JPS60231232A - Control method of magnetic tape device in multiprocessor - Google Patents

Abstract

PURPOSE:To reduce access to a main bus used by a main processor part in common by limiting a slow data transfer section of a magnetic tape device up to a buffer memory arranged on the way of a data transfer bus. CONSTITUTION:When the reading instruction of data corresponding to one block e.g. is generated from a main processor 1a to a magnetic tape device control part 3, a local processor 32 commands a DMA controller 35 to transfer the data of one block obtained from the device 4 to a buffer memory 34 through a local bus 11. The data read out from the device 4 are stored in the buffer memory 34 and a main bus 10 is not accessed. The main bus 10 is not accessed until the data are transferred from the buffer memory 34 to a main memory 2.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a multi-processor system having a plurality of processors, and more particularly to control of a magnetic tape device in a multi-processor system.

[Prior art]

Conventionally, in this type of magnetic tape device control unit, when writing to a magnetic tape, the memory area to read and the memory area to write data read from the magnetic tape device are both on the main memory.
When instructed to perform each operation, connect the main memory and magnetic tape device directly via the main bus, and
It was configured to send and receive data using MA (Direct Memory Access) transfer.

Therefore, as is well known, in this method, the data transfer speed on the main bus is slower than that of magnetic tape devices that can write and read data at high speeds, such as magnetic disk devices. The magnetic tape device control section occupies the main bus, which is determined by the read and write speeds and is shared by other local processor sections and the main processor section. As a result, instructions cannot be exchanged between other local processors and the main processor, resulting in a waiting state.

Also, for example, if you try to read data from a magnetic tape device, but the data size for 2 liters is larger than the data write size secured in main memory by the main exposure processor, other data in main memory will be destroyed. Unnecessary data transfer occurs when data needs to be retransferred, such as by interrupting data transfer to avoid data transfer, securing memory size sufficient to write one block of data in main memory, and then transferring the same data again. May/e
There was a drawback that the operational efficiency of the entire system decreased due to the occupied time of the /< space.

[Purpose of the invention]

The purpose of the present invention is to eliminate access to the main bus by a magnetic tape device, which requires low-speed data transfer.
By making it possible to exclude the magnetic tape device and further eliminating access to the main bus caused by sources other than the main bus, it has the effect of eliminating the above-mentioned drawbacks and realizing efficient system operation. , K provides a magnetic tape device control method in a multi-processor system.

[Structure of the invention]

The feature of the magnetic tape device control system of the present invention is that the main bus used for data transfer between the magnetic tape device and the main memory, which was conventionally directly connected for the data transfer function, is replaced by the four-cal processor of the magnetic tape device control section. By separating the data transfer function within the control range and inserting an MA controller that can access both buses used for data transfer and a pack memory that can be accessed from both the local processor and its DMA controller. - Transfer data can be stored in the buffer memory, and the local bus between the buffer memory and the magnetic tape device is a conventional DMA bus for both buses used for separated data transfer. In the transfer method, the main bus between the other backup memory and the main memory is a memory bus.
Since each memory transfer method can be set independently, accesses to the main bus shared by other local processor units and the main processor unit can be reduced.

[Explanation of Examples]

Hereinafter, the present invention will be specifically described with respect to a case where it is applied to a multiprocessor system as shown in FIG.

In FIG. 1, (l) is the main processor section, (l) is the main memory section, (31 is the magnetic tape device control section in one of several I10 controllers,
(Fuji is a magnetic tape device.

The main processor section (1) is the main processor (1a)
The magnetic tape device control unit @) includes a local processor (82), local memory (8a), dual port RAM (81), buffer memory IJ (84) and D
It is composed of an MA controller 85.

Main processor (LA), main memory color), dual port RAM (81), local processor (82)
The DMA controllers (85) are connected to each other by a main bus (10), and the DMA controller (85) and magnetic tape device (11) are connected to each other by a local bus (11).
) K is connected.

In addition, the DMA controller (85), buffer memory (
84) and local memory (8B) are connected to the local processor (8B) by a local control line (12).

In this application example, the local processor (82) normally polls the dual port RAM (81), sends and receives operation instruction commands from the main processor (la), and when the operation is completed, the local processor (8
21) is an example of setting report data for the main processor (la) in the dual port RAM (81) and applying it to a method of displaying completion to the main processor (la). Since the operation is well known, further explanation will be omitted.

Now, from the main processor (LA) to the magnetic tape device control unit, for example,
When an instruction (ins) to read out data (at) for 1 minute is issued, the local processor (82) first
:The magnetic tape device (4) is connected to the controller (85).
1 to 1 block of data (dt) is received through the local path (11), and an instruction (Sl) to pass it to the buffer memory (to 0) is received via the local path (11), and the DMA controller (8
5) Start up the magnetic tape device (small).

As a result, the data (at) read from the magnetic tape device (small) is stored in the buffer memory (84), and the main bus (10) is never accessed.
During this time, the main bus (10) can be used for exchanging instructions and data between other local processors (not shown) and the main processor (la). Note that similar effects can be obtained when reoperating in response to a malfunction that occurs during this time.

Next, when reading from the magnetic tape device (4) is completed, the size of the read data (at) as well as the data read result can be known by reading the internal counter of the DMA controller (85). Therefore, the main processor (la) is the main memory Q)
You can compare the memory size reserved for the magnetic tape device (
If the read data size from the beginning exceeds the size secured in the main memory ω), stop transferring this read data (at) to the main memory ω) and immediately transfer it to the main processor (1a). This command (, 1 ns) is terminated by reporting that the transfer cannot be performed (82).

This eliminates the process of transferring unnecessary data to the main memory (color), which eliminates the need to occupy the main bus (10) required in conventional data transfer. It can be used for exchanging instructions and data between the main processor (la) and the main processor (la).

In addition, when the main processor (1a) instructs (ins) to read the same block of data (at) again as a subsequent instruction, the main memory color is immediately read from the backup memory (84) that was previously stored. Since the data can be transferred to the magnetic tape device (4), the trouble of reading it out again from the magnetic tape device (4) can be saved.

Furthermore, from the buffer memory (84) to the main memory (21)
When transferring data (at) to the DMA controller, the main bus (10) is accessed for the first time, and the local processor (82)
Instructions (S8) to send the data (at) read from the rough memory (34) and pass it to the main bus (10) (main memory color), and the DMA controller (
85).

By using a high-speed memory-to-memory transfer method as the transfer method used at this time, the magnetic tape device 4' (4+)
The transfer rate determined by , is much faster and the transfer time for the same data frame is reduced. This means that the time occupied by the main/outlet (lO) can be shortened.

The above description of the present invention describes an example of data transfer from the magnetic tape device (21) to the main memory (2), and conversely, data transfer from the main memory (21 to the magnetic tape device (4)).
), it is of course applicable to data transfer as well, so a description thereof will be omitted.

[Effects of the present invention]

As explained above, the present invention extends the low-speed data transfer section by the magnetic tape device to the buffer memory provided in the middle of the data transfer bus, and transfers the data between the nokura memory and the main memory to the high-speed memory. By making each memory transfer method independent, even when data is being transferred using a low-speed magnetic tape device, access to the main memory is high, so conventionally the main memory is placed in a waiting state and cannot be used. The main bus that previously existed can now be used, making it possible to exchange instructions between the main processor and other local processors.

Note that even if the local processor of the magnetic tape device control unit occupies the main bus, it is possible to perform memory transfer between rough memory and main memory, so if data was transferred by directly connecting to a conventional magnetic tape device. It goes without saying that it can be opened in a shorter time than in the previous case.

As long as the previous transfer data can be held in the buffer memory, the same data can be retransferred from the buffer memory, so if there is a problem with data transfer related to either one of the data transfer buses, the data You can recover normally by simply transferring the .

Furthermore, since the data size is known when data is read from the magnetic tape device and stored in the buffer memory, the read data size is compared with the data write size secured in the main memory by the main processor. If the memory size reserved in the main memory is found to be smaller, it is possible to avoid starting data transfer access to the main bus. This eliminates the unnecessary occupation time of the main bus, which conventionally occurs when the main bus is found to be in use.

Furthermore, since the time required for rereading in response to a rereading command from the main processor in subsequent processing can be reduced to just data transfer between the buffer memory and main memory, reading can be performed by restarting a conventional magnetic tape device. It has the effect of being able to save time compared to the case where the

[Brief explanation of the drawing]

FIG. 1 is a schematic block diagram showing an example of a multiprocessor system according to the present invention. 1: Main processor section. la: Main-processor. 2 = Main-memory section. 8: Control unit of magnetic tape device. 4=Magnetic tape device. 10: Main (transfer) bus. 11: Local (transfer) bus. 12 = Local control line. 82: Local processor. 84: Buffer memory. 35: DMA controller. Patent applicant: NEC Corporation

Claims (1)

[Scope of Claims] A multiprocessor system comprising a main processor unit, a main memory unit, and at least a local processor included in a control unit of a magnetic tape device, and in which a transfer bus connects these units, at least the above-mentioned magnetic tape. A buffer memory for temporarily storing transfer data is provided on a transfer bus connected to the device, and data transfer between the main memory section and the magnetic tape device is separated, and the data stored in the buffer memory is separated. A magnetic tape device control method for a multiprocessor system, characterized in that transfer is controlled by the local processor.