JP3977573B2 - Disk controller - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a disk control apparatus that can improve performance by adaptively avoiding concentration of a load on a specific disk apparatus.
[0002]
[Prior art]
Conventionally, a technique such as disk striping is known as a method of distributing data to a plurality of disks in order to speed up disk access.
[0003]
This is a method in which an array is composed of a plurality of disk devices, the entire array is virtually regarded as one disk device, and data is distributed to the entire array by addressing in the virtual disk device.
[0004]
[Problems to be solved by the invention]
However, with such a method, there is a problem that when the access is concentrated on a specific disk device, the load cannot be distributed.
[0005]
In addition, when a computer system is actually operated, which disk device the load is concentrated on is affected by the operating environment, and it is difficult to predict the load status of each disk device at the time of system design. It was almost impossible to distribute on all disks on average.
[0006]
The present invention has been made in view of such circumstances, and it is desirable to provide a disk control device that can improve performance by adaptively avoiding concentration of load on a specific disk device. Objective.
[0007]
[Means for Solving the Problems]
In order to achieve the above-described object, the disk control device of the present invention provides data on a disk device with a high access frequency and data on a disk device with a low access frequency in order to correct an uneven access frequency among a plurality of disk devices. Is a disk control device for adaptively and almost uniformly distributing the access frequency to each of a plurality of disk devices. Based on the frequency information managed by the access frequency management means for managing frequency information related to the access frequency of the device for each disk block, the bias of the access frequency between the plurality of disk devices is predetermined. Monitors whether the allowable range is exceeded, and determines the predetermined allowable range A disk that exchanges data on a disk block of a disk device with a high access frequency selected by a predetermined rule and data on a disk block of a disk device with a low access frequency when a deviation in access frequency exceeding the limit is detected A control means, a mapping management means for managing a mapping table for associating an address of a disk block whose data has been exchanged by the disk control means with an address of a disk block that actually stores the data, and for any of the disk blocks An access control means for referring to a mapping table managed by the mapping management means when access is requested, and performing access to the plurality of disk devices after performing address conversion as necessary; and the mapping management means Based on the mapping table to be more manageable, characterized by comprising a disk restoring means for restoring all disk blocks to the state before the exchange, a.
[0008]
In the disk device of the present invention, since the access frequency to each disk device is dynamically distributed, the performance is improved as the system is operated, and there is no awareness of the host system. It does not involve any modification of the system. Furthermore, for example, even if it becomes necessary to move a disk device that has already been exchanged from another disk device to another system, it is possible to easily restore the disk device to its original state. Become.
[0009]
Further, the disk control device of the present invention preferably exchanges data having a multiple of the track size of each disk device.
[0010]
In the disk control device of the present invention, even if data is exchanged, the continuity of data on the track is ensured, so that there is almost no adverse effect on performance due to data distribution.
[0011]
In the disk controller of the present invention, it is preferable that the access frequency management means periodically initializes frequency information.
[0012]
In the disk control apparatus of the present invention, since the latest fluctuations in the disk load situation can be dynamically followed, appropriate data distribution that is not limited by past results is realized.
[0013]
In the disk control device according to the present invention, it is preferable that the access frequency management unit discards the number of accesses that are past for a predetermined period included in the frequency information.
[0014]
In the disk control device of the present invention, as described above, since it is possible to dynamically follow the latest fluctuations in the disk load status, appropriate data distribution that is not limited by past results is realized, and , To prevent the occurrence of unnecessary data exchange by absorbing temporary fluctuations to some extent.
[0017]
In the disk control device of the present invention, the mapping management means saves the mapping table to any storage area of the plurality of disk devices when the system is stopped, and restores the saved mapping table when the system is started. It is preferable to have a means to do.
[0018]
In the disk control device of the present invention, since the state at the time of shutdown can be taken over at the time of restart after the shutdown, post-processing is performed to restore the state of all the disk devices to the original state at the time of shutdown. There is no need, and even if this disk controller itself is replaced with another disk controller and restarted after shutdown, the shutdown state can be taken over without any problem.
[0019]
In the disk control device of the present invention, it is preferable that the mapping management unit includes a unit that stores the difference or the whole in any storage area of the plurality of disk units when the mapping table is updated.
[0020]
In the disk control device of the present invention, even if, for example, an unexpected power interruption occurs, the latest information on the data exchange status is not lost, and access to the disk device can be continued even after the power is restored. It becomes.
[0025]
DETAILED DESCRIPTION OF THE INVENTION
Embodiments of the present invention will be described below with reference to the drawings.
FIG. 1 is a configuration diagram of a computer system to which a disk control apparatus according to an embodiment of the present invention is applied.
[0026]
As shown in FIG. 1, in this computer system, a PCI bus 1 is laid as a system bus. A CPU 11, a system memory 12, a display controller 13, a keyboard controller 14, and a host bus adapter (HBA) are connected to the PCI bus 1. 15 is connected.
[0027]
The CPU 11 controls the entire computer system, and operates based on a description of an application program including an operating system and utilities stored in the system memory 12.
[0028]
The system memory 12 is a memory device serving as a main memory of the computer system, and stores an application system including an operating system and utilities describing the operation of the CPU 11 and various processing data used for executing these.
[0029]
The display controller 13 controls the output of the user interface of the computer system, and controls display data in the video RAM 131 drawn by the CPU 11 on a display device such as a CRT 132 or LCD 133.
[0030]
The keyboard controller 14 controls input of the user interface of the computer system, and converts the operation of the input device such as the keyboard 141 and the mouse 142 into data and delivers it to the CPU 11.
[0031]
The host bus adapter (HBA) 15 is a so-called disk control device that controls a plurality of disk devices 16a to 16n serving as external storage of the computer system, and is connected to the plurality of disk devices 16a to 16n via the SCSI bus 2. Execute data exchange between. The plurality of disk devices 16a to 16n controlled by the host bus adapter (HBA) 15 are independent drives recognized by the host system, and are logically single to the host system. It does not constitute an array to make it look like a disk.
[0032]
Then, the host bus adapter (HBA) 15 of this embodiment corrects the access frequency deviation among the plurality of disk devices 16a to 16n, and the data of the disk device with high access frequency and the data of the disk device with low access frequency. Is characterized in that it is dynamically replaced without making the host system aware of this, and this point will be described in detail below.
[0033]
FIG. 2 is a diagram showing the configuration of the host bus adapter (HBA) 15 according to this embodiment.
[0034]
As shown in FIG. 2, the host bus adapter (HBA) 15 includes a mapping table 152 and an access frequency management mechanism 153 in addition to the SCSI control mechanism 151.
[0035]
The SCSI control mechanism 151 is the core of the host bus adapter (HBA) 15, and access to the disk devices 16a to 16n requested from the host system via the PCI bus 1 is determined by the SCSI bus 2. Run according to protocol. The SCSI control mechanism 151 executes access control specific to the present invention, which will be described later, based on the mapping table 152.
[0036]
The access frequency management mechanism 153 records access statistical information for all the disk blocks of the plurality of disk devices 16a to 16n, and periodically resets the statistical information. The following processing is executed when the access is concentrated, for example, when access is concentrated on a specific disk device or when access to a specific disk device is not generated. I am trying to correct it.
[0037]
First, the contents of a disk block with a high access frequency are exchanged with the contents of any disk block of a disk device with a low access frequency. Next, it is registered in the mapping table 152 that this disk block has been exchanged. FIG. 3 is a conceptual diagram showing how the disk blocks are exchanged. Note that the size of the disk block to be exchanged is considered to be a multiple of the track size of each disk device. This is to ensure the continuity of the data on the track even after the exchange and to suppress the adverse effect on the performance due to the data distribution.
[0038]
The mapping table 152 is secured in the memory area of the host bus adapter (HBA) 15. The access frequency management mechanism 153 stores the mapping table 152 in the disk devices 16a to 16n when the system is stopped. The mapping table 152 saved at the time of system startup is loaded.
[0039]
On the other hand, when a disk access is requested from the host system, the SCSI control mechanism 151 first maps whether or not the disk block specified in the request has been moved to another disk device by the access frequency management mechanism 153. The determination is made with reference to the table 152.
[0040]
If the designated disk block is registered in the mapping table 152, the SCSI control mechanism 151 identifies the destination of this disk block from the mapping table 152, and converts this request into access to the destination block. .
[0041]
Here, with reference to FIG. 4, the operation procedure of the host bus adapter (HBA) 15 during disk access will be described.
[0042]
When disk access is requested from the host system, the SCSI control mechanism 151 determines whether or not the target is registered in the mapping table 152 (step A1). Here, when registered in the mapping table 152 (YES in step A1), the SCSI control mechanism 151 executes address exchange based on the mapping table 152 (step A2).
[0043]
Thereafter, the SCSI control mechanism 151 executes access to the disk (step A3), and the access frequency management mechanism 153 updates the access frequency information along with this access (step A4).
[0044]
Next, a data replacement algorithm of the host bus adapter (HBA) 15 executed when the access frequency management mechanism 153 periodically resets statistical information will be described.
[0045]
Assume that the allowable range of the access frequency difference is 50% and the number of accesses is the frequency shown in FIG.
[0046]
As shown in FIG. 5A, since the total number of accesses is 40 and the number of disk drives is 2, the average number of accesses per disk is 20. On the other hand, since the allowable range of the access frequency difference is 50%, the disk block replacement is performed so that the number of accesses to each disk is 10 to 30 times.
[0047]
First, among the blocks of the disk 1, a set of disk blocks whose total number of accesses is 10 or more is obtained. Here, only the block B can satisfy this condition. Next, a disk block is selected from each block of the disk 2 so as to be within the allowable range of 50% by replacing this block B. Here, by replacing the block F with the block B, the access frequency of the disks 1 and 2 becomes 15 times and 25 times, and it can be seen that the access frequency difference can be kept within 50%.
[0048]
As a result, the process of replacing the disk block B and the disk block F is started. FIG. 5B shows the state of the disks 1 and 2 after replacement.
[0049]
Next, a case where the allowable range of the access frequency difference is 50% and the access is biased toward the disk 1 as shown in FIG. Think.
[0050]
As shown in FIG. 6A, since the total number of accesses is 60 and the number of disk drives is 2, the average number of accesses per disk is 30. On the other hand, since the allowable range of the access frequency difference is 50%, the disk block is replaced so that the number of accesses to each disk is 15 to 45 times.
[0051]
In this case, even if any one disk block of the disk 1 is replaced with any one disk block of the disk 2, the access times of the disks 1 and 2 are 50 times and 10 times, which are not within the allowable range. . Therefore, it is necessary to exchange a plurality of disk blocks.
[0052]
Consider a case where two blocks A and B are selected from the disk 1 and replaced with the blocks F and G of the disk 2. At this time, the number of disk accesses is 30 times and 20 times, respectively, and it can be seen that the disk access is within the allowable range. FIG. 6B is a diagram showing the state of the disks 1 and 2 after replacement.
[0053]
Next, consider a case where the allowable range of the access frequency difference is 50% and the number of accesses is the frequency shown in FIG.
[0054]
As shown in FIG. 7A, since the total number of accesses is 20 and the number of disk drives is 2, the average number of accesses per disk is 10. On the other hand, since the allowable range of the access frequency difference is 50%, the disk block is replaced so that the number of accesses to each disk is 5 to 15 times.
[0055]
Among each block of the disk 1, a set of disk blocks is obtained such that the total number of accesses is 5 or more. Here, only the block B can satisfy this condition. However, even if this block B is replaced with any of the blocks D, E, and F of the disk 2, the access frequency difference does not fall within the allowable range. Similarly, it can be seen that the access frequency difference does not fall within the permissible range depending on the block exchange with the disk 2 for any set of disk blocks of the disk 1 where the access count is 5 or more.
[0056]
In such a case, abandon the correction of the access frequency bias by replacing the disk block. Therefore, the arrangement of the disk blocks does not change even after the replacement algorithm is completed. FIG. 7B shows the state of the disks 1 and 2 after completion of the replacement algorithm.
[0057]
Based on the above, the execution procedure of the data replacement algorithm of the host bus adapter (HBA) 15 will be described with reference to FIG.
[0058]
First, the access frequency management mechanism 153 obtains the average number of accesses per disk (step B1). Next, it is checked whether or not all subsets of the frequently accessed disk blocks have been inspected (step B2). If the inspection has been completed (YES in step B2), this processing is terminated.
[0059]
On the other hand, if the inspection has not been completed (NO in step B2), one set A of disk blocks belonging to the disk having the highest access frequency is selected (step B3). Here, it is checked whether or not all combinations having the same number of elements as the set A in the subset of the disk having the lowest access frequency have been checked (step B4), and if checked (YES in step B4), step Repeat the process from B2.
[0060]
If inspection has not been completed (NO in step B4), one set B of disk blocks having the same number of elements as the set A belonging to the disk having the lowest access frequency is selected (step B5).
[0061]
When the selected set A and set B are exchanged, it is determined whether or not the access frequency difference falls within the allowable range (step B6). If not, the process from step B4 is repeated. If it is within the range (YES in step B6), this replacement is executed (step B7), and this process ends.
[0062]
As described above, the host bus adapter (HBA) 15 of this embodiment realizes dynamically distributing the access frequency to the plurality of disk devices 16a to 16n on the SCSI bus 2.
[0063]
A disk device whose disk block has been replaced with another disk device by the access frequency management mechanism 153 cannot be accessed as a normal disk device. Therefore, in order to remove these disk devices and move them to another system, it is necessary to restore the moved disk blocks.
[0064]
Therefore, in this host bus adapter (HBA) 15, the access frequency management mechanism 153 is provided with a function of rewriting the moved disk block by tracing back the disk block movement history recorded in the mapping table 152. Can be accessed as a disk device.
[0065]
In the above-described embodiment, the statistical information is periodically reset, and the statistical information is aggregated before the reset to perform data exchange as necessary. This is to realize appropriate data distribution that dynamically follows the fluctuations in the latest disk load situation without being caught. For example, in order to absorb temporary fluctuations immediately after resetting to some extent, instead of this periodic resetting, the number of past accesses over a predetermined period is always discarded, and statistical information is aggregated and data exchanged. It is also effective to execute it periodically.
[0066]
In the above-described embodiment, the access frequency management mechanism 153 saves the mapping table 152 to one of the plurality of disk devices 16a to 16n when the system is stopped, and loads the saved mapping table 152 when the system is started. However, this eliminates the need to perform post-processing that restores the state of all disk units to their original state at the time of shutdown by allowing the state at the time of shutdown to take over when restarting after shutdown. At the same time, even if this host bus adapter (HBA) 15 is replaced with another host bus adapter (HBA) 15 and restarted after shutdown, the state at the time of shutdown can be taken over without any problem. Further, if the access frequency management mechanism 153 saves the difference or the whole in one of the plurality of disk devices 16a to 16n when the mapping table 152 is updated, for example, an unexpected power failure occurs. Even if it occurs, the latest information regarding the data exchange status is not lost, and access to the disk device can be continued even after the power is restored, which is effective.
[0067]
Even if there is no mechanism for taking into account the bias in access frequency, if a data exchange function using this mapping table 152 and an address conversion function after exchange are provided, for example, when the number of disk devices is increased, It is possible to automatically execute data distribution for performance improvement without the user himself / herself rearranging.
[0068]
【The invention's effect】
As described above in detail, according to the disk control device of the present invention, in order to correct the uneven access frequency among the plurality of disk devices, the data of the disk device with high access frequency and the data of the disk device with low access frequency are Since the access frequency to each disk unit is dynamically distributed, the performance improves as the system is operated, and the host system Is not conscious and does not involve any modification of the existing system.
[0069]
Also, by exchanging data having a multiple of the track size of each disk device, continuity of data on the track is ensured even after data exchange, and there is almost no adverse effect on performance due to data distribution.
[0070]
In addition, by periodically initializing the frequency information, it is possible to dynamically follow the latest fluctuations in disk load status, realizing appropriate data distribution that is not bound by past performance, If the number of accesses in the past that is included in the frequency information is discarded for a predetermined period or longer, temporary fluctuations are absorbed to some extent to prevent unnecessary data exchange. .
[0071]
Also, by providing a function to restore all disk blocks to the state prior to replacement based on this mapping table, for example, a disk device for which data exchange has already been performed is separated from other disk devices and moved to another system. Even if necessary, the disk device can be easily restored to the original state.
[0072]
In addition, by saving the mapping table to one of the storage areas in the multiple disk units when the system is stopped and restoring the saved mapping table when the system is started, the shutdown state is restored when the system is restarted after shutdown. It is possible to take over, and there is no need to perform post-processing that restores the original state of all disk units to the original state at the time of shutdown, and this disk controller itself can be replaced with another disk controller. Even if restarting after shutdown, the state at the time of shutdown can be taken over without any problem, and when the mapping table is updated, the difference or the whole is stored in any storage area in multiple disk units. If saved, for example, an unexpected power failure may occur. Even the, without losing the latest information related to the replacement status of the data, it is possible after power is restored to continue in the access to the disk device.
[Brief description of the drawings]
FIG. 1 is a configuration diagram of a computer system to which a disk control device according to an embodiment of the present invention is applied.
FIG. 2 is a diagram showing a configuration of a host bus adapter (HBA) according to the embodiment.
FIG. 3 is a conceptual diagram showing how a disk block is replaced in the embodiment.
FIG. 4 is an exemplary flowchart for explaining an operation procedure during disk access of the host bus adapter (HBA) according to the embodiment;
FIG. 5 is a first diagram for explaining a data replacement algorithm of the host bus adapter (HBA) of the embodiment;
FIG. 6 is a second diagram for explaining a data replacement algorithm of the host bus adapter (HBA) of the embodiment;
FIG. 7 is a third diagram for explaining a data replacement algorithm of the host bus adapter (HBA) of the embodiment;
FIG. 8 is an exemplary flowchart for explaining an execution operation procedure of a data replacement algorithm of the host bus adapter (HBA) according to the embodiment;
[Explanation of symbols]
1 ... PCI bus 2 ... SCSI bus 11 ... CPU
12 ... System memory 13 ... Display controller 14 ... Keyboard controller 15 ... Host bus adapter (HBA)
16a to n ... Disk device 131 ... Video RAM
132 ... CRT
133 ... LCD
141 ... Keyboard 142 ... Mouse 151 ... SCSI control mechanism 152 ... Mapping table 153 ... Access frequency management mechanism

Claims (6)

A disk control device for adaptively and almost uniformly distributing the access frequency to each of a plurality of disk devices,
Access frequency management means for managing frequency information regarding the number of accesses of each disk device for each disk block;
Based on the frequency information managed by the access frequency management means, it is monitored whether or not the deviation of access frequency between the plurality of disk devices exceeds a predetermined allowable range, and exceeds the predetermined allowable range. Disk control for exchanging data on a disk block of a disk device with a high access frequency selected by a predetermined rule and data on a disk block of a disk device with a low access frequency when an uneven access frequency is detected Means,
Mapping management means for managing a mapping table that associates addresses of disk blocks whose data has been exchanged by the disk control means with addresses of disk blocks that actually store the data;
Access control for executing access to the plurality of disk devices after referring to a mapping table managed by the mapping management means and performing address conversion as necessary when access to any disk block is requested Means,
Based on the mapping table managed by the mapping management means, disk restoring means for restoring all disk blocks to the state before replacement;
A disk control device comprising:   2. The disk control device according to claim 1, wherein the disk control means exchanges data having a multiple of the track size of each disk device.   2. The disk controller according to claim 1, wherein the access frequency management means periodically initializes frequency information.   2. The disk control apparatus according to claim 1, wherein the access frequency management unit discards a past access count for a predetermined period included in the frequency information.   The mapping management means includes means for saving a mapping table to one of the storage areas in the plurality of disk devices when the system is stopped and restoring the saved mapping table when the system is started. Item 4. The disk control device according to Item 1. 6. The disk control apparatus according to claim 5 , wherein the mapping management means includes means for saving the difference or the whole in a storage area of the plurality of disk devices when the mapping table is updated.

Images