JPS63120349A - File multiplex control system - Google Patents

Abstract

PURPOSE:To improve the processing performance with a file multiplex control system by carrying out simultaneously the read requests not only to different file data stored in the same volume but to the same file data in case a volume is shared by plural CPU. CONSTITUTION:A processing request 110 is given from a higher operating system 103 to a data group B1 simultaneously with a processing request 112 given from a higher operating system 104 to a data group B2. Thus both requests 110 and 112 can be carried out at a time without being kept waiting regardless of a read or write mode. Furthermore a read processing request 111 for the group B1 is given to the same group B1 from the system 104 during a read processing mode of the system 103 and also can be carried out at a time. A directory memory 106 receives accesses from memory directors 107 and 108. One of these two directors is kept waiting until the access given to the memory 106 by the other director is through. This waiting time is extremely short and is substantially neglected.

Description

[Detailed Description of the Invention] [Field of Industrial Application] The present invention relates to a file multiplexing control system, and in particular to an external system that uses a semiconductor memory as all or part of a storage medium, or as an intermediate buffer in a data transfer path. This invention relates to a file multiplexing control method applied to storage systems.

[Conventional technology]

Conventionally, in magnetic disk devices, polycomb shares, which were shared by multiple CPUs, were shared by one CPU.
(Operating System), even if there is an I/O request for the same volume from another CPTJ (Operating System),
This is not accepted and is kept on standby. In other words, in input/output processing within one operating system,
There is one receiver for input/output requests, and they are processed on a volume-by-volume basis. When there are multiple input/output instructions for a given volume, the order of input/output processing is controlled by creating a queue within the operating system. Additionally, when there are input/output requests to the same volume from multiple operating systems, the Z
ζ, its external storage device is controlling this.

Note that the unit known as volume recognized by the operating system is, for example, the volume in a conventional magnetic disk device, that is, one head arm equipped with multiple magnetic coating disks and a group of heads for reading and /&F. It physically corresponds to a unit (actuator) composed of Goku means that the head arm, etc. is occupied until the processing of the input/output request is completed, and even if there are other input/output requests, simultaneous processing is physically impossible. .

Against this background, in recent years, as volume capacity has increased, the number of files in a volume has also tended to increase, and the opportunities for 2J or so volumes to be shared by multiple systems have also increased. ing. As a result, waiting for a volume to become free under a single operating system7 or waiting for a volume to become free due to volume sharing among multiple operating systems causes a reduction in system throughput.9 On the other hand,7 high-speed data on external storage devices In order to achieve access, the storage medium is either a semiconductor memory device composed entirely of 4'-conductor memory, or Disk caches and the like have appeared in which semiconductor memory is installed as an intermediate buffer in the data transfer path.

[Problem that the invention seeks to solve]

As a conventional file multiple selection method, for example,
As described in Japanese Patent No. 9-45568, there is a system in which magnetic disk drives are selected within a range that can perform simultaneous operations and are caused to perform parallel operations.

That is, in the above method, in a magnetic disk subsystem equipped with a disk cache, when there are selection requests from multiple magnetic disk control devices for a magnetic disk device to be selected, those requests are sent to the magnetic disk device. When the selection request is for data on the cache memory of the same magnetic disk drive,
We are focusing on the fact that it is possible to divide these selection requests into levels, and to accept or accept the selection requests of Sowa 2 and others when each disk control device has a different selection level.

However, no consideration has been given to control when there are multiple processing requests for data expanded on the cache memory.

An object of the present invention is to improve such conventional problems, and to create a file system that can simultaneously execute read requests for the same data as well as different file data in the same volume when multiple CPUs share a volume. The purpose is to provide multiple control methods.

[Means for solving problems]

In order to achieve the above purpose, the file multiplex control method of the present invention registers and deletes the range of data groups used in input/output commands, read/write information, and the path of the input/output commands in the control device. A directory memory is provided to store a table for executing an input/output command, and when an input/output command is issued from another route to the same storage device that is already executing an input/output command from a higher-level device, the table in the directory memory is searched. The feature of the present invention is that the instructed information is compared with the information registered during the execution of the input/output command, and it is determined whether or not to permit the continuation of the input/output command. , when multiple input/output commands arrive at the same time for the same volume of data stored on semiconductor memory, (a) if they are for different data, execution is allowed regardless of read/write status; (b) In the case of processing requests for the same data, if a read request and a read solicitation are combined, the data will not be updated, so they are allowed to be executed at the same time. This results in
It is possible to improve throughput by eliminating the need for multiple input/output commands to wait for a volume to become available for the same volume.

By the way, the range of the data group instructed by the host device is as follows:
This is to define the data on the semiconductor memory used in the input/output command. The range of this data group includes, for example, a cylinder range and a track range, and can also be specified as control data by a program. Furthermore, data range information, that is, extent information indicating the size of a data group, which is conventionally used in operating systems, is also added to input/output instructions.

On the other hand, the control device stores the range of the designated data group and read/write information for each input/output instruction path when the input/output command starts, and deletes them when the input/output command ends. While this information is being registered for a certain route, if there is an input/output command for a different data group from another route, the execution of that input/output command is permitted, and the registered information, that is, the information being executed When the input/output command is a read process, if the read process request is for the same data group from another route, execution is permitted. However, the input/output command being executed is in the process of reading or writing.

If the input/output command received from another route is a write process,
By making the execution of the input/output command wait, it is possible to prevent inconsistencies due to incorrect file updates. Furthermore, regarding exclusive control on a volume-by-volume basis, by leaving the conventional reserve/release function for magnetic disks unchanged, no contradiction will occur.

〔Example〕

Hereinafter, five embodiments of the present invention will be described in detail with reference to the drawings.

FIG. 1 shows a computer system showing an embodiment of the present invention.
This is the essential diagram. Here, two CP TJlol, 10
Operating systems 103, 104 running on 2
and their operating systems 1.83.10
An input/output control device 105 shared by 4 and an external storage device, ie, a semiconductor memory 109, controlled by the control device Hios are provided. The control device 105 includes a storage director IQ7゜1 provided corresponding to each CPU.
08, and a directory memory 10G shared by these storage directors 1, 07, and 103.

Directory memory 106 is for storing the location and read/write information of data groups instructed by upper operating systems 103 and 104, and for managing data on semiconductor memory 109. semiconductor memory 1
09 is independently read/written from each storage director 107, 108. It is assumed that data of each volume A to N is stored on the semiconductor memory 109, and that a data group 1' (1, B2 exists in volume B).

Now, when a processing request 110 from the upper operating system 103 for data group B1 and a processing request 112 from the upper operating system 104 for data group B2 are issued at the same time, in this embodiment, read/write of the image processing request is executed. Regardless of the situation, image processing requests can be executed simultaneously without having to wait. Also, when the operating system 103 is in the middle of read processing for the data group B1, when a read processing request 111 for the same data group Bl is issued from the operating system 1.04, the read processing request 111 is executed simultaneously without making the request wait. Yes, it is possible (see solid arrow).

In addition, the operating system 103
1 during read or write processing, the operating system 104 writes to the same data group B1,
When a write processing request arrives, a processing command issued later is made to wait, so that there is no conflict due to a misunderstanding at the time of update.

Directory memory 106 is connected to each storage director 107
, 108, but while one storage director is accessing the directory memory 106,
Other storage directors must wait until their access is finished.

That is, the directory memory 106 is an exclusive memory. However, from each director the directory memory 106
Since the time to access is extremely short, there is almost no waiting time.

FIG. 2 is a diagram illustrating the Western movie configuration of the directory memory in FIG. 1.

The directory memory 10G is a control memory for managing data in the semiconductor memory 109J. As shown in FIG. 2, the directory memory 10G is an extent table in which extent information added to input/output commands from the upper operating system is registered. 301 to 303 are provided as many as the number of volumes.

This table 301 stores extent information that defines the range of data groups used in input/output commands from the operating system, update flags, and paths of input/output commands for the relevant extents. The extent tables 301 to 303 for each volume are composed of extent entries 201 of input/output commands for the volume. These extent tables 201 contain the start address/end address of the extent used in the relevant input/output instruction, the update flag indicating update processing, and the route to the upper CPU of the other person's output command requesting the relevant extent at the same time. It consists of a flag and a pointer section for managing extent entries in the table, and a plurality of these exist in the table. The number of entries in the table depends on the number of routes to the upper CPT, J for the same volume. That is, the greater the number of routes arriving from the third CPU to the control unit, the greater the number of entries. extension 1
- When an entry exists in Qizol, it indicates that data within the extent is being processed.
, at each start/end of an input/output command, the engine registers/
The search circuit 202 in FIG. 2 that is deleted is used to check whether the extension 1- is already in use or a page when receiving an input/output command. That is, the search circuit 202
When the extension l-[1ff start/end address and search table number are specified from each storage director 107 (:13: or 108), after storing this 71° in the search data register 204, A comparison circuit (not shown) searches for the magnitude of the extent address of the extent entry 1-201 with respect to the registered extent address, and the result is notified to the storage director 107 through the communication line 2/203.

FIG. 3 is an operational flowchart of a file multiplexing control system showing one embodiment of the present invention.

First, the storage director checks whether it has received an input/output command issued from the upper CPU that instructs extent and read/write (step).
When the command is received, a search circuit is activated to check whether the extent information in the command has already been registered in the extent table corresponding to the volume (step 3). If you have not received it, the extension 1 of the relevant input/output instruction
Since it is not possible to define the extent, the range of extent is assumed to be the entire one volume (step 2). This is to prevent conflicts from occurring during simultaneous access to the same volume. Next, check the search results of the extent table (step 4), and if the extent has not yet been registered, the start/end address of the extent 1- and the update flag indicating the processing request are added to the extent of the volume. It is registered in the table and processing is continued (step 6). At this point, a write processing request is made.

Turn on the update plug (step 11) and continue the process (step 12).

On the other hand, if the table search results show that the extent has already been registered, the update flag is checked (step 5).
), when the update flag is ON, the process cannot be accepted, so the process is made to wait by command retry, and the route flag of the route is set to ON (step 9.10). Further, even when the update flag is OFF, a write processing request is still not accepted, so the processing is made to wait by command retry. At the end of the input/output command, by deleting the relevant extent information from the extent table, the relevant extent 1- is released to the other person's output command waiting for the relevant extent (step 13).
, 1.4) At this time, execution of the extension 1-waiting external output command is resumed for the route whose route flag is ON (step 15.degree. 1G).

By performing such control, processing requests for different extents in the same volume or read processing requests for the same extent can be executed simultaneously and without contradiction.

〔Effect of the invention〕

As explained above, according to the present invention, when a volume is shared by multiple CPUs, read requests for the same file data as well as different file data in the same volume can be executed simultaneously. Therefore, processing performance is significantly improved.

[Brief explanation of the drawing]

Fig. 1 is a schematic diagram of a computer system capable of file multiplex control showing an embodiment of the present invention, Fig. 2 is a detailed configuration diagram of the directory memory shown in Fig. 1, and Fig. 3 is a diagram showing an embodiment of the present invention. 3 is a flowchart of multiple control operations shown in FIG. 10L, 102: Upper CPU, 103, 104: Operating system, 1o5: Input/output control device, 106
: Directory memory, 107° 108: Storage director, LO9: Semiconductor memory, 201: Extent entry, 202: Search circuit, 203: Result notification signal line, 204: Search data register, 301, 302, 3
03: Volume n extent table.

Claims (1)

[Claims] 1. A semiconductor storage device that stores file data or a cache device provided in a data transfer path with a host device, and a control device that performs data transfer with the semiconductor storage device or cache device via multiple paths. In the external storage subsystem having, in the control device,
A directory memory is provided to store a table for registering and deleting the range of data groups used in input/output commands, read/write information, and the path of the input/output commands, and a When an input/output command is issued to the same storage device through another route, the table in the directory memory is searched, the specified information is compared with the information registered during the execution of the input/output command, and the input/output command is executed. A file multiplex control method characterized by determining whether to permit continuation of an output command.