transmitted data can be ensured by connecting the used transfer memories (buffers) in order.
[0013]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, the details of the present invention will be described with reference to the drawings.
[0014]
First, a specific example of the mounting method of the present invention will be described with reference to the drawings. FIG. 1 shows an overall configuration diagram of the present invention. Reference numerals 101 and 102 denote PUs (processor units), 103 and 104 are their CPUs, and 105 and 106 are memories. Reference numeral 107 denotes a communication path (any network that can interconnect PUs) that connects them. Although the number of PUs is arbitrary, a parallel computer composed of two PUs is shown for explanation. Reference numerals 108 and 109 denote OSs (operating systems). When executing the user program, first, the user program is loaded onto the memory from an auxiliary storage device or the like not shown in FIG. 1 (110, 111). It is assumed that the user program is linked in advance with the message passing library (112, 113) of the present invention. In the message passing library, remote memory write areas (114, 115) in which remote memory can be written from other PUs are provided. Further, in the remote memory writing area, there are a static buffer (116, 117) in which an address is assigned in advance for each communication partner PU and a dynamic buffer (118, 119) in which the address dynamically changes.
[0015]
Among the above components, the message passing library is a component that characterizes the present invention. This will be described in detail below.
[0016]
(A) Buffer configuration
A message passing buffer configuration according to the present invention will be described with reference to FIGS. As described above, the present invention uses two types of buffers, the static buffer (301) and the dynamic buffer (401). The static buffer 301 corresponds to the static buffers 116 and 117 in FIG. 1, and the dynamic buffer 401 corresponds to the dynamic buffers 118 and 119 in FIG.
[0017]
First, the static buffer is provided with a plurality of blocks (6 blocks are shown for PU # 1 in FIG. 3) for each communication partner PU (# 0, # 1, # 2,... #N). Has been. Each block of the static buffer is roughly divided into a header (302) and a message body (303). Further, in the header, tag (304), length (305), first address (306), The information of last address (307) is included. tag is an identifier for selecting the corresponding send / receive pair, length is the length of the message to be communicated, first address is the start address when the dynamic buffer is allocated, and last address is the final address when the dynamic buffer is allocated Is an area for storing. Note that the identification information of the transmission destination PU and the transmission source PU in the communication path (network) is separately managed, and the message is transferred in the network. As the static buffer, buffers of the same shape for transmission (301) and reception (309) are prepared for each PU, and information such as the buffer usage status is shared with the communication partner PU. The static buffer is set in advance for each communication partner PU, and each PU can know the address at the time of initialization. The transmitting side can always transfer data to the receiving buffer on the receiving side as long as there is an empty space in the transmitting buffer.
[0018]
The dynamic buffer includes a plurality of blocks (401) that are not distinguished for each communication partner PU. Each block is divided into a header (402) and a message body (403), and the header is used by the receiving side to check whether data has arrived. The configuration of the header portion area is the same as that of the static buffer, and the identification information (address) in the network of the transmission destination PU and the transmission source PU is separately managed. Further, in order to select the buffer amount according to the message length, some blocks are managed in a bundle (in a plurality of blocks 401 in FIG. 4, this bundle is indicated by a bold frame). Each bundle is registered in the management header (404), and the receiving side PU obtains a buffer bundle of an optimal size according to the message length (in the example of FIG. 4, a bundle of 1 block and a bundle of 4 blocks are respectively obtained. Multiple planes are prepared. When the message is smaller than the size of one block, a bundle of 1 block is obtained. When the message is larger, a bundle of 4 blocks is obtained. Be done ). The reception side PU stores the acquired start address and end address of the buffer bundle in the first address and last address in the header of the static buffer, and transmits them to the transmission side PU. In addition, the transmission side has two dynamic buffers (405), and by using them alternately, pipeline processing is possible. Details of the pipeline processing will be described later.
[0019]
(B) Communication protocol
In general, when the message length is short, message transfer is required to be performed at a higher speed (low latency), while when the message length is long, a larger amount of messages are transferred per unit time (high throughput). Is required. In order to satisfy these two incompatible requirements, the present invention provides two communication protocols, a short protocol used when transmitting a message with a short message length and a long protocol used when transmitting a message with a long message length. However, data transfer with a small delay is realized by switching and using this.
[0020]
FIG. 5 shows a timing chart of the short protocol. The short protocol uses a static buffer for message transfer. When the transmission function is called by the user program (501), the transmission side PU transmits the header of the static buffer and the message body to the reception side (503). On the other hand, when the reception function is called (502), the reception side PU receives the message in the static buffer, and transmits a reception completion notification to the transmission side (504). Message communication can be completed with one round-trip data communication.
[0021]
However, since the remote memory writing area is limited, the length and quantity of the static buffer are limited. For this reason, when all messages are transmitted in a static buffer, when a large number of messages are communicated, it is necessary to wait for the static buffer to be freed, and the communication speed decreases. Therefore, when the message length is long, the message is transferred by using a dynamic buffer that is not distinguished for each PU and can be prepared in large quantities. This is a long protocol. When sending a message with a message length smaller than the static buffer capacity when sending a message with a message length smaller than the static buffer capacity, using the message body length of the static buffer as a boundary, and sending a message with a message length larger than the static buffer capacity It is controlled to use the long protocol.
[0022]
FIG. 6 shows a timing chart of the long protocol. When the transmission function is called by the user program (601), the transmission side PU first detects the length of the data length to be transmitted (data amount of data to be transmitted), and this data amount is larger than the capacity of the static buffer. It is determined that the long protocol is used. When the amount of data to be transmitted is smaller than the capacity of the static buffer, the aforementioned short protocol is used. In the case of the long protocol, the header of the static buffer is transmitted to the receiving side (603). On the other hand, when the reception function is called (602), the receiving side PU receives the message information in the static buffer, and sends back the dynamic buffer address information corresponding to the information using the static buffer (604). A message is transmitted based on the buffer information received by the PU (605), and finally the receiving PU transmits a reception completion notification to the transmitting side (606). Although two round-trip data transfer is required and the latency is higher than that of the short protocol, the dynamic buffer can transfer a large amount of data compared to the static buffer, so that the throughput can be increased.
[0023]
Hereinafter, the operation of each protocol will be described in detail.
First, the operation of the short protocol will be described with reference to FIG. The sending PU stores the message length and identifier in the header of the static buffer (701). Further, the message is copied from the user area to the message body (702). Next, the remote buffer is written into the static buffer from the transmission side to the reception side (703). On the other hand, the receiving PU receives the message in the static buffer (705), copies it to the user area, and completes the reception (706). Finally, the reception side sends out a reception completion notification using a static buffer (707), and upon receipt thereof, the transmission side also ends the processing (704).
[0024]
Next, the operation of the long protocol will be described with reference to FIG. First, the transmission side PU stores the message length and the identifier in the header of the static buffer as in the short protocol (801). In the case of the long protocol, the message cannot be sent in the static buffer, so only the header is written in the remote memory to the receiving side (802). When the receiving PU receives the message information in the static buffer (807), it secures a dynamic buffer having an appropriate length according to the length and obtains its start address and final address (808). These addresses are set in the first address and last address of the static buffer and sent back to the transmission side (809). The transmission side receives the buffer information (803), repeats the loop until all the messages are transmitted (804), and copies and transmits the dynamic buffer block from the user area to the buffer (805). Further, the receiving side repeats the loop until all messages are received (810), and copies the message from the received block to the user area (811). Finally, the receiving side sends out a reception completion notification using a static buffer (812), and the transmitting side also ends the processing (806).
[0025]
(C) Pipeline transfer in long protocol
In message passing using remote memory writing, the message is transferred three times: copying from the user area to the remote memory writing area, remote message transfer from the sending side to the receiving side, copying from the remote memory writing area to the user area, and so on. Necessary. Therefore, at least about three times the time required for remote memory writing is required. Therefore, in the present invention, two dynamic buffers on the transmission side are prepared and performance is improved by performing pipeline processing.
[0026]
The operation during pipeline processing will be described below with reference to FIG. In FIG. 9, the buffer in the remote memory writing area of the sending PU ( 202 ) Represents the dynamic buffer (405) on the transmission side in two sides in FIG. 4, and the buffer (203) in the remote memory writing area of the reception side PU is simplified to represent the dynamic buffer (401) on the reception side. ing. Although there are many types of dynamic buffers on the receiving side, only the four sides necessary for transferring ABCD data are shown here. In step 1, message A is copied from the user area to the remote memory writing area on the transmission side (901). Next, in step 2, message A is transmitted from the transmission side to the reception side by remote message transfer (902), and at the same time, message B is copied to the remote memory writing area (903). In step 3, message A is copied from the remote memory writing area to the user area on the receiving side (904), message B is transferred from the transmitting side to the receiving side in remote memory (905), and message C is transferred from the user area on the transmitting side. Copy to the remote memory writing area (906). In step 4, message B is copied (907) on the receiving side, remote memory write transfer (908) from the transmitting side to the receiving side of message C, and message D is copied (909) on the transmitting side at the same time.
[0027]
FIG. 10 shows operations for each of the transmission side PU and the reception side PU in the pipeline processing. On the transmitting side, first, data A is copied from the user area to the remote memory writing area (1001), then data A is transmitted to the receiving side, and at the same time, data B is copied (1002). (1003, 1004), and finally data D is transmitted (1005). On the other hand, the receiving side first receives data A from the transmitting side (1006), then receives data B, and at the same time performs a memory copy of data A (1007), and thereafter the same operation continues (1008, 1009). Finally, memory copy of data D is performed (1010).
[0028]
(D) Message passing library interface
The message passing library is a group of functions for performing message passing in the form of function calls from the user program. The interface of the message passing library function and its operation in the present invention will be described below. Note that the function name, argument name, and the like are arbitrary and do not necessarily have the same specifications as described here.
[0029]
(1) Init ()
In this function, the message passing library performs necessary initialization operations. When using the message passing library, all PUs must always call this function first. When this function is called from the user program, a static buffer (FIG. 1: 116, 117) and a dynamic buffer (FIG. 1: 118, 119) are created in the remote memory write area, and each buffer is initialized. . The static buffer has a fixed address for each PU, and all PUs can communicate with each other at the time of initialization. The functions listed below can only be used after calling the initialization function.
[0030]
(2) Send (buf, dest, tag, length)
Where buf is the start address of the user memory in which the message to be transmitted is stored, dest is the destination PU number (information identifying the destination PU in the network), tag is the identifier of the message, and length is the length of the message Represents. When the user calls this function, the library transmits a message using the short protocol (FIG. 7: 701 to 704) or the long protocol (FIG. 8: 801 to 806) depending on the length.
[0031]
(3) Recv (buf, src, tag)
In this function, buf is the start address of the user memory that stores the received message, src is the source PU number (information identifying the PU in the network), and tag is the message identifier. When the user calls this function, the length of the message is received in the header portion of the static buffer with the matching transmission source PU number, and the short protocol (FIG. 7: 705 to 707) or the long protocol (FIG. 8: 807) is matched accordingly. ˜812) to receive the message.
[0032]
(E) Guarantee of order in non-blocking operation
There are a blocking function and a non-blocking function for transmission and reception in the message passing library. The blocking function is a function that blocks (stops) the operation of the program until the transmission / reception is completed after the transmission / reception function is called.The non-blocking function is returned before the transmission / reception is completed after the function is called. It is a function that can perform other operations in the meantime. Up to the previous section, a blocking function was assumed. This section describes an additional mechanism for realizing non-blocking. In the non-blocking function, a plurality of transmission / reception functions may be issued before transmission / reception is completed. At this time, in the transmission / reception function with the same tag, the transmission function and the reception function may not correspond in the order in which they are issued. The method for guaranteeing the order of transmission / reception functions according to the present invention will be described below with reference to FIG.
[0033]
When transmitting, the message passing library of the present invention first sets message information (tag, length) in the header of the static buffer and communicates with the receiving side (when the protocol is short, the message body is also transmitted simultaneously). At this time, a member next (1101) is added to the header of the static buffer, and the next block of the static buffer to be transmitted is designated by this next. Each block of the static buffer in use is chained by next, and is transmitted in the order of the chain. The static buffer received by the receiving side is sent from the transmitting side, and is connected by next like the transmitting side. Therefore, it is possible to search in the order of the chain and to determine the order in which the transmission functions are issued.
[0034]
If the static buffer is not empty when the sending function is issued on the sending side, or the header of the static buffer is not yet sent when the receiving function is issued on the receiving side, the order of the sending / receiving functions is changed to the static buffer. It cannot be expressed in the order of the blocks. Therefore, in the present invention, in order to maintain the order of functions that cannot be processed immediately upon function issuance, a request object for managing the order of issuing unprocessed functions is introduced (1102, 1103). The request object has a total of four elements: tag (1104, 1105) holding message information, dest (src) (1106, 1107), length (1108, 1109) and next (1110, 1111) holding the order. have. Unfinished functions are chained by next, as in the static buffer block, and are processed in that order on the sending and receiving sides. After the blocks of the static buffer are processed in order, the processing proceeds in the order of request objects. The ordering in the non-blocking function is guaranteed by the above method.
[0035]
(F) Additional interface for non-blocking operation
If ordering is guaranteed, a non-blocking function can be realized by adding the following interface.
[0036]
(1) Isend (buf, dest, tag, length)
Each argument has the same specifications as Send. When the user calls this function, if there is space in the static buffer, the buffer to be used is transferred after being chained. If there is no space in the static buffer, a request object is created and this function is chained. connect. Request object chains are processed in order of issue. After the static buffer is transferred to the receiving side, data is transferred asynchronously using the short protocol (FIG. 7: 701 to 704) or the long protocol (FIG. 8: 801 to 806).
[0037]
(2) Irecv (buf, src, tag)
Each argument has the same specifications as Recv. When the user calls the function, a request object is created and connected to the chain. It is processed in order from the top function of the chain, and it is searched whether a transmission function having a tag corresponding to the static buffer chain from the top is issued. After the corresponding transmission is searched, data is transferred asynchronously using the short protocol (FIG. 7: 704 to 707) or the long protocol (FIG. 8: 807 to 812).
[0038]
(3) Wait ()
This function is a function for waiting for completion of the non-blocking function. Non-blocking functions return immediately after a function call, so the user does not know when the function completes. Therefore, this function is used to clearly indicate the completion of the non-blocking function. When this function is issued, the PU is blocked until a non-blocking function that has not been confirmed for completion is completed. This function is also completed when all functions are completed. After this function is completed, Send / Isend and Recv / Irecv are newly issued and communication is resumed.
[0039]
【The invention's effect】
According to the remote memory transfer control method of the present invention, in data transfer using remote memory writing, data is transferred using an area to which an address is assigned in advance for each PU according to the length of data, or dynamically assigned at the time of transfer. It is possible to select whether data is transferred by pipeline processing using the area to be transferred, thereby enabling high-speed data transfer. When the pipeline operation is introduced as shown in FIG. 12, three message transfers are overlapped except for the first and last two times. Therefore, about three times the performance can be obtained as compared with the message passing using the conventional remote memory writing.
[0040]
Further, according to the remote memory transfer control system of the present invention, it is possible to ensure that data is transferred in the order in which the transmission functions are issued and that the transferred data is received in the order in which the reception functions are issued. This makes it possible to guarantee the order of data in the transmission / reception function that performs the non-blocking operation.
[Brief description of the drawings]
FIG. 1 is an overall configuration diagram of an embodiment of the present invention.
FIG. 2 is a data transfer control method using conventional remote memory writing.
FIG. 3 is an explanatory diagram of a static buffer.
FIG. 4 is an explanatory diagram of a dynamic buffer.
FIG. 5 is a timing chart of a short protocol.
FIG. 6 is a timing chart of a long protocol.
FIG. 7 is a flowchart of a short protocol.
FIG. 8 is a flowchart of a long protocol.
FIG. 9 is an explanatory diagram of pipeline operation.
FIG. 10 is a flowchart of pipeline operation.
FIG. 11 is an explanatory diagram for guaranteeing order.
FIG. 12 is an operation diagram of pipeline operation.
[Explanation of symbols]
101, 102 ... Element computer, 103, 104 ... CPU, 105, 106 ... Memory, 107 ... Communication path, 108, 109 ... Operating system, 110, 111 ... User program, 112,113 ... Message communication library, 114,115 ... Memory area for data transfer, 116,117,301,309 ... Static buffer, 118,119,401,405,910,911 ... Dynamic buffer, 201,204 ... User buffer, 202,203 ... Data transfer buffer, 302 ... Static buffer header, 303,308 ... Static buffer message body, 304 .. An area for storing the identifier of the message, 305... An area for storing the message length, 306... An area for storing the start address of the reserved dynamic buffer, 307. Store the address Area 402: header for confirming whether a message has arrived in the dynamic buffer 403: body of the message in the dynamic buffer 404: management header for the dynamic buffer 1101 ... Area for storing the order of static buffers, 1102, 1103... Request object, 1104, 1105... Request object area for storing message identifier, 1106, 1107... Object area, 1108, 1109... Request object area for storing message length, 1110, 1111... Area for storing the order of request objects.

Claims (4)

A data transfer method between a plurality of computers and a parallel computer comprising a communication path interconnecting them,
In each of the computers, a static reception buffer area that is fixed for each communication partner computer and a dynamic reception buffer area that is not fixed for each communication partner computer are secured in advance. And steps to
Is shorter than the data length of transmission data is predetermined value, and performing data transfers by writing transmission data to the static buffer area of the destination by the remote memory write from a source computer,
When the data length of the transmission data is longer than a predetermined value, the step of transmitting information including the data length of the transmission data from the transmission source computer to the transmission destination computer, and at the transmission destination computer, Determining a dynamic buffer area to receive the transmission data according to the information, and returning the address of the determined dynamic buffer area to the transmission source computer by remote writing using the static buffer area; A data transmission / reception method in a parallel computer, comprising the step of writing the transmission data in a dynamic buffer area indicated by a returned address by remote writing from the transmission source computer . A remote memory writing mechanism that connects multiple element computers via a communication path and writes data in the data transfer memory area on any element computer to the data transfer memory area on any other element computer A data transmission / reception method between two arbitrary element computers in a parallel computer having:
Each element computer is triggered by a communication initialization request from a user program, and communication occurs in the data transfer memory area with a static receive buffer area whose address is fixed in advance for each computer as a communication partner. Reserving a dynamic receive buffer area from which buffers are dynamically allocated each time,
In response to a transmission request from the user program, the transmitting side element computer copies the transmission data on the user memory to the data transfer memory area, configures a header including information on the data length, When the data length is shorter than a predetermined value, the header and the data are written to the static buffer area prepared on the element computer on the receiving side using a remote memory writing mechanism, and the data length of the transmission data Is longer than a predetermined value, the header is written to the static buffer area prepared on the element computer on the receiving side using a remote memory writing mechanism;
The element computer on the receiving side refers to the header when the header arrives and acquires the data length. If the data length of the transmission data is longer than a predetermined value, the data length is necessary. Securing a buffer on the dynamic buffer area and notifying address information of the buffer to the element computer on the transmission side;
When the data length of the transmission data is longer than a predetermined value, the transmission-side element computer refers to the address information when the address information arrives, and sends the data to the reception-side element computer. Writing to a reserved dynamic buffer using a remote memory writing mechanism;
The reception-side element computer, when triggered by a reception request from the user program, copies the data written in the static buffer area or the dynamic buffer area to the reception buffer on the user memory. Transmission / reception method. When multiple transmission requests are issued from the user program, the address information of the static reception buffer used at the time of the next transmission request is stored in the header information, thereby guaranteeing the order of issuing the plurality of transmission requests. The data transmission / reception method according to claim 2, further comprising steps . Between two arbitrary element computers in a parallel computer having a remote memory writing mechanism in which a plurality of element computers are connected by a communication path and data on an arbitrary element computer is written to the memory of another arbitrary element computer A data transmission / reception method,
Each element computer reserves on the data transfer memory area a dynamic reception buffer area to which buffers are dynamically allocated each time communication occurs, and two transmission buffers. Configuring the dynamic reception buffer area to be composed of a plurality of predetermined fixed-length blocks;
The transmission-side element computer divides the transmission data into a plurality of predetermined fixed-length packets triggered by a transmission request from the user program, configures a header including information on the length of the data, A step of notifying the receiving side element computer of the header; and upon receiving the header information, the receiving side element computer refers to the header information and sets a reception buffer including the necessary number of blocks. Securing on the reception buffer area and notifying address information of the buffer to the element computer on the transmission side;
The transmission side element computer, upon the arrival of the address information of the buffer, copies the nth packet of the plurality of packets to one of the two transmission buffers on the data transfer memory area, and The (n−1) th packet copied to the other of the two transmission buffers on the data transfer memory area is transferred to the dynamic reception buffer n on the element computer on the reception side by using a remote memory writing mechanism. -Writing to the first block and applying the above two steps sequentially for all the plurality of packets;
A data transmission / reception method including a step in which a reception side element computer sequentially copies the data written in a plurality of blocks of the reception buffer to a reception buffer on a user memory in response to a reception request from a user program.

Images