JPS6373351A - Scheduling method for look-ahead and write in mass - Google Patents

Abstract

PURPOSE:To continuously utilize a data transfer path and to improve an utilization factor by executing a scheduling for look-ahead and write in a mass every time total data transfer quantity between a channel and a control device reaches previously specified quantity. CONSTITUTION:A channel 12, which controls a data transfer between input/ output units (a cartridge type MT device illustrated as an example) 15a-15n specified according to a command from a CPU 10 and a main storage device 11, is connected with an MT control device 13, which has a buffer 14 and is connected to the units 15-15n through a data transfer path 16 and a controlled information path 12. And whenever the total data transfer quantity between the channel 12 and the device 13 reaches a scheduled quantity, the device 13 executes the next scheduling. In the scheduling the data quantity for look-ahead and write in a mass is decided to be the same quantity to the data transfer quantity after starting former scheduling as a basic value. And if excessive quantity of data transfer occurs in the units 15a-15n, the basic value is made to be corrected.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to the control of input/output devices in data processing systems, and in particular, to control of input/output devices in data processing systems, particularly those of the type that require a certain amount of time for pre-processing and/or post-processing of data transfer. Concerning control of multiple input/output devices.

[Conventional technology]

Certain input/output devices require a certain amount of time for pre-processing and/or post-processing of data transfer, and this pre-processing and post-processing is performed off-line with the control device. For example, cartridge-type magnetic tape devices do not have a buffer mechanism such as a vacuum column in order to be smaller and lighter.
Therefore, sudden starting and stopping cannot be performed. As a result, this type of magnetic tape device (hereinafter abbreviated as MT) requires start-up processing to drive the tape to a predetermined speed as pre-processing for data transfer. Repositioning to correct the problem requires treatment. This start-up process and repositioning process are performed off-line from the M and T control devices, and the time required for them is known.

The fact that the time required for startup processing is known means that any data transfer start time can be set by appropriately setting the startup processing start time. on the other hand,
If data transfer is performed in blocks and the amount of data in each block is known, it is possible to calculate the transfer end time of the MT currently transferring data. Under such conditions, as described in Japanese Patent Application No. 60-39868, the MT group is controlled so that the start-up process of other MTs is completed exactly at the transfer end time of the currently operating MT. By providing a mechanism, data transfer to a plurality of MTs can be performed continuously, and as a result, data transfer paths are effectively utilized and the overall processing speed of the system is improved.

[Problem that the invention seeks to solve]

The control mechanism does not cause any inconvenience when a suitable number of MT devices operate approximately equally. However, each MT is
Once the data transfer is interrupted, the data transfer cannot be restarted until at least the time required for repositioning and start-up processing has elapsed, so the control mechanism alone does not allow continuous use of the data transfer path. There may be cases where this cannot be achieved.

For example, for repositioning, the total time required for processing and start-up processing is Looms, and the transfer time for one block is 10m5.
If one MT transfers data one block at a time, the utilization rate of the data transfer path is 10×2/(100+10) L:0.18, or only about 18%.

To solve this problem, basically, a buffer is provided in the MT control device, and an appropriate number of blocks are read ahead and written in bulk through the buffer, so that the repositioning and startup processing of each MT can be done by others. This may be completed during the pre-read/collective write period for the various MTs. For example, in the above example, if 10 blocks each are pre-read and collectively written to both MTs, continuous use of the data transfer path is achieved. However, in an actual MT system, the number of active MTs changes dynamically, and the access frequency of those MTs, that is, the required data transfer amount in a certain period of time, is uneven, and the ratio also changes over time. Change. Therefore, in order to fully utilize the data transfer path effectively in practical use, it is necessary to dynamically determine the amount of data read in advance and written in batches for each MT, that is, to perform scheduling.

The present invention is applicable to a type of MT, such as a cartridge type MT, in which at least one of pre-processing and post-processing requires a non-negligible amount of known time, and these processes are performed offline from the control device. The purpose of this paper is to realize practical and sophisticated read-ahead/batch-write scheduling for output devices that can achieve continuous use of data transfer paths in the complex practical environment described above. do.

[Means for solving problems]

The present invention performs pre-read/collective write scheduling every time the total data transfer amount between a channel and a control device reaches a predetermined amount, and outputs the pre-read/collective write amount to each input/output device. The determination is made in relation to the amount of data transferred in the past between the channel for the device and the control device. For example, if the amount of data transferred between the channel and the control device for each input/output device after the previous scheduling is used as the basic value for the amount of data read ahead and written in batches for each input/output device, especially if an excessive amount of data is transferred, If there is a device that has been activated or if there is an increase or decrease in the number of active devices, corrections are made from the basic value.

[Effect]

According to the present invention, appropriate prefetch/collective write scheduling is performed according to the frequently changing operating status of the input/output system, and therefore the utilization rate of the data transfer path is always maintained at a high level. Furthermore, since scheduling is performed every time the total data transfer amount reaches a certain value, the required buffer size can be kept to a reasonable value.

〔Example〕

Prior to a detailed explanation of the embodiments, an overview of the present invention will be explained from a theoretical aspect using MT as an example.

In short, each MT performs its repositioning and startup processing.
It is necessary to complete the process within a time that does not exceed the total of the pre-read/collective write times of all other MTs. That is, the number of MTs in the active state is N, MT; (i=1,
2. ......, N)'s pre-read/collective write data amount is α5, the maximum among α1 to α8 is MAX (α,), M
If the data transfer rate of T is expressed by t and the time required for repositioning and startup processing is expressed by S, then when the following equation holds, continuous use of the data transfer path between the MT control device and the MT group is achieved. Ru.

The left side of equation (1) is the MT with the maximum amount of look-ahead/collective writing.
(hereinafter referred to as MT, A) is the sum of the amount of pre-reading and collective writing of all MTs. Therefore, equation (1) means that after MTMA finishes one pre-reading and collective writing, After completing the pre-reading and summary writing of all MTs, MT,
The time it takes for □ to start the next read-ahead/summary-write is
This means that it is no shorter than the time required for repositioning and startup processing. If formula (1) holds true, M T , A,
For any MTJ other than the Continuous execution of batch writing, in other words, continuous use of the data transfer path is possible.

In order to read ahead and write in batches, a two-sided buffering method is used in which buffering between the channel and MT control device and buffering between the MT control device and MT are performed in parallel. is effective. If the data transfer rates of the channel and MT are equal, the buffer and M
The data transfer between the buffer and the MT group is continuous as long as the amount of data transferred between the buffer i and the channel is at most 1 during the period in which Σ α1 data is transferred between T. This is because the transfer rate is the same as that of the channel. Therefore, the basics of buffer scheduling are as described above.

However, in actual MT systems, the following problems occur.

(i) When the access frequency of a certain MT reaches 111'', the required buffer size approaches infinity.

(ii) It is necessary to deal with frequent changes in the number of active MTs.

The basic idea of the present invention is that whenever the total data transfer amount between the buffer and the channel exceeds a predetermined value, (1)
The key point is that the scheduling of pre-reading and collective writing is carried out in a way that satisfies the formula.

Let E be the predetermined total data transfer amount. The value of E is preferably selected around 2 St. When the total amount of data transferred between the buffer and the channel reaches E, this E
The amount of data whose source or destination is MTi (hereinafter, the amount of data transferred between the buffer and channel is referred to as the amount of data movement) is βi (i=1.2.
N). According to the present invention, in principle, the prefetch/collective write amount α for each MT is set equal to β.

That is, α = β1...
(2) In such scheduling, if equation (1) holds true, after the scheduled prefetch/collective write is completed, the next prefetch/collective write can be performed successively. That is, the following equation must hold true.

E-MAXβn S-t...
(3) However, MAXβ represents the maximum value of βi.

If the access frequency of a certain MT is abnormally high, (3
) does not hold and therefore requires adjustment. (1
) can be established at the time when the total amount of data movement for all MTs other than the MT for which the data movement amount MAXβ was observed (hereinafter referred to as MAX MT ) reaches S·t.

Therefore, the amount of data movement for MAX MT at such a point in time is estimated by proportional calculation.

This ratio is equal to the ratio of E-MAXβ and S-t.

However, it is assumed that the access frequency of each MT does not change over time. Then, this estimated data movement amount is calculated as the pre-read/collective write amount MA for MAX MT.
Set as Xα. That is, MAXα=MAXβ・
S-t/(E-MAXβ)...(4) And for other MTs, at the time when the amount of read-ahead/collective writing of MAX MT reaches MAXα given by equation (4), Scheduling and execution sequences are adjusted so that the amount of prefetching and bulk writing equal to each amount of data movement is performed. in this way,
Even if equation (1) is not satisfied, the total data movement amount is E
When this is reached, the read-ahead/collective write process is started.

Therefore, there is no need to increase the buffer size too much, and problem (i) above is solved.

Next, scheduling when a change occurs in active MTs will be explained. In the following, write MT means an MT that performs a write operation, and read MT means an MT that performs a read operation.

(a) If there is a free area in the write MT's increase buffer, data transfer from the channel can be executed immediately. Therefore, for a light MT that is newly requested to be activated, the old active MT
Similarly, the amount of data movement βi at the time of scheduling may be set as the bulk writing amount α.

Data transfer between the buffer and the channel for this new write MT causes a change in the data movement amount of the old active MT, so the amount of pre-reading and collective writing must be corrected. However, regarding light MT,
At all times, only data actually written from the channel to the buffer is subject to batch writing, and therefore, the data movement amount β 2 may be set as the batch writing amount α according to basic scheduling. On the other hand, for the light MT, since the read-ahead was performed by prediction, correction is necessary. To be more specific, the look-ahead and
The bulk writing amount is set by the following formula.

α, = β, −α'i (m/ (n + m)) (γ
, -βi) ...(5) where n: Number of conventional active MTs m: Number of MTs whose activation was requested between the previous schedule and the current schedule α': Previous MTi Amount of read ahead γ,: Amount of data from MT that existed in the buffer at the time of the previous schedule The second term on the right side of equation (5) is an adjustment for the amount of excessive read due to the previous read ahead. The previous read-ahead amount α' is the data movement amount predicted on the assumption that there are n active MTs. However, now the number of active MTs has increased to n+m, and the access frequency of each conventional active MT should decrease accordingly. Therefore, the previous read-ahead amount α' was excessive with respect to the amount of data (data movement amount) that would actually be read to the channel between now and the next scheduled time. To estimate this excess data amount, at the time when the total data movement amount next reaches E (the next schedule point), the data movement amount for the new active MT group will be on average E/ (
n+m), and the total can be expected to be E-m/(n+m), which is the reduction in the amount of data movement for the conventional active write. This reduction amount can be considered to be distributed to the data movement amounts for the old active MTs in proportion to their old values. Therefore, Δα′
,=α'H(E-m/E (n+m))=α'・(
m/(n0m)) is estimated as the excess lookahead amount for MTi, and this is reduced from the basic value β of the lookahead schedule.

The second term on the right side of equation (5) is an adjustment of the resulting excessive lookahead amount by reducing the data movement amount β· from the expected value.

As a result of the previous look-ahead, the amount of data prepared in Batsubua at the time of the previous schedule was γ based on the prediction. However, the amount of data actually read out from there to the channel is β1. As a result, the difference γ, -β1 currently exists in the buffer as the resulting excess read-ahead amount. Therefore, this quantity is also the basic value β.

will be reduced from Furthermore, if the amount of data in the buffer at the time of this schedule is 7, then γ, = γ, −β
i+α′.

That is, γi−βi=γ, −α″i Therefore, 7.−α′i can be used instead of γ and −βi, and this is adopted in the embodiments described below.

(b) Increase in read MT M whose activation is newly requested due to read operation
For T, read-ahead from MT to buffer must be performed before reading from buffer to channel. Therefore, assuming that there is already an average data movement amount E/(n+m) for this new lead MT, the total data movement amount is updated, and when scheduling, the read-ahead amount of this new lead MT is calculated based on the above assumption. The data movement amount is set to E/(n+m). The amount of read ahead of the conventional active read MT is determined according to equation (5). The bulk write amount of the write MT is set to the actual data movement amount β1, as in the case (a), and there is no need to modify it.

(Q) Decrease in write MT In this case, the bulk write amount of the conventional active write MT is set to the actual data movement amount β, and there is no need to modify it. However, the read MT requires modification. The reason for this is that the reduction in the number of active MTs creates a margin in the utilization rate of the data transfer path, and as a result, the access frequency of the remaining active MTs may increase. That is, the decrease in total data movement that would result from a reduction in the number of active MTs is offset by the increase in data movement for the remaining active MTs, and each MT
The increase in is expected to be proportional to the previous amount of data movement. Therefore, if the set of MTs released from the active state is represented by 0, the amount of read ahead for the remaining read MTs is: This is to compensate for the shortcomings that may occur.

(d) Reduction of lead MT In this case, the same measures as mentioned in paragraph (C) are taken for the same reasons.

If the above (a) to d) are combined, active MT
The pre-read/collective write amount of the conventional active MT when there is a change in the amount of data is set as follows.

Lead MT = α1 = β, - (γ1 - β,) - α'7 (m + / (n m + - m -)) + (Σα
′)・β・/E・・・・・・(7)jEθ where m+: MT newly requested to be activated
Number of MTs m-: Number of MTs newly released from active state Light MT: αi=β1 ・・・・・・(7
') FIG. 1 shows an overview of the above-mentioned scheduling in the form of a flowchart. That is, in step 1,
When the total data movement amount reaches a predetermined value E, prefetch/collective writing scheduling is started, and step 2
If there is no change in the active MT and no MT that does not satisfy equation (3) due to excessive access frequency is found in swing 3, pre-reading and collective writing using equation (2) cannot be performed in step 4. However, if there is a change in the active MT, prefetching and collective writing are performed in step 5 using equations (7) and (7'), and on the other hand, if the active MT does not satisfy equation (3),
If it is discovered, in step 6, the pre-reading and summary writing as described in relation to equation (4) is performed.

Examples of the present invention will be described in detail below. FIG. 2 shows an overview of an example of a computer system to which the present invention is applied. CPU
l0 accesses main memory 11 to execute programs stored therein, and calls input/output devices via channel 12 as necessary. Channel 12
0 controls data transfer between the specified input/output device and the main storage device 11. Channel 12 is connected to MT control device (hereinafter simply referred to as control device) 13. The control device 13 has a buffer 14 and is connected to a plurality of cartridge type MT devices 15a to 15n through a data transfer path 16 and a control information path 17.

As described above, in a cartridge type MT device (hereinafter simply referred to as MT), startup processing and repositioning require a certain amount of time, and these processings are performed off-line from the control device. As detailed in the above-mentioned Japanese Patent Application No. 60-39868, mechanisms for continuously advancing data transfer to successive MTs include a type in which each MT is provided with a timer, and a mechanism in which a start-up request is made asynchronously with the data transfer operation. There are two types: one that provides a dedicated control information path for issuance. In this embodiment, the latter type is used, and the control information path 17 includes the dedicated control information path described above. However, it goes without saying that the former type of mechanism may also be used.

The control device 13, as shown in FIG.
A processor A20 performs data transfer control and overall control between the buffer 14 and the buffer 14;
Processor B that controls data transfer between 5a and 15n
21 and a memory 22 shared by these processors. The processor B21 includes a timer 23 for controlling the timing of issuing an MT startup request, and a timer 23 for controlling the timing of issuing an MT startup request,
5 a to 15 n and the buffer 14 . The shared memory 22 has areas for an MT management information section 25, a prefetch/collective write MT information section 26, a buffer management information section 27, and a prefetch/collective write control information section 28.

FIG. 4 shows the format of the MT management information section 25.

A set of such management information is prepared for each MT. MTID30 is the identification information of the corresponding MT,
The T state 31 indicates the state of this MT (active, rewind, etc.). Buffer address 32 indicates the start address of a buffer area used for data transfer between buffer 14 and channel 12 for this MT.

In other words, for an MT in a read operation, it indicates the beginning of the buffer area that holds data that has been read ahead but has not yet been read out to the channel, and for an MT in a write operation, it indicates the beginning of the buffer area that holds data that has been read ahead but has not yet been read out to the channel. Indicates the beginning of the buffer area that holds data that has not yet been written to the MT. The read/write state 33 indicates whether this MT is in a read operation state or a write operation state. The data movement amount 34 indicates the amount (βi) of data transferred (read or written) between the channel 12 and the buffer 14 for this MT after the previous schedule time. As described in relation to equations (3) and (4), the summary writing addition flag 35 indicates that the access frequency of this MT is excessive,
This flag is turned on when part of the data transferred from the channel after the scheduled time is also subject to batch writing, and the additional batch writing amount 36 indicates how much data will be written in the future in the above case. should be added to. The previous read ahead/collective write amount 37 is the read ahead/collective write amount (α
′, ). This information is used to determine the read-ahead amount of the lead MT using equation (5) when a change occurs in the active MT at the next schedule time. The wait flag 38 indicates whether or not this MT is in a wait state. The buffer amount 39 indicates the amount of data existing in the buffer, and is used to determine the prefetch amount of the read MT using equation (5).

FIG. 5 shows the format of the prefetch/collective write MT information section 26. A set of such information is prepared for each MT. MTID4.0 is identification information of the corresponding MT. The pre-read/collective write amount 41 indicates the data amount of pre-read or batch write determined by scheduling for this MT. The pre-read/collective write buffer address 42 indicates the start address of the buffer area to be used for pre-read/collective write. The buffer shortage flag 43 is a flag that is turned on when pre-reading is started when the entire buffer capacity required for pre-reading cannot be secured for the read MT. The prefetch/collective write MT pointer 44 points to the prefetch/collective write MT information section 26 corresponding to the next MT in the prefetch/collective write sequence. The pre-read/collective write sequence for successive MTs is performed cyclically, and this cyclic sequence is defined by cyclic chaining by the pre-read/collective write MT pointers 44a to 44m, as shown in FIG. The chain formed by the pointer 44 in this manner will hereinafter be referred to as a read-ahead/collective write chain. Look ahead first
The prefetch/collective write MT information field corresponding to the MT on which collective writing is performed is specified by the head prefetch/collective write pointer 611 in the prefetch/collective write control information section 28, which will be described later. The mount request flag 45 and rewind request flag 46 indicate that mounting and rewinding of this MT are requested, respectively, and the mount completion flag 47 and rewind completion flag 48 indicate that they have been completed, respectively. The save buffer address 49 is a field in which the read-ahead/collective write buffer address 42 is saved after this MT start-up process is requested. The access frequency MT flag 50 is a flag indicating that the access frequency of this MT is too high and that it is necessary to correct the amount of pre-reading and bulk writing using equation (4).

The format of the buffer management information section 27 is shown in the upper left part of FIG. The buffer capacity 51 is equal to the buffer 1
The total capacity of 4 is shown. The free buffer amount 52 indicates the capacity of the buffer that is currently free.

The empty buffer pointer 52 points to the start address of the currently empty buffer area. As shown in FIG. 7, the buffer 14 has a large number of areas 5 of a fixed length called slots.
This slot is divided into four slots, and this slot forms the unit of buffer area allocation. An empty slot is an empty buffer pointer 5
3, and the buffer area and pre-read/collective write area for each MT are chained with the buffer address 32 and pre-read/collective write buffer address 42, respectively, by pointers, and thereby the buffer area for various MTs is can be allocated dynamically.

FIG. 8 shows the format of the prefetch/collective write control information section 28. The number of active MTs 600 indicates the total number of active MTs at present, and the number of read active MTs 601 and the number of write active MTs 602 indicate the number of each MT that is currently active for read processing and write processing. . The prefetch/collective write flag 603 is a flag that is turned on when the processor A 20 issues a prefetch/collective write request to the processor B 21 and is turned off when the processor B completes the request, and the prefetch/collective write completion flag 604 is This flag is turned on when processor B completes read-ahead/collective writing, and turned off when processor A receives notification of its completion. The total data movement amount 605 is the total amount of data transferred between the channel 12 and the buffer 14 after the previous scheduling time, and is equal to the sum of the data movement amounts 34 of all MTs. The schedule data amount 606 indicates a reference value (E) for starting the schedule when the total data movement amount 605 exceeds this value. Increased number of lead MT units: 60
7. Increased number of write MTs: 608, decreased number of read MTs: 6
09 and the number of write MTs decreased 610 respectively indicate the numbers of read MTs and write MTs that have increased or decreased since the time of the previous schedule.

As shown in FIG. 6, the head pre-read/collective write MT pointer 611 indicates the head of the pre-read/collective write chain, that is, the MT point where the pre-read/collective write is performed first.
It refers to the prefetch/collective write MT information section 26 corresponding to T. The schedule waiting flag 612 indicates a state in which the schedule is kept waiting even though the total amount of data movement has reached the value at which scheduling should be performed. The access frequency flag 613 indicates that one of the active MTs has an excessive access frequency and that the amount of pre-reading and bulk writing needs to be corrected using equation (4). Access frequency post-processing flag 61
4 indicates that when there is an MT whose access frequency is too high, post-processing for adjusting the amount of pre-reading and collective writing is being executed. The start-up process MT pointer 615 points to the read-ahead/collective write MT information section 26 corresponding to the MT on which the start-up process is to be performed during execution of read-ahead/collective write. Startup processing time 616 and repositioning processing time 61
7 indicates the respective times required for the MT startup process and repositioning. In a cartridge type MT, the processing time for repositioning is usually about four times longer than the start-up processing time. These times for a cartridge type MT are constant, but for the present invention, they only need to be known at the time of scheduling and do not need to be constant. Data transfer rate 618
indicates the data transfer rate of MT. The decreased MT total data movement amount indicates the total amount of data movement that was planned for the decreased read MT and decreased write MT, and the flow of processing by the processor A20 and processor B21 is explained in the previous read-ahead/collection of these MTs. do. 9A to 11 are flowcharts of processing by processor A, and FIGS. 12A to 12C are flowcharts of processing by processor B.

As shown in FIG. 9A, when processor A receives a processing request from channel 12 (800), processor A receives various processing completion notifications from processor B, such as mount processing completion notification (801), rewind Processing completion notification (
a O2) or prefetch/collective write processing completion notification (80
3) starts the operation when received. The processing when a processing request is received from a channel is shown in FIGS. 9B and 9C.

Referring to FIG. 9B, processing requests from channels are classified into three types: rewind requests, read requests, and write requests. Although several other types of processing requests are actually included, they are omitted because they are not directly relevant to the present invention. In step 804, it is determined whether rewind processing is requested. Processing when rewind processing is requested will be described later. If it is not a rewind request, it is a read request or a write request in this embodiment. Step 805 checks whether this read request or write request is for a new MT, and
If the request is for T, the process jumps to step 814 (FIG. 9C). Processing after step 814 will be described later. If it is not a request for a new MT, step 80
At step 6, the schedule wait flag 612 is checked, and if it is on, the wait flag 38 of the requested MT is turned on at step 816, and the process ends. If the schedule wait flag is not on, step 807
Then, it is determined whether the request is a read request or a write request.

In the case of a write request, an appropriate number of empty slots 54 are secured in step 808, and one block of data is transferred from the channel 12 thereto. The free buffer amount 52 and the free buffer pointer 53 are updated. Next, step 8
In step 09, the bulk write addition flag 35 of the requested MT is checked, and if it is off, in step 811, the slot that has stored the data from the current channel is linked to the end of the slot sequence pointed to by the buffer address 32. The data movement amount 34 and the total data movement amount 605 are increased by this amount of data. However, if the bulk write addition flag is on, in step 810, the slot storing the data from the channel is linked to the end of the slot string pointed to by the read-ahead/batch write buffer address 42, and the additional bulk write amount 36 is reduced by this amount of data. If the reduced additional batch writing amount is 0'' or less, the batch writing addition flag 35 is turned off.

On the other hand, in the case of a read request, in step 812, one block of data from the beginning of the slot sequence pointed to by the buffer address 32 is transferred to the channel, the slot in which this data was stored is released, and,
Buffer address 32 is updated. Further, the data movement amount 34 and the total data movement amount 605 are increased by this amount of data, and the buffer amount 39 is decreased by the same amount. After completing steps 810, 811, or 812, the process proceeds to the process shown in FIG. 10 in order to determine whether or not the data transfer has reached a state where pre-read/collective write scheduling should be performed.

FIG. 10 shows the flow of processing in prefetch/collective write scheduling. First, in step 900, the access frequency flag 613 is checked. If it is on, as will be described later, special read-ahead/collective writing for MTs that are accessed too frequently is being executed, so the process ends immediately. Next, in step 901, the total data movement amount 605 is calculated as the schedule data amount (E) 606.
It is checked whether or not it has been reached, and if it has not been reached, the process ends. If the total amount of data movement has reached the scheduled amount, it is checked in step 902 whether the pre-read/collective write flag is on and the pre-read/collective write completion flag is off. If this condition is met, the prefetch/collective write process is in progress, and therefore, in step 903, the schedule wait flag 612 is
is turned on and the process ends. On the other hand, if the conditions are not met, scheduling is initiated.

First, in step 904, it is determined whether or not there has been a change in active MT after the previous schedule time.
This is determined by checking the number of T units 607 or the number of reduced light MT units 610, and if there is a change, a jump is made to step 918. If there is no change in the active MT, then in step 905 it is checked whether there is an MT that satisfies the following formula.

(Data movement amount 34) 〉(Schedule data amount 606) - [(Data transfer rate 618) This is an MT that has an excessive access frequency that does not satisfy equation (3) (hereinafter referred to as an excessively accessed MT). If an excessively accessed MT is found, a jump is made to step 912. On the other hand, if there is no excessively accessed MT, basic scheduling using equation (2) is performed in steps 906 and subsequent steps.

First, in step 906, the read-ahead/collective write amount 41 of the MTs linked in the read-ahead/collective write chain (Fig. 6) is set to a value equal to the data movement amount 34, and then, in step 907, these The MTs are classified into a read MT group and a write MT group, and a list for each group is created. In step 908, for each read MT, the amount of read ahead is 41.
The required number of slots 54 are secured and linked to the prefetch/collective write buffer address 42. If the required number of slots cannot be secured, the buffer shortage flag 43 is turned on. In step 909, each light M
For T, the slot linked to the buffer address 32 (the slot that holds data transferred from the channel) is the read-ahead/collective write buffer address 42.
linked to. Next, in step 910, the data movement amount 34 is transferred to the previous prefetch/collective write amount 37, and then the data movement amount 34 and the total data movement amount 605 are cleared. Finally, in step 911, the prefetch/collective write flag 603 is turned on, and a prefetch/collective write request is issued to the processor B21.

If an excessively accessed MT exists, scheduling as described for equation (4) is performed. By the way, as mentioned above, the prefetch/collective write processing sequence is performed cyclically along the chain as shown in Figure 7, and the next MT that last operated in the previous prefetch/collective write process is This time's preview from the
Summary writing begins. Therefore, in many cases, there are MTs that must be pre-read and collectively written before over-accessed MTs. Therefore, first, a pre-read/collective write is performed for these preceding MTs in an amount equal to the data movement amount at the time of scheduling, and for excessively accessed MTs, a pre-read/collective write is performed in an amount calculated by equation (4). will be carried out. Then, in the automatically started post-processing, for the MTs in the order following the over-accessed MT, the amount of data transferred from the previous schedule to the time when the pre-reading and collective writing of the over-accessed MT is completed. The amount of read-ahead/collective writing is performed equal to
As a result, the amount of read-ahead/collective writing for MTs other than the over-accessed MTs will be increased from the time of the previous schedule when the amount of read-ahead/collective writing is equal to the amount of data moved up to the point when the pre-reading/collective writing is completed. It is equal to the amount of data moved up to the point in time when MT's pre-read/collective write is completed.

First, in step 912, the pre-read/collective write amount 41 of each MT that should be pre-read/collectively written before the over-accessed MT is set to a value equal to the data movement amount 34. Next, in step 913, the data movement amount of the excessively accessed MT is corrected according to the following equation, and this value is set as the pre-read/collective write amount.

Amount of read ahead/collective writing 41 = (data movement amount 34) × (data transfer rate 618) × ((startup processing time 616) + (processing time for repositioning 617)) ÷ ((schedule data amount 606) −( Data movement amount 34)) (8) Next, in step 914, the access frequency MT flag 50 of the excessively accessed MT is turned on, and the access frequency flag in the prefetch/collective write control information section 28 is further turned on. 613 is also turned on. Then, in step 915, it is checked whether the excessively accessed MT is a write M or T. If it is a write MT, in step 916, the collective write addition flag is turned on, and the pre-read/collective write amount and the The difference in data movement amount is stored in the additional bulk write amount 36. Subsequently, in step 917, the excessively accessed MTs and the MTs to which pre-reading/collective writing should be performed are classified into a read MT group and a write MT group, and a jump is made to step 908. Thereafter, the processes of steps 908 to 911 are performed on these MTs, and a prefetch/collective write request is issued. However, step 9
Clearing the data movement amount in 10 is due to excessive access M
This is done only for T and the MT that precedes it. Therefore, the prefetch/collective write process that has been started is performed only for the MTs that precede the overaccessed MT in order and the overaccessed MT, and ends when the prefetch/collective write for the overaccessed MT is completed. This will be followed by post-processing, which will be detailed later.

If there is a change in the active MT after the previous schedule, the amount of read-ahead/collective writing of the write MT (including newly activated ones) is the amount of data actually transferred from the channel. As in the case where there is no change, the determination may be made through the process in step 906. However, the amount of read-ahead/collective writing of the conventional read MT must be calculated using equation (7). To do this, in step 918, the data movement amount 34 of each lead MT whose MT state 31 indicates the active state is corrected using the following equation.

Parameter a = (Data movement amount 34) - ((Buffer amount 39) - (Previous read ahead/collective write amount 37))
9) Parameter b = ((Increased number of read MTs 607) + (Increased number of write MTs 608)) ÷ ((Number of active MTs 600) + (Increased number of read MTs 607) + (Increased number of write MTs 608) - (Decreased lead Number of MTs: 609) - (Reduced number of light MTs: 610) × (Previous read ahead/collective write amount: 37) ...... (10) Parameter C = (Reduced MT total data movement amount: 619) × (Data movement amount: 34 ) ÷ (Schedule data amount 606) ...... (11) Modified data movement amount 34 = (Parameter a) - (Parameter b) + (Parameter C) ...... (12
) In addition, in parameter a ((buffer amount 39) -
(Previous read-ahead/collective writing amount 37)) is (γi-βi) in equation (5), as mentioned in the explanation of equation (5).
is equivalent to

Next, in step 919, the MT whose MT state 31 indicates the data waiting state is linked into the read-ahead/collective write chain (FIG. 6). The MT in the data waiting state is
This is a lead MT whose activation was requested after the previous scheduled time.

For this MT, the processing after mounting is completed (described later)
In step 824), the data movement amount 34 is set to the expected value. Then, in step 920,
Number of active MTs increased to 600. Number of lead MTs increased to 607.
and the increased number of read MTs 608, and subtract the decreased number of read MTs 609 and the decreased number of write MTs 610, and clear the increased and decreased MT numbers 607 to 610 and the decreased MT total data movement amount 619. Then, a jump to step 906 described above is performed, and the prefetch/collective write amount 41 is set to a value equal to the data movement amount 34.

Thereafter, the processes 907 to 911 described above are performed, and finally, a prefetch/collective write request is issued to processor B.

In addition, to explain the MT to which a rewind request has been issued, in the case of a read MT, when a rewind request is issued, the MT is immediately removed from the read-ahead/write-at-all chain as described later, and as a result, the MT becomes the target of scheduling. naturally excluded from. In addition, in the case of a write MT, it is necessary to write the data transferred from the channel to the buffer before the rewind request to the MT.
As far as collective writing is concerned, the same processing as active MT is performed.

Next, a description will be given of processing when a new read request or write request is issued to an MT to which no access request has been made so far. Such a new request is detected in step 805 of FIG. 9B and processed in step 814, as described above.
causes a jump to. Referring to Figure 9C, step 8
14, the MT management information section 25 and pre-read/collective write MT information section 26 are prepared for this newly requested MT, and the identification information of this MT is stored in its MTIDs 30 and 40, and read/write operations are performed. A state 33 is set according to the type of request, and an MT state 31 is set to the mounted state. Next, in step 815, the mount request flag 45 is turned on and a mount request is issued to processor B.

A rewind request from a channel is detected in step 804 of FIG. 9B, resulting in a jump to step 817. Referring to FIG. 9C, in step 817, the previous read-ahead/collective write amount 37 of the MT for which rewind is requested.
is added to the reduced MT total data movement amount 619. Next, in step 818, it is determined whether this MT is a read MT or a write MT. In the case of a read MT, in step 819, the MT state 31 is set to the rewind state, all linked slots are released, the reduced number of read MTs 609 is increased by LL I I+, and furthermore, this MT is set to the rewind state.・Excluded from the summary chain. Subsequently, in step 820, rewind request flag 4 is set.
6 is turned on, and a rewind request is issued to processor B. On the other hand, in the case of a write MT, rewind cannot be executed until after the data that has been transferred from the channel to the buffer has been written by the next read-ahead/collective write process.

Therefore, in step 821, MT state 3
1 is set to a rewind wait state, the decreased number of write MTs 610 is increased by 1'', and the process ends here.

Next, the processing performed by the processor A20 when receiving the processing completion notification from the processor B21 will be described. Figure 9A
, when a mount completion notification is detected in step 801, the mount completion flag is turned off in step 822.

This flag is turned on when processor B confirms the completion of mounting, as will be described later. Then, in step 823, it is determined whether this MT is a read MT or a write M. In case of lead MT, step 8
24, the number of increased lead MTs 607 was increased by 1",
The MT state 31 is set to a data waiting state, and the data movement amount 34 is set to a value obtained by dividing the schedule data amount 606 by the number of active MTs 600 plus 1″′, and furthermore, the total amount is set by this value. Data movement amount 60
5 is increased. Then, a jump is made to step 813 (FIG. 9B) described above in order to check whether the prefetch/collective write schedule condition is satisfied by updating the total data movement amount.

On the other hand, if the mounted MT is a write MT, in step 825, the number of additional write MTs 608 is incremented by "1", the MT state 31 is set to the active state, and this MT is in the read-ahead/collective write chain. Linked. A jump is then made to step 808 to transfer one block of data from the channel to the buffer.

When a rewind completion notification is detected in step 802, the rewind completion flag (which was turned on when processor B confirmed the rewind completion, as described below) is turned off in step 826; Channel 12 is notified of rewind completion. Thereafter, the MT management information section 25 and the prefetch/collective write MT" information section 26 for this MT are cleared.

Finally, when a prefetch/collective write completion notification from processor B is detected in step 803, step 82
At 7, the procedure shown in FIG. 11 is called. First, in step 1000, the prefetch/collective write completion flag 604 and the access frequency post-processing flag 614 are turned off. The read-ahead/collective write completion flag is turned on when processor B completes read-ahead/collective write, as described later, and the access frequency post-processing flag is
Step 1006, which will be described later, starts the post-processing of read-ahead/collective writing when there is an excessively accessed MT.
It was turned on. Next, in step 1001, an MT whose MT state indicates a rewind waiting state and whose data movement amount 34 is 110 II is searched. For such MTs, pre-reading and bulk writing are performed after step 821 (C in FIG. 9), and as a result, write MTs in which writing of the last data transferred from the channel has been completed.
It is. Therefore, a rewind request for such MT is issued to the processor.

Subsequently, in step 1002, the MT states 31 of these MTs are set to the rewind state, and further, in step 1003, these MTs are removed from the read-ahead/collective write chain. Next, in step 1004, it is checked whether the schedule wait flag 612 is on, and if it is on, a jump is made to step 1010. If the schedule wait flag is off, it is checked in step 1005 whether the access frequency flag 613 is on, and if it is off, the process advances to step 1015, which will be described later.

If the access frequency flag is on, excessive access MT
Post-processing of pre-reading and collective writing in the case where there is exists is performed in step 1006 and subsequent steps. First, in step 1006, the access frequency flag 613 is turned off,
The access frequency post-processing flag 614 is turned on. In step 1007, the read ahead/collective write amount 41 is set to the value of the data movement amount 34 for all MTs other than the excessively accessed MTs (that is, the MTs for which the access frequency MT flag 50 is turned on). Then step 10
In step 08, all MTs other than the excessively accessed MTs are classified into a read MT group and a write MT group, and then step 1009
calls the procedure from step 9.08 shown in FIG. As a result, all MTs other than excessively accessed MTs
As mentioned above, for the MT that precedes the over-accessed MT in the pre-read/collective write order, the pre-read/collective write of the excessively accessed MT is completed from the current schedule. For MTs that follow an over-accessed MT, it is equal to the amount of data moved from the time of the previous schedule to the time when pre-reading and collective writing of the over-accessed MT is completed.

If the schedule wait flag 612 is on, the schedule wait flag is turned off in step 1010;
In order to execute the awaited prefetch/collective write scheduling, the procedure shown in FIG. 10 is called in step 1011. Thereafter, in step 1012, if there is an MT whose wait flag 38 is turned on, that flag is turned off. Such an MT is placed in a wait state in step 816 (FIG. 9B) in response to a read request or a write request that occurs during a schedule wait state. Then, if there is a write MT among such MTs, in step 1013, the processes of steps 808 to 811 (FIG. 9B) are performed on that MT, and one block of data is transferred from the channel. Also.

If there is a lead MT, in step 1014, the process of step 812 (FIG. 9B) is performed on that MT,
One block of data is transferred to the channel.

In the next step 1015, an MT whose MT state 31 indicates a data waiting state is searched for.

Such an MT receives a new read request between the time of the scheduling for prefetching/collective writing that has just ended and the previous scheduling time, and at that time, the MT performs step 8.
24 and is placed in a data waiting state. For such an MT, in step 1016, the data movement amount 34 is changed from the scheduled data amount 606 to the active MT.
The total data movement amount 605 is increased by this value, and the MT state 31 is changed to the active state. Subsequently, in step 1017, the process of step 812 is performed. done, 1
The block's data is transferred to the channel. Step 1
The procedures following 015 are performed even when the access frequency flag 613 is off.

Next, processing by processor B21 will be explained.

Processor B starts processing from processor A20.
When a processing request is received from one of the MTs 15a to 15n, when a timer completion notification is received from the timer 23, or when a data transfer completion notification is received from the data transfer device 24. Either when you receive it or when you receive it.

Referring to FIG. 12A, steps 1100, 1101 and 1102 detect a mount request, a rewind request, and a read ahead/collective write request from processor A, respectively. Step 1103 detects a timer completion notification. Step 1104 detects various completion notifications from the MT. The completion notification from the MT includes a mount completion notification, a rewind completion notification, and a start-up processing completion notification. Step 1105 detects a data transfer completion notification from the data transfer device.

When a mount request or a rewind request is received from processor A, steps 1106 or 110 are performed, respectively.
At 7, the corresponding request is sent via the control information path 17.
Issued to the applicable MT. When a prefetch/collective write request is received, in step 1108, the start-up processing MT pointer 615 is set to the value of the first pre-read/collective write MT point 611, and the start-up processing MT pointer is started toward the MT pointed to. An increase request is issued. Next, in step 1109, the time at which it is decided whether or not to issue a start-up request to the next MT is calculated based on the pre-read/collective write amount 41 of this MT, and the timer completion notification is received at this time. A timer 23 is set. An example of the method of calculating this time is
No. 9868. Next, step 111
0, the contents of the prefetch/collective write buffer address 42 are moved to the save buffer address 49, and the process ends.

The process performed when a timer completion notification is received is shown in FIG. 12B. First, in step 1111, the access frequency flag 613 is checked, and if it is on, a jump is made to step 111'7, which will be described later. Next, in step 1112, access frequency post-processing flag 6
14 is checked, and if it is on, a jump is made to step 1120, described below.

If both flags are off, in step 1113, the contents of the start-up processing MT pointer 615 are replaced with the contents of the read-ahead/collective write MT pointer 44 of the MT it points to. Then, in step 1114, this new content of the startup processing MT pointer is
It is compared with the contents of T pointer 611. If they match, the prefetching/collective writing for all predetermined MTs has been completed, and therefore, in step 1115, the prefetching/collective writing is completed.
The collective write flag 603 is turned off, the prefetch/collective write completion flag 604 is turned on, and the processor A20 is notified of the completion of prefetch/collective write. On the other hand, if the new contents of the start-up process MT pointer and the contents of the head prefetch/collective write pointer do not match, step 1116
Then, a startup processing request is issued for the next MT pointed to by the new contents of the startup processing MT pointer, and
The timer 23 is newly set in order to determine the time for determining whether or not it is necessary to start up the subsequent MT.

If the access frequency flag 613 is on, the first half of the prefetch/collective write when there is an excessively accessed MT is in progress, and will end when the excessively accessed MT (the access frequency MT flag 50 is on) ends. There must be. Therefore, in step 1117, the startup process MT
Access frequency MT of MT pointed to by pointer 615
Flag 50 is checked and if it is on, step 1
At 118, the contents of the read-ahead/collective write MT pointer 44 of this MT are transferred to the head read-ahead/collective write MT pointer 61.
Replace the contents of 1. As a result of this replacement, the head read ahead/collective write MT pointer 611 points to the next MT of the excessively accessed MT. Then, step 111 described above
A jump to 5 is performed, and the first half of the pre-reading and summary writing ends. On the other hand, if the access frequency MT flag is off, in step 1119 the contents of the start-up processing MT bonneter 615 are replaced with the contents of the read-ahead/collective write MT pointer 44 of the MT it points to, and then in step 1116 described above. At this point, a start-up request for the next MT is issued55).

If the access frequency flag 613 is off and the access frequency post-processing flag 614 is on, pre-read/collection post-processing is in progress when there is an excessively accessed MT, which means that the It must end when MT ends. Therefore, step 11
20, the contents of the start-up process MT pointer 615 are updated in the same way as in steps 1113 and 1119, and then, in step 1121, the access frequency MT flag 50 of the MT pointed to by the updated start-up process MT pointer is checked. . If it is on, the next MT is an overaccess MT. Therefore, in step 1122, the contents of the first read-ahead/collective write MT pointer 611 are replaced with the contents of the start-up processing MT pointer 615 (which points to the excessively accessed MT), and then a jump is made to step 1115. , the post-processing ends. On the other hand, if the access frequency MT flag is off,
In step 1116, a startup request for the next MT is issued.

When various completion notifications from the MT are detected in step 1104, a jump is made to step 1123 shown in FIG. 12C. In step 1123,
Completion notifications are analyzed. In the case of a mount completion notification or a rewind completion notification, in step 1124, the mount completion flag 4 of the corresponding MT is set according to the contents of the completion notification.
7 or the rewind completion flag 48 is turned on and the respective completions are notified to the processor A20. On the other hand, if startup is complete, in step 1125, the required number of slots are secured for the MT whose buffer shortage flag 43 is on, and then the type of operation (read/write) and slot address are secured. Necessary information such as a list is transmitted to the data transfer device 24, and a data transfer request is issued.

When a data transfer completion notification is received from the data transfer device 24, it is determined whether it is a read or a write in step 1126 in FIG. 12C. For light processing,
In step 1127, the preread/collective write amount 41 is reduced by the amount of transferred data, and the empty slot is released. In case of read processing, step 112
In step 8, the slot that accommodated the transferred data is linked to the end of the slot string linked to the buffer address 32, and the read ahead/collective write amount 41 and the buffer amount 39 are
They are respectively decreased and increased by the amount of data transferred. In either case, the read ahead/collective write amount 41 is checked in step 1129, and if it is not "0", a jump is made to step 1125 described above, and a transfer request for the next block is issued. be done. However, if the amount of read ahead/collective write is "0", the scheduled read ahead/collective write for this MT has finished, so
Processing ends.

〔Effect of the invention〕

According to the present invention, continuous use of a data transfer path for a group of input/output devices of a type that requires a considerable amount of time for preprocessing and/or postprocessing, such as a cartridge type magnetic tape device, can be dynamically changed. It is easily achieved under such conditions. In particular, if there is an extreme imbalance in access frequency, or if the numbers of devices performing input operations and devices performing output operations change frequently and irregularly, the data transfer rate will be sufficiently high. The road utilization rate can be maintained at all times. Therefore, even if relatively low-grade equipment of the above-mentioned type is used, the time required for pre-treatment and post-treatment is negligible, and the total processing time is as high as when using high-grade equipment (for example, an MT with a vacuum column). performance can be obtained.

[Brief explanation of the drawing]

FIG. 1 is a flowchart showing the concept of the present invention, FIG. 2 is a block diagram of an example of a computer system having a cartridge type MT group, FIG. 3 is a block diagram of the MT control device in FIG. 2, and FIG. FIG. 5 is a diagram showing the format of the MT management information section in FIG. 3, FIG. 5 is a diagram showing the format of the pre-read/collective write MT information section in FIG. 3, FIG. 6 is a diagram showing the pre-read/collective write chain, and FIG. 7 is a diagram showing the format of the MT management information section in FIG. is a diagram showing the link form of the buffer management information section and the buffer slot in FIG. 3, FIG. 8 is a diagram showing the format of the prefetch/collective write control information section in FIG. 10 is a flowchart showing the basic processing of processor A in the figure, FIG. 10 is a flowchart of prefetch/collective writing scheduling, FIG. 11 is a flowchart of processing of processor A after completion of prefetching/collective writing, and FIGS. 3
3 is a flowchart of processing by processor B in the figure. 1...Step of checking whether the total amount of data movement has reached the amount that requires prefetch/collective write scheduling, 4...Basic scheduling, 5...Scheduling when there is a change in active MT ,6・
Scheduling when there is an MT with excessive access frequency, 12... Channel, 13... MT control device, 14... Buffer, 15a to 15n... Cartridge type MT.

Claims (1)

[Claims] 1. A channel, a plurality of input/output devices, and a control device for controlling data transfer between the channel and the plurality of input/output devices, wherein the channel has access to the input/output devices. The input/output device issues a request in units of data blocks, and the input/output device performs at least one of pre-processing and post-processing of data transfer offline with the control device, and requires a known length of time to do so, and the control device In an input/output system that has a data transfer buffer and performs pre-reading and batch writing to the input/output devices, the amount of data read in advance and batch writing to each of the input/output devices requested to be accessed from the channel is calculated. A scheduling step that is determined in relation to the past data transfer amount between the channel and the control device for the output device, and data transfer between the channel and the control device and pre-reading/collective writing for the input/output device in parallel. and starting the next scheduling step when the total amount of data transferred between the channel and the control device after starting the scheduling step reaches a predetermined scheduled amount. Pre-reading/collective writing scheduling method. 2. In claim 1, the scheduling step sets the pre-read/collective write data amount to the data transfer amount between the channel and the control device for each of the input/output devices after the start of the previous scheduling step. A method for scheduling read-ahead and bulk writes that determines equal amounts. 3. In claim 2, in the scheduling step, when the amount of data transferred to a certain input/output device is larger than a predetermined limit amount, the amount of data to be read in advance and written in bulk to each of the input/output devices is determined respectively. A prefetch/collective write scheduling method that corrects based on the data transfer amount. 4. In claim 3, the limit amount is a value obtained by subtracting the product of the known time required for preprocessing etc. and the data transfer rate of the input/output device from the scheduled amount.
Pre-reading/collective writing scheduling method. 5. In claim 2, the scheduling step includes pre-reading and collective writing for at least some of the input/output devices when there is a change in the active input/output devices after the start of the previous scheduling step. A pre-read/collective write scheduling method that corrects the amount of data based on the amount of data transferred. 6. The pre-read/collective write scheduling method according to claim 5, wherein the modification is performed only on input/output devices for which a read operation is requested.