JP3698031B2 - Remote access method to secondary storage device - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a method for remote access to a secondary storage device, and more specifically, in an environment where a plurality of computers are connected by an inter-computer connection network, an application program running on one computer is connected to another computer. The present invention relates to a method for accessing data on a secondary storage device.
[0002]
[Prior art]
There are several ways to access a secondary storage device (remote secondary storage device) connected to another computer.
[0003]
One method is a method based on the Network File System (NFS) as shown in FIG. Data access in NFS is realized by software by providing a communication protocol layer called RPC (reference document “NFS & NIS” ASCII Publishing Bureau).
[0004]
In NFS, a read / write call from a client program is converted into an RPC request and transmitted to a process called an NFS daemon running on a server computer using the RPC protocol. For example, in the case of a write call, write data is transmitted together with the request, but this data is captured by the NFS daemon, then written to the secondary storage device, and the response is transferred to the client program. If it is a read call, the NFS daemon reads the target data from the secondary storage device and returns the data as a response to the RPC request.
[0005]
When a read / write call from a client program is converted into an RPC request, it is broken down into a plurality of RPC requests as read / write requests for fixed-size data. Normally, the size of data accessed by a single RPC request is 8 Kbytes. Access requests are sent to the NFS daemon one by one, and after waiting for a response, the next RPC request is sent. In other words, since the NFS daemon processes only one RPC request at the same time, it is only necessary to secure an area for capturing 8 Kbytes of data. However, since one NFS daemon is occupied for one client program, it is necessary to start a plurality of NFS daemons on the size calculator in order to process requests from a plurality of client programs.
[0006]
As another method, for example, there is a high-speed LAN disk server as disclosed in JP-A-4-92943. In this method, as shown in FIG. 17B, an inter-computer connection network is a high-speed LAN, and an environment comprising a disk server for managing a secondary storage device (limited to a disk) and a plurality of workstations. Is assumed.
The disk server that manages the disk is dedicated hardware, and has a structure that can process a plurality of read / write requests simultaneously.
[0007]
A client program runs on the workstation, and a read / write request to the remote disk from the client program is transmitted to the disk server by communication. For example, if it is a write request, data is transmitted together with the request. On the other hand, the disk server temporarily stores the received data in the data buffer, writes the data to the disk, and notifies the result to the requesting workstation through communication. If it is a read request, the disk server checks whether there is target data in the data buffer. If not, the data is read from the disk and stored in the data buffer. Further, the stored data is transmitted to the requesting workstation by communication.
[0008]
In this method, a data buffer is provided in the disk server in order to relay data between the inter-computer connection network and the secondary storage device. The data buffer is used by dividing an area for each workstation. A read / write request is made in a fixed length unit called a frame.
[0009]
Both of the above-described two methods decompose the read / write call from the client program as a fragmented read / write request to a plurality of fixed length data, and manage the secondary storage device (server computer , Disk server), it is possible to process read / write calls to the remote secondary storage device.
[0010]
[Problems to be solved by the invention]
However, since each fragment read / write request that has been disassembled is transmitted one by one, and after waiting for the processing result, the next fragment read / write request is transmitted, the communication is performed many times. The flow of processing for one read / write call is as shown in FIG. 17C, and (fragment read / write request transmission) → (secondary storage device read / write) → (response return) → (fragment read) / Write request transmission) →... Is processed, and after all fragment read / write requests are completed, the result of the client program read / write call is returned. Therefore, it seems that the time required for accessing the remote secondary storage device from the viewpoint of the client program takes an extra round-trip communication time corresponding to the number of fragment read / write requests.
[0011]
Therefore, in the present invention, most of the communication time in the data access time to the remote secondary storage device is hidden from the client program, and the access is as if the secondary storage device is connected to the computer on which the client program operates. Aim to achieve time.
[0012]
[Means for Solving the Problems]
In order to achieve the above object, according to the present invention, a client program operating on a first computer is connected to a second computer that can communicate with the first computer through a connection network between computers. In a method of accessing data on a secondary storage device, the first computer and the second computer (3) Any one of the processes (4) to (4) is executed.
[0013]
(1) The first computer divides a read / write call by a client program into read / write requests for data of a certain size, and the read / write requests obtained by the division are all at once described above. After confirming that responses are returned from the second computer for all the read / write requests transmitted, the result of the read / write call is returned to the client program. On the other hand, the second computer receives the read / write request from the first computer, accesses the requested data on the secondary storage device for one received read / write request, and then receives the request source. The process of sending one response to the client program is multiplexed and executed for a plurality of read / write requests.
[0014]
In this way, a read / write call by a client program is divided into read / write requests for data of a certain size and transmitted to the second computer all at once, and a plurality of read / write requests are multiplexed in the second computer. By performing the processing, the physical read / write processing of the secondary storage device in the second computer and the communication processing in the inter-computer connection network are performed in parallel, and most of the communication time is the physical read / write. Overlapping with the processing time, the entire processing time is shortened.
[0015]
(2) The first computer divides the read / write call by the client program into read / write requests for data of a certain size, and the number of read / write requests obtained by the division is predetermined. When the threshold value is less than the threshold value, all are transmitted to the second computer at once. When the threshold value is exceeded, first the read / write request for the threshold value is transmitted to the second computer, and then the read / write is performed. Each time one response from the second computer corresponding to the request is confirmed, one remaining read / write request is transmitted to the second computer, and the second read / write request is transmitted for all the transmitted read / write requests. After confirming that a response is returned from the computer, the result of the read / write call is returned to the client program. The second computer executes the same processing as (1) above.
[0016]
In this way, the threshold value is set as the number of requests that can be received at one time by the second computer, and the number of read / write requests to be transmitted first is kept below the threshold value. The load can be kept below a certain value. Furthermore, since one read / write request is transmitted in response to one response, the number of read / write requests received by the second computer is always less than the threshold value, and the load on the second computer continues to be a constant value. The following can be suppressed.
[0017]
(3) The first computer divides the read / write call by the client program into read / write requests for data of a certain size, and transmits the read / write request obtained by the division prior to the transmission of the read / write request. A read / write request is transmitted to the second computer, a threshold value is determined from the number of reservations included in the response from the second computer to the transmitted reservation request, and the division / read obtained When the number of requests is less than or equal to the threshold value, all of the requests are transmitted to the second computer at once. When the number of requests exceeds the threshold value, a read / write request corresponding to the threshold value is first transmitted to the second computer. Thereafter, every time one response from the second computer corresponding to the read / write request is confirmed, one remaining read / write request is sent to the second computer. And confirming that responses have been returned from the second computer for all the read / write requests transmitted, requesting the second computer to cancel the reservation, and sending the request to the client program Returns the result of a read / write call. On the other hand, in response to a reservation request from the first computer, the second computer uses the number of read / write requests allowed for one client program according to the current load status of the own computer as the reservation number. 1 is notified to the computer and a memory amount corresponding to the number of reservations is secured, a read / write request is received from the first computer, and the requested secondary storage device is received for one received read / write request. After accessing the above data, a process of sending one response to the requesting client program is executed in a multiplexed manner for a plurality of read / write requests, and a reservation cancellation request from the first computer On the other hand, the reserved memory amount is released.
[0018]
In this way, the first computer is notified of the number of reservations allowed for one client program in accordance with the dynamically changing load status of the second computer, and the first computer displays the number of reservations as a threshold value. Thereafter, by making a read / write request according to this threshold value, it is possible to control so that the load generated in the second computer does not exceed a certain value. Therefore, even if the second computer is a partner of a plurality of client programs, the processing capacity of the second computer is not exceeded.
[0019]
(4) The first computer divides the read / write call by the client program into read / write requests for data of a certain size, and reads / writes one of the read / write requests obtained by the division. Reservation with a reservation request added to a write request + read / write batch request is transmitted to the second computer, and included in the reservation response from the second computer with respect to the transmitted reservation + read / write batch request A threshold value is determined from the number of reservations to be reserved, and when the remaining number of read / write requests obtained by the division is equal to or less than the threshold value, a reservation release request is added to one read / write request and reserved. After making a cancellation + read / write batch request, send all to the second computer all at once, and if the threshold value is exceeded, the threshold is set first. After a read / write request corresponding to a read value is transmitted to the second computer, one remaining read / write request is sent each time one response from the second computer corresponding to the read / write request is confirmed. The last one read / write request transmitted to the second computer is transmitted as a reservation cancellation + read / write batch request with a reservation cancellation request added, and the second request is transmitted for all the transmitted requests. After confirming that a response is returned from the computer, the result of the read / write call is returned to the client program. On the other hand, the second computer receives a request from the first computer. When the received request is a reservation + read / write batch request, the second computer allows one client program according to the current load status of the own computer. Notifying the first computer of the number of read / write requests as the number of reservations and securing a memory amount corresponding to the number of reservations, then accessing the requested data on the secondary storage device and requesting the client Processing to send one response to the program. When the received request is a read / write request, after accessing the requested data on the secondary storage device, send one response to the requesting client program. If the received request is a reservation cancellation + read / write batch request, after accessing the requested data on the secondary storage device The process of the read / write request being processed generated from the requesting client program releases the amount of memory to the reserved upon completion all, with respect to multiple requests, executes multiplexes.
[0020]
In this way, by introducing a reservation + read / write batch request that requests a reservation request and a read / write request in a lump, the first computer does not have to send a reservation request as a separate request first. Since the second computer performs read / write processing following the reservation response, the communication time of the reservation response can be overlapped with the physical read / write time. If a subsequent read / write request is transmitted from the first computer during the physical read / write in the second computer, the communication time may overlap the physical read / write time. it can. Furthermore, by introducing a reservation cancellation + read / write batch request that requests a reservation cancellation request and a read / write request at once, it is not necessary to send a reservation cancellation request as a separate request at the end. Time can be saved.
[0021]
DETAILED DESCRIPTION OF THE INVENTION
Next, exemplary embodiments of the present invention will be described in detail with reference to the drawings.
[0022]
Referring to FIG. 1, an example of a computer system to which the present invention is applied includes a computer 1 on which a client program operates, a server computer 2, an inter-computer connection network 3 that connects these computers so that they can communicate with each other, and a server computer 2 And a secondary storage device 4 connected to the.
[0023]
A process called a file server 10 is operating on the server computer 2. The file server 10 receives a read / write request by communication, reads / writes data on the secondary storage device 4 connected to the server computer 2 according to the request content, and returns a response by communication. In the example of FIG. 1, the functions that the server computer should have are realized by operating a process called the file server 10 on a general-purpose computer, but other similar functions are implemented in the OS of the general-purpose computer. Embodiments and embodiments realized by adding firmware to a dedicated computer are also conceivable.
[0024]
In the computer 1 on which the client program operates, a client program 5 that is an application program is operating. In the example of FIG. 1, a dedicated program called a library 11 is linked to the client program 5 in order to transmit a read / write call from the client program 5 to the server computer 2. The library 11 generates an appropriate read / write request in accordance with a read / write call from the client program 5 and transmits it to the file server 10. Request / response exchange between the library 11 and the file server 10 follows the transfer method described later, but the transmission of the request is controlled by the library 11. In addition, instead of the library 11, an embodiment in which the same function as that of the library 11 is installed in the OS of a general-purpose computer is also conceivable.
[0025]
Several embodiments will be described below, but before that, a basic concept for performing multiple processing of a plurality of requests in the file server 10 which is the premise thereof will be described.
[0026]
When attention is paid to one request, the typical processing sequence is made up of a series of actions with several interruptions, as shown in FIG. That is, a corresponding initial action is determined according to the type of request, and after passing through several intermediate actions as the process proceeds, a response is returned to the computer 1 in which the client program 5 operates in the final action. One request is completed. Accordingly, multiple processing of requests can be realized by processing actions of other requests at the interruption point.
[0027]
The file server 10 manages the actions as described above using a queue (action queue), and executes a main loop as shown in FIG. The file server 10 has a mouth (reception port) for receiving a request, and receives the request by reading this (S1). Next, the initial action corresponding to the received request is registered at the end of the action queue (S2). Next, one action is extracted from the head of the action queue (S3). Then, the action is processed (S4), and the process returns to step S1. By such a main loop, the file server 10 alternately processes a plurality of request actions, thereby realizing multiple processing of requests.
[0028]
One of the interruption points of request processing is physical I / O issuance. At this time, the I / O is issued asynchronously, the action process is terminated, and the multi-process is realized by processing the action of another request. When the asynchronous I / O is completed, an asynchronous I / O completion handler (not shown) operates. The processing shown in FIG. 2C is executed by this asynchronous I / O completion handler.
[0029]
First, an action subsequent to the action that issued the I / O is determined. There are several possible methods for this determination as follows. 1) A method of directly determining by specifying an action as an argument of the handler. 2) A method in which an action subsequent to the action that issued the I / O is registered in the queue, and the handler searches the queue to determine the corresponding action.
[0030]
Next, the determined action is registered at the head of the action queue. This registered action is then taken out and processed in the main loop of the file server 10.
[0031]
Note that the asynchronous I / O completion handler interrupts and starts at an arbitrary timing during the operation of the file server 10, but the action queue is exclusive so that the file server 10 and the asynchronous I / O completion handler are not operated simultaneously. .
[0032]
Hereinafter, a method for accessing data on the secondary storage device 4 connected to the server computer 2 from the client program 5 operating on the computer 1 will be described with some examples.
[0033]
(Embodiment 1) In this embodiment, when there is a read / write call from the client program 5, the library 11 executes the processing shown in FIG.
[0034]
First, a read / write call from the client program 5 is analyzed and divided into read / write requests for data of a certain size (L11). For example, when a certain fixed size is 8 Kbytes, and the read / write call from the client program 5 is, for example, reading or writing of 24 Kbytes of data, a read / write request for 8 Kbytes from the top of the requested data And a read / write request for the next 8 KB, and a read / write request for the last 8 KB.
[0035]
Next, all the read / write requests obtained by the division are transmitted all at once to the file server 10 of the server computer 2 via the inter-computer connection network 3 (L12). Then, it waits for a response returned from the server computer 2 (L13). Waiting for a response is a process of reading a response reception port of the client program 5, and if a response is not received, waits until it is received. When a plurality of read / write requests are transmitted, a plurality of responses are returned.
[0036]
Next, when a response from the server computer 2 is received, the number of responses that have been returned so far is checked to see if all responses have been returned (L14). If the number of responses is smaller than the division number of the read / write request, that is, if all responses have not been returned, the process returns to step L13 and waits for responses. If all responses have been returned, the process returns to the call of the client program 5.
[0037]
On the other hand, the file server 10 receives a read / write request from the library 11, accesses the requested data on the secondary storage device 4 for one received read / write request, and then responds 1 to the request source. The process of transmitting one is multiplexed and executed for a plurality of read / write requests. Since the method of multiplexing and executing requests has been described above, a sequence focusing on one read / write request will be described with reference to FIG.
[0038]
First, the file server 10 secures a memory (data buffer) serving as a relay for temporarily storing read / write data relating to the read / write request (S11). Next, the physical read / write is issued asynchronously to the secondary storage device 4 and waits for its completion (S12). Here, the first action ends, and the request processing is interrupted. Next, when the physical read / write is completed, a read / write response is transmitted to the request source (S13). This is the final action of the read / write request.
[0039]
As described above, in the present embodiment, the read / write call by the client program 5 is divided into read / write requests for data of a certain size and transmitted all at once to the server computer 2, and the plurality of read / write requests received are transmitted. The server computer 2 multiplexes the write request, and the physical read / write processing of the secondary storage device in the server computer 2 and the communication processing in the inter-computer connection network 3 are performed in parallel. As a result, the communication time overlaps with the physical read / write processing time, and the overall processing time is shortened.
[0040]
A state in which the communication processing time and the physical read / write processing time overlap will be described with reference to FIG. FIG. 5 shows a case where a read / write call from the client program 5 is divided into four read / write requests and requested. First, the library 11 first sends four read / write requests to the server computer 2 at a time. Then, the file server 10 of the server computer 2 multiplexes four read / write requests, but since physical read / write is started when the first request is received, the communication time of the subsequent request is the first read. / Overlapping with write processing time. This makes most of the requested communication time invisible to the client program 5. Further, at the end of each request, a response is returned to the library 11. During that time, the file server 10 of the server computer 2 processes the physical read / write of the next request in an overlapping manner. As a result, most of the response communication time is invisible to the client program 5.
[0041]
Referring to FIG. 5, it can be seen that most of the communication time overlaps with the physical read / write time in the server computer 2. As a result, most of the communication time is invisible to the client program 5, and the processing time of the read / write call is shortened. However, although the communication time 24 in FIG. 5 is invisible to the client program 5, the communication time 23 of the first request and the last response is visible to the client program 5. However, as the amount of data to be accessed increases and the number of divisions increases, this time 23 becomes relatively smaller than the total of the physical read / write times 22, and the read / write call processing time 21 becomes the physical read / write time. It is equivalent to the sum total of the time 22.
[0042]
Responses are returned one after another from the server computer 2 to the computer 1 on which the client program 5 runs. On the other hand, the library 11 can determine the completion of the read / write call from the client program 5 by grasping that the processing of all the requests has been completed.
[0043]
(Embodiment 2) In this embodiment, when there is a read / write call from the client program 5, the library 11 executes the processing shown in FIG.
[0044]
First, as in the first embodiment, a read / write call from the client program 5 is analyzed and divided into read / write requests for data of a certain fixed size (L21).
[0045]
Next, it is examined whether or not the number of divisions is larger than a predetermined threshold value (L22). If the number of divisions is not greater than the threshold value, all read / write requests obtained by the division are transmitted all at once to the file server 10 of the server computer 2 via the inter-computer connection network 3 (L24). If the number of divisions is larger than the threshold value, among the read / write requests obtained by the division, only the number of read / write requests equal to the threshold value are sent to the server computer 2 via the inter-computer connection network 3 at a time. It transmits to the file server 10 (L23). At this time, read / write requests that have not yet been transmitted are held inside. Then, it waits for a response returned from the server computer 2 (L25).
[0046]
Next, when a response from the server computer 2 is received, it is confirmed whether or not there is a read / write request that has not been transmitted yet (L26). If there is still a read / write request that has not been transmitted, one of the read / write requests is transmitted to the file server 10 of the server computer 2 and the process waits for a response in step L25. If it is determined in step L26 that there are no untransmitted read / write requests, the number of responses returned so far is checked to determine whether all responses have been returned (L28). If the number of responses is smaller than the division number of the read / write request, that is, if all responses are not returned, the process returns to step L25 again to wait for responses. If all responses have been returned, the process returns to the call of the client program 5.
[0047]
The processing of the file server 10 is the same as that of the first embodiment, receives a read / write request from the library 11, and accesses the requested data on the secondary storage device 4 for one received read / write request. Thereafter, the process of transmitting one response to the request source is executed in a multiplexed manner for a plurality of read / write requests.
[0048]
Similar to the first embodiment, the basic effect of the present embodiment includes the effect that most of the communication time is overlapped with the physical read / write to be invisible to the client program. However, in the first embodiment, all read / write requests obtained by division are transmitted all at once. Therefore, when the data amount of the read / write call from the client program 5 is large, the read / write request transmitted at once. And the load on the server computer 2 and the inter-computer communication network 3 becomes excessive. If the load on the server computer 2 or the like increases abnormally, some of the read / write requests may disappear and may not be processed, so the read / write call from the client program 5 may fail.
[0049]
On the other hand, in this embodiment, the threshold value is set as the number of requests that can be received at one time by the server computer 2, and the number of read / write requests to be transmitted first is kept below the threshold value. The initial load at can be kept below a certain value. Furthermore, since one read / write request is transmitted in response to one response, the number of read / write requests received by the server computer 2 is always below the threshold value, and the load on the server computer 2 continues to be below a certain value. Can be suppressed.
[0050]
An example of the exchange of request / response that occurs between the server computer 2 and the client program 5 in this embodiment is shown in FIG. In this example, the read / write call from the client program 5 is divided into eight read / write requests, and the threshold value is set to 4.
[0051]
First, since the threshold value is 4, the library 11 first transmits only four read / write requests to the server computer 2. Thereby, the number of read / write requests received first by the server computer 2 can be suppressed to 4 or less.
[0052]
Thereafter, the file server 10 of the server computer 2 multiplexes each read / write request and returns a response one by one. On the other hand, the library 11 transmits one read / write request to the server computer 2 every time one response is received. As a result, the number of read / write requests being multiplexed by the server computer 2 is always 4 or less.
[0053]
(Embodiment 3) In this embodiment, when there is a read / write call from the client program 5, the library 11 executes the processing shown in FIG.
[0054]
First, as in the first embodiment, a read / write call from the client program 5 is analyzed and divided into read / write requests for data of a certain size (L31).
[0055]
Next, a reservation request designating this division number is transmitted to the file server 10 of the server computer 2 via the inter-computer connection network 3, and a response is waited (L32).
[0056]
Next, when a response to the reservation request is received from the file server 10 of the server computer 2, the number of reservations included in the response is extracted and set as a threshold value (L33).
[0057]
Next, it is checked whether or not the number of divisions in step L31 is larger than this threshold value (L34). If the number of divisions is not greater than the threshold value, all read / write requests obtained by the division are all transmitted at once to the file server 10 of the server computer 2 via the inter-computer connection network 3 (L36). If the number of divisions is larger than the threshold value, among the read / write requests obtained by the division, only the number of read / write requests equal to the threshold value are sent to the server computer 2 via the inter-computer connection network 3 at a time. It transmits to the file server 10 (L35). At this time, read / write requests that have not yet been transmitted are held inside. Then, it waits for a response returned from the server computer 2 (L37).
[0058]
Next, when a response from the server computer 2 is received, it is confirmed whether or not there is a read / write request that has not been transmitted yet (L38). If there is a read / write request that has not been sent yet, one of the read / write requests is sent to the file server 10 of the server computer 2 (L39), and the process goes to the waiting for response in step L37. If it is determined in step L38 that there is no untransmitted read / write request, the number of responses that have been returned is checked to see if all responses have been returned (L3a). If the number of responses is smaller than the division number of the read / write request, that is, if all responses are not returned, the process returns to step L37 and waits for responses.
[0059]
If all responses have been returned, a request (reservation cancellation request) for canceling the reservation in the server computer 2 is transmitted to the file server 10 of the server computer 2 via the inter-computer connection network 3, and the response is awaited. (L3b). Then, after a response to the reservation cancellation request is returned, the process returns to the call of the client program 5.
[0060]
In this embodiment, the file server 10 performs a reservation request process and a reservation release request process in addition to the read / write request process. Since the read / write request process is almost the same as in the first and second embodiments, the reservation request process and the reservation release request process will be described below.
[0061]
FIG. 9 shows a sequence focusing on one reservation request process. As described above, when a reservation request is transmitted from the library 11, the file server 10 first calculates the number of read / write requests that are allowable for the client program (S31).
[0062]
This allowable number depends on the current load status of the file server 10. Specifically, it is almost determined by the current remaining memory capacity that can be used by the file server 10 to process read / write requests. For example, if the total amount of memory that can be used by the file server 10 to process a read / write request is A and the amount of memory required to process one read / write request is B, [A / B] is a file. It becomes the processing load upper limit of the server 10. In addition, when reservations for C read / write requests are accepted at this time, the current allowable total number (total allowable number) is C, and the current allowable number of read / write requests is the processing load upper limit − C. Thus, the file server 10 obtains the allowable number for the client program by subtracting the current allowable number (total allowable number) from the upper limit of the processing load.
[0063]
Next, the smaller value of the allowable number obtained in step S31 and the number of divisions specified in the reservation request is determined as the reservation number (S32). Then, the total of the current allowable number is increased by the number of reservations this time (S33), and a response specifying the number of reservations obtained in step S32 is transmitted to the request source (S34). By increasing the total number of current allowances by the number of reservations this time, it is possible to prevent the amount of memory commensurate with the current number of reservations from being used in read / write request processing generated by other client programs. . Therefore, it is possible to prevent a memory shortage from occurring when a memory serving as a relay for temporarily storing read / write data related to a read / write request from the client program is secured.
[0064]
An embodiment in which the processing content of step S32 is changed to the number of reservations = min (allowable number, number of divisions, maximum allowable number given per client program) is also conceivable. In this way, the maximum allowable number given to one client program can be limited, and one client program can be prevented from occupying all the current allowable numbers.
[0065]
The reservation request process as described above consists of one action.
[0066]
On the other hand, FIG. 10 shows a sequence focusing on the processing of one reservation cancellation request. As described above, when a reservation cancellation request is transmitted from the library 11, the file server 10 updates the current allowable number by subtracting the reservation number given to the client program (S35). As a result, the reserved memory amount is released, and the amount of memory that can be used for processing read / write requests from other client programs is increased accordingly. Then, the end of the reservation cancellation request process is notified to the library 11 by a response (S36). Such reservation cancellation request processing consists of one action.
[0067]
As in the second embodiment, when the threshold value is determined in advance by the processing capacity of the server computer 2, if there is one client program, the number of read / write requests exceeding the processing capacity of the server computer 2 will come. There is no. However, when there are a plurality of client programs that transmit read / write requests to the server computer 2, the processing capacity of the server computer 2 may be exceeded.
[0068]
On the other hand, in this embodiment, the number of reservations allowed for one client program is notified to the library 11 in accordance with the load status of the file server 10 of the server computer 2 that changes dynamically, and the library 11 reserves the reservation. Since the number is set as a threshold value and a read / write request is subsequently made according to the threshold value, the load generated in the server computer 2 can be controlled so as not to exceed a certain value. That is, the server computer 2 does not exceed the processing capacity of the server computer 2 even if a plurality of client programs are dealt with. Further, since the number of reservations is reset when all the read / write requests from the client program are completed, the memory can be effectively used for the reservation request of the client program that occurs thereafter.
[0069]
FIG. 11 shows an example of request / response exchange that occurs between the server computer 2 and the client program 5 in this embodiment. Except for the reservation process (A) and the reservation cancellation process (B), the exchange is the same as in the second embodiment. In the example of FIG. 11, the read / write call from the client program 5 is divided into five read / write requests, but it is assumed that the number of reservations from the file server 10 is three. For this reason, the library 11 sets 3 as the threshold value, and first sends three read / write requests to the server computer 2. The subsequent processing is the same as the processing procedure of the second embodiment. However, after all read / write requests have been processed, the reservation cancellation processing (b) is performed, and the processing returns to the read / write call from the client program 5. .
[0070]
(Embodiment 4) In this embodiment, when there is a read / write call from the client program 5, the library 11 executes the processing shown in FIG.
[0071]
First, as in the first embodiment, a read / write call from the client program 5 is analyzed and divided into read / write requests for data of a certain size (L41).
[0072]
Next, it is checked whether or not this division number is larger than 1 (L42). If the number of divisions is one, without making a reservation request, one read / write request is transmitted to the file server 10 of the server computer 2 via the inter-computer connection network 3, and a response is received (L44).
[0073]
On the other hand, when the number of divisions is greater than 1, one request is a reservation + read / write batch request in which a reservation request is added to one read / write request among the read / write requests obtained by the division. To the file server 10 of the server computer 2 and wait for reception of a reservation response (L43). Waiting for reception of a reservation response is a process of reading a response reception port of the client program 5, and if a reservation response has not been received, it waits until it is received.
[0074]
Next, when a reservation response is returned from the server computer 2, the number of reservations included in the response is taken out and set as a threshold value (L45).
[0075]
Next, it is checked whether or not the division number -1 in step L41 is larger than this threshold value (L46). If the division number −1 is not larger than the threshold value, all the remaining read / write requests obtained by the division are transmitted to the file server 10 of the server computer 2 all at once via the inter-computer connection network 3. (L48). However, one of the read / write requests is added with a reservation release request to make a reservation release + read / write batch request, which is transmitted as one request. When the division number −1 is larger than the threshold value, only the number of read / write requests equal to the threshold value among the remaining read / write requests obtained by the division are transmitted via the inter-computer network 3 at a time. To the file server 10 of the server computer 2 (L47). At this time, read / write requests that have not yet been transmitted are held inside. Then, it waits for a read / write response returned from the server computer 2 (L49).
[0076]
Next, when a read / write response is received from the server computer 2, it is confirmed whether or not there is a read / write request that has not been transmitted (L4a). If there is still a read / write request that has not been transmitted, it is checked whether the number is 1 (L4c). If it is not 1, there are two or more read / write requests that have not yet been transmitted. If it is, one of the read / write requests is transmitted to the file server 10 of the server computer 2 (L4d), and the process waits for a response in step L49. On the other hand, if there is one read / write request that has not been transmitted yet, a request for canceling reservation is added to the request and a request for canceling reservation + read / write batch request is transmitted to the file server 10 of the server computer 2 ( L4e), the process waits for a response in step L49.
[0077]
If it is determined in step L4a that there are no untransmitted read / write requests, the number of responses returned so far is checked to determine whether all responses have been returned (L4b). Is less than the division number of the read / write request, that is, if all the responses are not returned, the process returns to step L49 and waits for a response. If all responses have been returned, the process returns to the call of the client program 5.
[0078]
In this embodiment, the file server 10 multiplexes and executes reservation + read / write batch request processing and reservation cancellation + read / write batch request processing in addition to read / write request processing.
[0079]
FIG. 13 shows a sequence focusing on one read / write request. The file server 10 first secures a memory serving as a relay for temporarily storing read / write data relating to the read / write request (S41). Next, physical read / write is asynchronously performed on the secondary storage device 4, and the completion is waited (S42).
[0080]
Next, when the physical read / write is completed, it is checked whether the reservation cancellation + read / write batch request processing is interrupted by the request from the same client program 5 (S43). If it is interrupted, it is checked whether there is another read / write request being processed from the same client program 5 (S44), and if it is interrupted, the reservation release interrupted + read / write batch request is interrupted. Is registered in the request queue (S45), and a response to the read / write request for which the physical read / write has been completed is transmitted to the request source (S46). On the other hand, if the reservation cancellation + read / write batch request processing is not interrupted, or if there is no such reservation cancellation + read / write batch request, or there is a suspended reservation cancellation + read / write batch request. However, if there is a read / write batch request being processed from the same client program, step S45 is skipped and a response to the read / write request is transmitted to the request source (S46).
[0081]
FIG. 14 shows a sequence focusing on processing of one reservation + read / write batch request. As described above, when a reservation + read / write batch request is transmitted from the library 11, the file server 10 first calculates the number of read / write requests that can be accepted for the client program (S47). This allowable number depends on the current load status of the file server 10. Specifically, it is calculated as a value that is the upper limit of the processing load of the file server 10 minus the total of the current allowable number (total allowable number).
[0082]
Next, the smaller value of the allowable number obtained in step S47 and the division number -1 specified in the reservation request is determined as the reservation number (S48). Then, the total number of current allowable numbers is increased by the number of reservations this time (S49), and a response specifying the number of reservations obtained in step S48 is asynchronously transmitted to the request source (S4a).
[0083]
An example in which the processing content of step S48 is changed to the number of reservations = min (allowable number, division number-1, maximum allowable number given per client program) is also conceivable. In this way, the maximum allowable number given to one client program can be limited, and one client program can be prevented from occupying all the current allowable numbers.
[0084]
The processing so far is substantially the same as the reservation request processing in the file server 10 of the third embodiment. However, in this embodiment, since the reservation response is transmitted asynchronously and then the read / write request processing is continued, the read / write request processing time and the reservation response communication time overlap.
[0085]
In the read / write request processing, first, a memory serving as a relay for temporarily storing read / write data related to the read / write request is secured (S4b). Next, physical read / write is performed asynchronously to the secondary storage device 4, and the completion is awaited (S4c). When the physical read / write is completed, a read / write response is transmitted to the request source (S4d).
[0086]
On the other hand, FIG. 15 shows a sequence focusing on one reservation cancellation + read / write batch request. As described above, when a reservation cancellation + read / write batch request is transmitted from the library 11, first read / write request processing is performed, and then reservation cancellation processing is performed. In the read / write request processing, a memory serving as a relay for temporarily storing read / write data related to the read / write request is secured (S4e), and the physical read / write is asynchronously performed on the secondary storage device 4, Wait for the completion (S4f). When the physical read / write is completed, it is checked whether there is another read / write request being processed from the same client program 5 (S4g), and if there is any, it is suspended until there is no read / write request being processed. . The interruption method is performed by registering the request in the interruption queue. The request registered in the suspension queue is reregistered from the suspension queue to the request queue in step S45 of FIG.
[0087]
When there is no other read / write request being processed, the reservation is canceled and the current total allowable number is updated by subtracting the reservation number (S4h). As a result, the memory amount reserved for the client program is released. Then, a reservation cancellation response is included in the read / write response and transmitted to the request source (S4i).
[0088]
According to this embodiment, since the reservation + read / write batch request is introduced, the library 11 does not need to send the reservation request as a separate request first, and the file server 10 reads / writes following the reservation response. Since processing is performed, the communication time for the reservation response can be overlapped with the physical read / write time. If a subsequent read / write request is transmitted from the client program during physical read / write in the server computer, the communication time can be overlapped with the physical read / write time.
[0089]
In addition, since the reservation cancellation + read / write batch request is introduced, it is not necessary to send the reservation cancellation request as a separate request lastly, and the communication time can be reduced accordingly.
[0090]
From the above, the communication time for the first and last reservation / reservation cancellation can be reduced. FIG. 16 shows an example of request / response exchanges that occur between the server computer 2 and the client program 5. The first reservation + read / write batch request (c) is sent to the server computer 2 as one request. The server computer 2 first performs reservation processing, returns a reservation response, and proceeds to read / write processing at the same time. At this time, the physical read / write and reservation response communication time overlap.
[0091]
The read / write request processing after the reservation response is received by the library 11 of the computer 1 on which the client program 5 operates is the same as in the third embodiment. However, the difference is that the last request is a reservation cancellation + read / write batch request. The last request processing corresponds to the reservation cancellation + read / write (d) processing shown in FIG. In the example of FIG. 16, the read / write call from the client program 5 is divided into five read / write requests, but it is assumed that the number of reservations from the file server 10 is three. Therefore, the library 11 sets the threshold value to 3, and after receiving the reservation response, first sends three read / write requests to the server computer 2. Subsequent processing is the same as the processing procedure of the second and third embodiments, but when the remaining read / write request becomes one, a request that combines reservation cancellation and read / write (reservation cancellation + read / write). (Collective request). In response to this request, the server computer performs read / write processing and reservation cancellation processing, and returns it as one response. Thereby, the last communication time can be reduced.
[0092]
Although the embodiments of the present invention have been described above, the present invention is not limited to the above embodiments, and various other additions and modifications can be made. For example, in FIG. 1, a computer 1 on which one client program 5 operates is connected to one server computer 2 via an inter-computer communication network 3, but a computer on which one or more client programs 5 operate. It is also applicable to a computer system in which a plurality of computers are connected to the inter-computer communication network 3. Further, although an example in which only one secondary storage device 4 is connected to the server computer 2 has been shown, a plurality of secondary storage devices may be connected.
[0093]
【The invention's effect】
As described above, according to the present invention, the following effects can be obtained.
[0094]
According to the present invention, most of the communication time at the time of data access to the remote secondary storage device is made invisible to the client program, and the access is as if the secondary storage device is connected to the computer on which the client program operates. Time can be achieved.
[0095]
Further, according to the present invention, even when the client program accesses a large amount of data, that is, when the number of divided read / write requests increases, the read / write requests requested at a time can be performed. The amount can be suppressed, and the processing load on the second computer can be prevented from becoming excessive.
[0096]
Furthermore, according to the present invention, even when a plurality of client programs make read / write requests to one second computer, the amount of read / write requested at a time according to the load on the second computer is reduced. This can suppress the processing load of the second computer from becoming excessive.
[0097]
Furthermore, according to the present invention, the first computer side does not need to send the reservation request as a separate request first, and the second computer side performs read / write processing following the reservation response. Communication time can be overlapped with physical read / write time. If a subsequent read / write request is transmitted from the first computer during the physical read / write in the second computer, the communication time may overlap the physical read / write time. it can. Further, the first computer side does not need to send a reservation cancellation request as a separate request lastly, and the communication time can be reduced accordingly.
[Brief description of the drawings]
FIG. 1 is a block diagram showing an example of a computer system to which the present invention is applied.
FIG. 2 is a flowchart illustrating an example of a process for multi-processing a plurality of requests in a file server.
FIG. 3 is a flowchart illustrating an example of library processing according to the first exemplary embodiment.
FIG. 4 is a flowchart illustrating an example of processing focusing on one read / write request of the file server according to the first embodiment.
FIG. 5 is a timing chart showing request / response exchange between the client program and the server computer in the first embodiment.
FIG. 6 is a flowchart illustrating an example of library processing according to the second exemplary embodiment.
FIG. 7 is a timing chart showing request / response exchanges between a client program and a server computer in Embodiment 2.
FIG. 8 is a flowchart illustrating an example of library processing according to the third exemplary embodiment.
FIG. 9 is a flowchart illustrating an example of processing focusing on one reservation request of the file server according to the third embodiment.
FIG. 10 is a flowchart illustrating an example of processing focusing on one reservation cancellation request of the file server according to the third embodiment.
FIG. 11 is a timing chart illustrating request / response exchanges between a client program and a server computer according to a third embodiment.
12 is a flowchart illustrating an example of library processing in Embodiment 4. FIG.
FIG. 13 is a flowchart illustrating an example of processing focusing on one read / write request of the file server according to the fourth embodiment.
FIG. 14 is a flowchart illustrating an example of processing focusing on one reservation + read / write batch request of the file server according to the fourth embodiment.
FIG. 15 is a flowchart illustrating an example of processing focusing on one reservation cancellation + read / write batch request in the file server according to the fourth embodiment.
FIG. 16 is a timing chart illustrating request / response exchanges between a client program and a server computer according to a fourth embodiment.
FIG. 17 is an explanatory diagram of a conventional access method to a remote secondary storage device.
[Explanation of symbols]
1. Computer that runs client program
2 ... Server computer
3. Computer network
4 ... Secondary storage device
5 ... Client program
10 ... File server
11 ... Library

Claims (3)

First 1 From the client program running on the other computer 1 That can communicate with each other through the computer network 2 A method of accessing data on a secondary storage device connected to a computer of
  Said 1 In this computer, the read / write call by the client program is divided into read / write requests for data of a certain size,
  Prior to transmission of the read / write request obtained by the division, the total number of the read / write requests is set as the division number, and is included in the reservation request. 2 Send to
  Said request for said transmitted reservation request; 2 The threshold value is determined from the number of reservations included in the response from the computer.
  When the number of read / write requests obtained by the division is equal to or less than the threshold value, the first 2 If the threshold value is exceeded, the first read / write request for the threshold value is sent. 2 After sending to the other computer, the first corresponding to the read / write request 2 Response from the calculator 1 The remaining read / write requests 1 Said first 2 To the calculator
  For all the read / write requests sent 2 After confirming that a response has been returned from 2 Request the computer to cancel the reservation, and return the result of the read / write call to the client program,
  Said 2 In the calculator of
  Said 1 In response to a reservation request from another computer, 1 The quotient divided by the data size used in one read / write request is defined as the upper limit of the processing load, and the difference obtained by subtracting the total number of reservations currently reserved from the upper limit of the processing load is defined as the allowable number. 1 The number of divisions included in the reservation request from the computer, and 1 Of the maximum allowable number assigned per client program, the minimum value is determined as the reservation number, and the reservation number is determined as the number of reservations. 1 And ensure the amount of memory commensurate with the number of reservations,
  Said 1 Received read / write requests from other computers and received read / write requests 1 After accessing the requested data on the secondary storage device, it responds to the requesting client program. 1 The process of sending one is multiplexed for multiple read / write requests,
  Said 1 In response to a reservation cancellation request from the other computer, the reserved memory amount is released,
  And said first 2 Multiplexing of processing in the computer of the system from receiving the request to returning the response 1 It is realized by executing an action of another request at a point of interruption of a series of actions related to one request, I / O A method for remote access to a secondary storage device, characterized in that an action subsequent to an action issued is preferentially processed. First 1 From the client program running on the other computer 1 That can communicate with each other through the computer network 2 A method of accessing data on a secondary storage device connected to a computer of
  Said 1 In the calculator of
  The read / write call by the client program is divided into read / write requests for data of a certain size,
  Of the read / write requests obtained by the division 1 A reservation request including the total number of read / write requests as a division number is added to one read / write request, and the above-mentioned reservation + read / write batch request is added. 2 Send to
  Said request for said transmitted reservation request; 2 The threshold value is determined from the number of reservations included in the response from the computer.
  When the remaining number of read / write requests obtained by the division is less than or equal to the threshold value, 1 A reservation cancellation request is added to one read / write request to make a reservation cancellation + read / write batch request. 2 When the threshold value is exceeded, the first read / write request for the threshold value is sent. 2 After sending to the other computer, the first corresponding to the read / write request 2 Response from the calculator 1 Every time you check Read / write requests 1 Said first 2 And send it to the last computer 1 One read / write request is sent as a reservation cancellation + read / write batch request with a reservation cancellation request added,
  For all the read / write requests sent 2 After confirming that a response is returned from the computer, the read / write call result is returned to the client program,
  Said 2 In the calculator of
  Said 1 If the request received from this computer is a reservation + read / write batch request, the total memory amount of the local computer is 1 The quotient divided by the data size used in one read / write request is defined as the upper limit of the processing load, and the difference obtained by subtracting the total number of reservations currently reserved from the upper limit of the processing load is defined as the allowable number. 1 The number of divisions included in the reservation request from the computer, and 1 Of the maximum allowable number assigned per client program, the minimum value is determined as the reservation number, and the reservation number is determined as the number of reservations. 1 After confirming the amount of memory corresponding to the number of reservations, access the requested data on the secondary storage device and respond to the requesting client program. 1 If the received request is a read / write request, access the requested data on the secondary storage device and then respond to the requesting client program. 1 If the received request is a reservation cancellation + read / write batch request, the data stored in the requested secondary storage device is accessed, and the request generated by the client program When all the read / write requests are completed, the process of releasing the secured memory amount is executed in a multiplexed manner for a plurality of requests,
  And said first 2 Multiplexing of processing in the computer of the system from receiving the request to returning the response 1 It is realized by executing the action of another request at the interruption point of a series of actions concerning one request, I / O A method for remote access to a secondary storage device, characterized in that an action subsequent to an action issued is preferentially processed. Said 2 A plurality of the above-mentioned computers 1 3. The method for remote access to a secondary storage device according to claim 1, wherein said computers are connected through said inter-computer connection network.

Images