JP2003099384A - Load-sharing system, host computer for the load-sharing system, and load-sharing program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a multi-path load-sharing system in a host computer, which accesses a data storage device with unshared cache method. SOLUTION: In the load-sharing system, connecting a host computer 100 and a data storage device 200 with the unshared cache method via a network 300, a host computer 100 is constituted with a load-sharing control part 130 which groups plural physical paths with the data storage device 200 for each of caches which are to be used by a host control part, for executing prefetch within the data storage device 200 and manage them as a path group, map a sequential readout request into a single path group and properly distributing input output requests into the plural physical paths.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to multipath load balancing of a host computer when there are a plurality of physical access paths between the host computer and the data storage device, and more particularly, the data storage device is a non-shared cache system. In this case, the present invention relates to a load balancing system, a host computer of the load balancing system, and a load balancing program.

[0002]

2. Description of the Related Art In the case of a system that requires high availability,
Multiple physical paths are provided for accessing the host computer and the data storage device. The reason is that when a failure occurs in a certain path, a different path is used as its alternative path.

In order to efficiently use these plural physical paths, multipath load distribution is widely performed on the host computer side. The purpose is to improve the throughput between the host computer and the data storage device by appropriately distributing the load of access to the data storage device among a plurality of physical paths.

The means for controlling the multi-path load balancing is located between the file system layer and the device driver layer and receives an I / O request (input / output request) to the data storage device from the file system of the host computer. And the path to be used based on the load balancing algorithm, and I / O to the appropriate device driver.
It is a mechanism to pass the request.

Next, the operation of the conventional load balancing system will be described with reference to the drawings. FIG. 12 is a block diagram showing the configuration of a conventional load balancing system.

The file system 120 in the host computer 100e receives I / O from the application 110.
When the request is received, it is converted into a block unit I / O request and sent to the load distribution control unit 130e. The load distribution control unit 130e receives the I / O from the file system 120.
When the O request is received, the path to be used is determined based on the load balancing algorithm. Finally, the device driver 16
Send an I / O request to the appropriate one of 0-165.

As the load balancing algorithm, the generated I / O request is sent in a round-robin manner to a plurality of paths, or is sent to the path with the smallest number of pending I / O requests. , The method of sending to the path with the smallest total number of blocks of the pending I / O requests is used (the pending I / O request is an I for which an acknowledgment has not been returned). / O request).

Further, as an example of the prior art, Japanese Patent Laid-Open No. Hei 07
The technique disclosed in Japanese Patent Publication No. 334449 is cited.
This method is based on the input / output path identification unit that identifies physical paths, the load measurement unit that measures the load of each physical path, the load statistical information storage unit that stores load information, and the load information that is stored. An input / output route determination unit that determines a path is provided.

The input / output path determination unit compares the threshold value designated at the time of an input / output request to the peripheral device with the load status of the input / output control device indicated by the statistical information accumulated by the load statistical information accumulation unit. Based on the results
An information providing unit that determines whether or not the I / O route needs to be changed, and a plurality of I / O routes when the information providing unit determines that the load condition exceeds a threshold value and the I / O route needs to be changed. And a low load route selecting unit for selecting a route with a low load, and an input / output request is made via the input / output route selected by the low load route selecting unit.

[0010]

However, in the above-described conventional load balancing system, when multipath load balancing is used for a non-shared cache type data storage device, the prefetch performance of the data storage device deteriorates. There was a point. Hereinafter, this problem will be described in detail.

A high-performance data storage device uses a RAID (Redundant Array) to receive I / O requests from a host computer.
Before executing in a disk array by Inexpensive Disks), some functions are realized by processing using a processor or the like. Hereinafter, the part that performs this processing will be referred to as the host control unit.

As a function of the host control unit, a virtual volume function that enables the volume presented to the host computer side to be flexibly configured, and a RAM (Random Access Memory) for improving throughput and response time are provided.
ry) etc. and the caching function, etc.

The caching is a function of hiding the access time to the magnetic disk from the host computer to improve the throughput and the response time. In other words, if a read request causes a cache hit, it is not necessary to access the magnetic disk, and if a write request is cached for the time being and a confirmation response is returned to the host computer at that time, the host computer sends a magnetic disk to the magnetic disk. Access seems to be not happening.

In many cases, a plurality of host control units are provided in parallel. There are two reasons.

The first is high availability. That is, when a failure occurs in a certain host control unit, a different host control unit is used as its substitute.

The second is prevention of bottlenecks. That is, in order to prevent the processing power of the host control unit from becoming lower than the other capabilities of the system, a plurality of host control units that can perform the same processing are arranged in parallel to improve the processing power.

As described above, in the case of the data storage device having a plurality of host control units, it can be classified into a shared cache system and a non-shared cache system depending on the cache configuration system. FIG. 13 is a block diagram showing configurations of a conventional shared cache data storage device 200f and a non-shared cache data storage device 200g.

A shared cache system is a system in which all host control units refer to a single cache and a single cache control information, and a system in which several host control units refer to different caches or different cache control information. Is a non-shared cache method.

Here, the cache control information is information in which address information of cached data and order relation of caching are arranged and described, and which data is cached or which data is swept out is determined. It is used as information for.

In the shared cache method, since the data in the same address area is not cached twice, the cache capacity can be effectively used, and since only one cache needs to be prepared, the capacity is relatively large. There is an advantage that can be done. As a result, the cache hit rate becomes higher than that of the non-shared cache method.
Disadvantages are that there is a slight delay due to competition among a plurality of processors and that the cost tends to increase due to complicated control.

On the other hand, the non-shared cache method has the advantages that the competition between a plurality of processors is unlikely to occur, the structure is simple and the cost can be reduced, but the cache coherency is maintained.

As a countermeasure for this problem, a method of copying only dirty data between caches is generally used. Here, the dirty data refers to data that is cached for the purpose of being written to the magnetic disk later. When data is written to some cache and becomes dirty, it is copied to another cache. This method keeps the cache coherent.

Further, in the non-shared cache system, since cache control information is handled separately for each cache, a function such as prefetch works in units of host control units sharing the cache.

The prefetch function is a sequential read I from the I / O request from the host computer.
/ O (read request) is selected, and the function is to read data from the magnetic disk and store it in the cache in advance. Specifically, when read I / Os of adjacent addresses arrive at the data storage device continuously to some extent, it is determined from the cache control information that the I / Os are sequential read I / Os. An appropriate amount of data at the subsequent addresses is read and stored in the cache.

For example, when the host computer sends sequential read I / Os to the data storage device, the host controller receives the read I / Os of the first three or four adjacent addresses and thereafter. The read I / O corresponding to the address is sent to the magnetic disk. The read I / O that sequentially arrives from the host computer has a mechanism in which a cache hit occurs and the response time is significantly reduced.

The problem to be solved by the present invention is the problem that the prefetch performance of the data storage device is significantly deteriorated when the multipath load balancing is used for the non-shared cache type data storage device.

For example, when the round-robin is used for the load balancing algorithm of the multipath load balancing, the sequential read I / O generated in the host computer
Are sequentially distributed to each path, so that read I / O to intermittent addresses is recorded in each of the plurality of cache control information. In this case, there are two factors that cause performance degradation.

The first point is that the read I / O of consecutive addresses is hard to be recorded in the cache control information, so that the accuracy of selecting the sequential read I / O is lowered. This point reduces the cache hit rate of sequential read I / O.

The second point is that read I / Os of intermittent addresses are recorded in a plurality of cache control information.
Sequential read I / O by multiple host control units
Therefore, there is a possibility that prefetching to the same address area will be performed multiple times.
This point causes congestion of the magnetic disks belonging to the address area. That is, the response time of sequential read I / O is significantly deteriorated.

The above example is based on the load balancing algorithm.
The same applies when a method other than round robin, such as that disclosed in Japanese Patent Laid-Open No. 07-334449, is used.
This is because when multi-path load balancing is used for a non-shared cache type data storage device, read I / O to intermittent addresses is recorded in each of a plurality of cache control information.

An object of the present invention is to solve the above-mentioned drawbacks of the prior art and to reduce the performance of prefetch performed by the host controller of the data storage device by the host computer accessing the data storage device of the non-shared cache system. An object of the present invention is to provide a load balancing system, a host computer of the load balancing system, and a load balancing program that can perform multipath load balancing without the need.

In particular, the problem that the accuracy of the host controller selecting the sequential read I / O is reduced, and the problem that multiple prefetches from a plurality of host controllers to the same address area are performed in a multiple manner does not occur. It is to provide a load balancing system capable of performing path load balancing, a host computer of the load balancing system, and a load balancing program.

[0033]

To achieve the above object, a load balancing system of the present invention comprises a host computer,
In a load balancing system in which a non-shared cache type data storage device is connected via a network, address information of an input / output request is mapped to a plurality of physical paths between the data storage device and sequential reading is performed. A request is mapped to the single path, mapping information indicating the contents of the mapping is generated, and the input / output request is appropriately distributed to a plurality of physical paths by referring to the mapping information. And

According to a second aspect of the load balancing system of the present invention,
In a load balancing system in which a host computer and a non-shared cache type data storage device are connected via a network, a plurality of physical paths between the host computer and the data storage device are grouped for each cache in the data storage device. Managed as a path group, generating path group information indicating the information of the path group, mapping the address information of the input / output request and the path group, and mapping a sequential read request to a single path group. However, it is characterized in that mapping information indicating the contents of the mapping is generated, and the input / output requests are appropriately distributed to a plurality of physical paths by referring to the path group information and the mapping information.

According to a third aspect of the load balancing system of the present invention,
In a load balancing system in which a host computer and a non-shared cache type data storage device are connected via a network, communication between the data storage device and the host computer is relayed, and caching of the data storage device is executed. A host controller is provided, and the address information of the input / output request is mapped to a plurality of physical paths between the host computer and the host controller, and sequential read requests are mapped to the single path. , Generating mapping information indicating the contents of the mapping, referring to the mapping information, and appropriately distributing the input / output requests to a plurality of physical paths.

According to a fourth aspect of the load balancing system of the present invention,
In a load balancing system in which a host computer and a non-shared cache type data storage device are connected via a network, communication between the data storage device and the host computer is relayed, and caching of the data storage device is executed. A host controller is provided, and a plurality of physical paths between the host computer and the host controller are grouped for each cache in the host controller and managed as a path group, and information of the path group is indicated. Generates path group information, performs mapping between the address information of the input / output request and the path group, maps a sequential read request to the single path group, and generates mapping information indicating the contents of the mapping, Referring to the mapping information, Characterized in that it suitably distribute output requests to a plurality of physical paths.

According to a fifth aspect of the load balancing system of the present invention,
In the generation of the mapping information, the logical volume number of the input / output request is used as the address information of the input / output request.

According to a sixth aspect of the load balancing system of the present invention,
Partitioning the logical address number of the input / output request into appropriate widths, and using the partition to which the logical address number of the input / output request corresponds as the address information of the input / output request in the generation of the mapping information. Characterize.

According to a seventh aspect of the load balancing system of the present invention,
It is characterized by comprising means for obtaining information on an optimum partition width used for partitioning the logical address number of the input / output request via the network.

According to the load balancing system of the present invention of claim 8,
In a load balancing system in which a host computer and a non-shared cache type data storage device are connected via a network, the host computer sets a plurality of physical paths between the host computer and the data storage device in the data storage device. A path group table that records path group information that is grouped for each cache and managed as a path group, and mapping of sequential read requests to a single path group, address information of input / output requests, and the path group Is provided with a mapping table that records mapping information indicating
The I / O request is appropriately distributed to a plurality of physical paths by referring to the path group information and the mapping information.

According to a ninth aspect of the present invention, in a host computer of a load balancing system, which is connected to a non-shared cache type data storage device via a network, the input / output request address information and the data storage device are provided. Mapping with a plurality of physical paths between them, mapping a sequential read request to a single path, and mapping management means for generating mapping information indicating the contents of the mapping, and referring to the mapping information. In addition, a load distribution control means for appropriately distributing the input / output requests to a plurality of physical paths is provided.

According to a tenth aspect of the present invention, in a host computer of a load balancing system in which a data storage device of a non-shared cache system is connected via a network, a plurality of physical paths with the data storage device are provided. Is grouped for each cache in the data storage device and managed as a path group, and path group management means for generating path group information indicating information of the path group, address information of an input / output request, and the path group. Mapping management means for mapping sequential read requests to the single path group and generating mapping information indicating the contents of the mapping, and referring to the path group information and the mapping information, Fill-out request to multiple physical paths Characterized in that it comprises a load distribution control means for earnestly dispersed.

In the host computer according to the present invention of claim 11, the path group management means comprises means for acquiring, from the data storage device, configuration information of data stored in the data storage device.

According to a twelfth aspect of the present invention, in the host computer according to the present invention, the path group management means comprises means for acquiring configuration information of data stored in the data storage device from a designated external information device. And

In the host computer of the thirteenth aspect of the present invention, the mapping management means performs the mapping by using the logical volume number of the input / output request as the address information of the input / output request.

In the host computer of the fourteenth aspect of the present invention, the mapping management means partitions the logical address number of the input / output request into appropriate widths, and the input / output request is used as address information of the input / output request. The mapping is performed by using the partition to which the logical address number of.

In the host computer of the fifteenth aspect of the present invention, the mapping management means comprises means for obtaining, from the data storage device, information on an optimum partition width used for partitioning the logical address number of the input / output request. It is characterized by

In the host computer of the sixteenth aspect of the present invention, the mapping management means obtains information on an optimum partition width used for partitioning the logical address number of the input / output request from a designated external information device. It is characterized by comprising means.

According to a seventeenth aspect of the present invention, the host computer of the present invention uses the external information device as a name server.

In the host computer of the eighteenth aspect of the present invention, the mapping management means periodically updates the mapping information by referring to the load information of each path.

According to a nineteenth aspect of the present invention, in a host computer of a load balancing system in which a non-shared cache type data storage device is connected via a network, a plurality of physical paths are provided between the host computer and the data storage device. , A path group table for recording path group information for managing each cache in the data storage device as a path group, and an address of an input / output request in which a sequential read request is mapped to a single path group. A mapping table that records mapping information indicating mapping between information and the path group, and load distribution that appropriately distributes the I / O request to a plurality of physical paths by referring to the path group information and the mapping information. A control means is provided.

A load balancing program according to a twentieth aspect of the present invention is a load balancing program for executing load balancing by controlling a host computer connected to a non-shared cache type data storage device via a network. Mapping the address information of the request and a plurality of physical paths between the data storage device, mapping a sequential read request to the single path, and generating mapping information indicating the contents of the mapping. A mapping management function and a load distribution control function for appropriately distributing the input / output request to a plurality of physical paths by referring to the mapping information are provided.

According to a twenty-first aspect of the present invention, there is provided a load balancing program for executing load balancing by controlling a host computer connected to a non-shared cache type data storage device via a network. A plurality of physical paths to and from the storage device are grouped for each cache in the data storage device and managed as a path group, and a path group management function that generates path group information indicating information of the path group, A mapping management function that performs mapping between the address information of the input / output request and the path group, maps a sequential read request to a single path group, and generates mapping information indicating the contents of the mapping, and the path group. Refer to the information and the mapping information , Characterized in that it comprises a load distribution control function to properly distribute the output request into a plurality of physical paths.

According to a twenty-second aspect of the present invention, there is provided a load balancing program for executing load balancing by controlling a host computer connected to a non-shared cache type data storage device via a network. A path group table in which a computer records path group information for managing a plurality of physical paths to and from the data storage device as a path group by grouping for each cache in the data storage device, and a sequential read request A mapping table that records mapping information indicating the mapping between the address information of the input / output request and the path group, which is mapped to the single path group, and refers to the path group information and the mapping information. Multiple entry output requests Characterized in that it comprises a load distribution control function to appropriately disperse the physical path.

In the present invention, the path group management unit and the mapping management unit impose a restriction on sending a sequential read request to a single cache. By adding such a restriction, a sequential read request can be processed by the host control unit group sharing the cache. That is, by performing load distribution on that, the problem that the host control unit reduces the accuracy of selecting sequential read requests, and the problem that multiple prefetches from the multiple host control units to the same address area are performed It can be resolved.

Further, in the second embodiment of the present invention, the administrator previously sets the path group information, the address information of the input / output request, and the mapping information of the path group as a path group table and a mapping table as a host computer. The load balancing control unit refers to these tables and determines the path to be used based on the load balancing algorithm.

[0057]

BEST MODE FOR CARRYING OUT THE INVENTION Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

FIG. 1 is a block diagram showing the configuration of a load balancing system according to the first embodiment of the present invention. Figure 1
Referring to, the first embodiment of the present invention comprises a host computer 100 that operates under program control, a non-shared cache data storage device 200, and a network 300.

The host computer 100 has an application 110, a file system 120, a load distribution control unit 130, a load distribution information table 131, a path group management unit 140, and a path group table 141.
, Mapping management unit 150, and mapping table 1
51, device drivers 160 to 165, and communication ports 170 to 175.

The data storage device 200 includes a host controller 2
10, 211, caches 220 and 221, communication ports 230 to 235, and a disk array 240.

In this embodiment, the application 11
For simplicity, only one file system 120 is shown, but a plurality of applications and a plurality of file systems may be operating on a single host computer.

The device drivers 160 to 165, the communication ports 170 to 175, and 230 to 235 (both the host computer side and the data storage device side) are 6 for simplicity.
Although they are shown one by one, the case where each is provided with about 2 to 50 is also conceivable.

Although only one data storage device 200 is shown for the sake of simplicity, it may be connected to a plurality of data storage devices.

Only two host control units 210 and 211 are shown for the sake of simplicity, but there may be a case where the host control units 210 and 211 are provided with about 2 to 20. Although each host control unit 210, 211 has three communication ports, it may be possible to have about 1 to 10 communication ports.
There may be cases in which there are different numbers of communication ports, or there are mixed communication ports with different performance. Also, this figure 1
In this case, one host control unit has one cache, but a plurality of host control units may share one cache.

Various forms of the network 300 can be considered. As a concrete example, Fiber Channel
Arbitrated Loop, Fiber Channel Fabric, Ethernet, P
arallel SCSI (Small Computer System Interface),
ATM (Asynchronous Transfer Mode), FDDI (Fi
ber Distributed Data Interface) and other data links. It is also conceivable that interconnection devices such as repeaters, bridges, routers, gateways, etc. are provided along the way route.

Each of these parts operates roughly as follows.

When the file system 120 receives an I / O request (input / output request) from the application 110, the file system 120 converts the I / O request into a block unit I / O request and sends it to the load distribution control unit 130. LU for each I / O request
Address information such as N and LBA is added.

Upon receiving the I / O request from the file system 120, the load distribution control unit 130 first refers to the address information of the I / O request and the mapping table 151 to determine the path group to be used.

Then, the paths belonging to the path group are confirmed by referring to the path group table 141, and the path to be used is determined from among them based on the load balancing algorithm. Finally, the I / O request is sent to the appropriate one of the device drivers 160 to 165.

The path group management unit 140 first enumerates the paths to each data storage device, and confirms which cache of the data storage device is used by the host control unit to which each path is connected. Then, the paths are grouped for each cache used, and the information is described in the path group table 141.

The mapping management unit 150 first specifies the range of the address information of the I / O request handled by the file system 120, and maps the address information and the path group. Then, the information is stored in the mapping table 1
51.

The data storage device 200 is provided with a plurality of caches 220 and 221 and employs a non-shared cache system. Regarding dirty data, by copying between the distributed caches 220 and 221, the cache consistency between the caches is maintained. Since the other cached data is not copied, the content of the cached data differs between caches.

Upon receiving the I / O request from the communication ports 230 to 235, the host control units 210 and 211 refer to the address information and send the I / O request to the disk array 240. Also, the cache 220, 221
Read caching, write caching, and prefetching are performed using.

The disk array 240 includes a plurality of magnetic disk devices, and I / O from the host control units 210 and 211.
When the O request is received, the I / O request is sent to an appropriate magnetic disk by referring to the address information. Further, it may be possible to provide a plurality of RAID controllers inside and operate as a RAID device.

Next, the overall operation of this embodiment will be described in detail with reference to the flow charts of FIGS.

FIG. 2 shows the load distribution control unit 1 of this embodiment.
30 is a flowchart for explaining the operation of 30. The load distribution control unit 130 is first started up by a script or the like when the host computer boots. Then, it waits until an I / O request comes from the file system 120 (step 201).

When an I / O request comes from the file system 120, the path information to be used is selected by referring to the address information of the I / O request and the mapping table 151 (step 202). The mapping table 151 describes mapping information such that a path group can be uniquely selected from the LUN or LBA that is the address information of the I / O request.

Next, the path to be used is selected by referring to the path group table 141 and the load distribution information table 131 (step 203). Path group table 141
Describes which path belongs to each path group, and by referring to this, the paths belonging to the path group can be confirmed.

The load distribution information table 131 describes the usage status of each path and which load distribution algorithm is used in each path group, and the paths used in combination with the information in the path group table 141 are described. You can choose.

The path utilization status here means the path to which the I / O request was sent immediately before in each path group, the number of I / O requests pending in each path, and the pending status in each path. It refers to the total number of blocks of I / O requests that have been made.

The load balancing algorithm of the path group mentioned here is a method of sending by a round robin,
It refers to a method of sending to the path with the smallest number of pending I / O requests, a method of sending to the path with the smallest total number of blocks of pending I / O requests, and the like.

For example, a specific operation procedure when the round robin method is used for load distribution will be shown below. First, the load balancing information table 1 is used as the load balancing algorithm used in the path group selected in step 202.
31 (see the result of the round robin method in this example). Next, path group table 1
The path belonging to the selected path group is checked with reference to 41. Finally, the path to be used is determined by referring to the usage status of the path in the load balancing information table 131 (in this example, the next path is determined by referring to the information of the path that sent the I / O request immediately before). it can).

This procedure is almost the same even when different load balancing algorithms are used. It just refers to information about different usage situations.

When the path to be used is determined, the I / O request is sent to the appropriate device driver (step 20).
4). Finally, the usage status of the path changed by sending the new I / O request is described in the load balancing information table 131 (step 205). After this, it waits until an I / O request comes from the file system 120 again.

FIG. 3 is a flow chart for explaining the operation of the path group management unit 140 of this embodiment. The path group management unit 140 is first started by a script or the like when the host computer boots.
Then, the paths to the respective data storage devices are listed (step 301).

The resolution for enumerating paths is a combination of end-to-end communication ports, that is, a combination of each communication port on the host computer side and a communication port on the data storage device side accessible from the host computer side. For example, if each of the six communication ports on the host computer side can access the two communication ports on the data storage device side, a total of 12 paths are listed.

The meaning of being accessible includes not only the meaning that it is physically possible, but also that it is not restricted by access control such as zoning. That is, in this step 301, all accessible end-to-end paths are listed.

When the paths are listed, it is confirmed which cache of the data storage device is used by the host control unit to which each path is connected (step 302). In this step 302, first, it is specified which communication port on the data storage device side each path uses. Then, the cache used by the host control unit using the communication port is checked.

For example, in this embodiment, the communication port 2
The cache 220 is used when 30 to 232 are used, and the cache 221 is used when the communication ports 233 to 235 are used.

At step 302, the method is divided into two methods depending on how the configuration information of the data storage device, that is, the information on the correspondence between the communication port and the cache used is known. That is, there are a method of having a data storage device teach it and a method of having a third party such as a name server teach it.

The first method is to specify, for example, which communication port of the data storage device each path uses.
Since communication occurs between the host computer and the data storage device, the configuration information of the data storage device is passed from the data storage device side to the host computer at that time.

Second, when inquiring to the name server when specifying the communication port on the data storage device side to be accessed, the corresponding information is given to the name server.
The host computer knows the correspondence between the communication port and the cache used through the name server.

After confirming which cache each path uses, the paths are grouped for each cache used (step 303). The grouped paths are treated as a path group. Finally, information on the path group is stored in the path group table 141.
(Step 304).

FIG. 4 is a flow chart for explaining the operation of mapping management section 150 of this embodiment.
The mapping management unit 150 is first started up by a script or the like when the host computer boots. Then, first, the range of the address information of the I / O request handled by the file system 120 is specified (step 401).
Specifically, it is to specify what range the LUN or LBA is used in.

Next, the address information and the path group are mapped (step 402). The purpose of the mapping in step 402 is to introduce a sequential read I / O to a single path group, and to use a load balancing algorithm similar to the conventional one, with random I / O not having such a restriction as much as possible. It is to enable multipath load balancing. Specifically, the unique path group is derived from the LUN and LBA that are the address information of the I / O request. Two examples are given below.

The first is the logical volume number (hereinafter LU
N) is used. In this case, each LUN is appropriately mapped (for example, in ascending order) to the path group. With this method, the sequential I / O is almost L
It utilizes the property that UN is the same. That is, sequential I / Os are mapped to almost the same path group.

The second method is to use logical address (hereinafter referred to as LBA) partitions. In this case, the LBA is partitioned into certain appropriate widths, and each partition is mapped to the path group appropriately (for example, in ascending order). This method
Sequential I / O takes advantage of the close proximity of LBAs. That is, sequential I / Os are mapped to almost the same path group.

Although the method using one of them has been described here, a method of combining the LUN and LBA to derive the path group is also conceivable.

When the LBA partition is used in step 402, two methods can be considered for determining the partition width. That is, there are a method of having a data storage device teach it and a method of having a third party such as a name server teach it.

The first method is to specify, for example, which communication port of the data storage device each path uses.
Since communication occurs between the host computer and the data storage device, the optimum partition width is passed from the data storage device side to the host computer at that time.

The second is that the host computer knows the optimum partition width information through the name server.

After mapping the address information and the path group, the information is described in the mapping table 151 (step 403).

Next, the effect of this embodiment will be described. As described above, according to this embodiment, the path group management unit 140 and the mapping management unit 150 together provide information for restricting sending sequential read I / O to a single cache. is doing.
By applying such a restriction, sequential read I / O will be processed by the host control unit group (210, 211) sharing the cache. Then, since the load distribution control unit 130 performs the load distribution on that, the problem that the host control units 210 and 211 reduce the accuracy of selecting the sequential read I / O, There is no problem of prefetching multiple times in the same address area.

Next, a second embodiment of the present invention will be described in detail with reference to the drawings. FIG. 5 shows the second aspect of the present invention.
3 is a block diagram showing a configuration of a load balancing system according to the embodiment of FIG.

Referring to FIG. 5, the difference between the configuration of the load balancing system of the present embodiment and that of the first embodiment is as follows.
The host computer 100a does not include the path group management unit 140 and the mapping management unit 150 according to the first embodiment.

That is, the connected host controller 2
Path group information in which paths are grouped for each cache of the data storage device 200 used by
The administrator previously registers the address information of the I / O request and the mapping information of the path group in the path group table 141.
The mapping table 151a is provided to the host computer 100a.

The load distribution control unit 130a refers to these tables and determines the path to be used based on the load distribution algorithm.

As described above, according to the present embodiment, in addition to the effects of the first embodiment, it is possible to perform very detailed setting because information is set by a person, and a system is provided. The point is that it can be configured very simply.

Next, the third embodiment of the present invention will be described in detail. FIG. 6 is a block diagram showing the configuration of the load balancing system according to the third embodiment of the present invention.

The configuration of the load balancing system of the third embodiment of the present invention is the same as that of the first embodiment, but the mapping table 151 is periodically updated in the mapping management unit 150b of the host computer 100b. The feature is that it has a function to do.

FIG. 7 is a flow chart for explaining the operation of the mapping management unit 150b of this embodiment. Referring to FIG. 7, the mapping management unit 150b of this embodiment is first started up by a script or the like when the host computer boots. Then, first, the range of the address information of the I / O request handled by the file system 120 is specified (step 701).

Next, the process waits until the time set by the administrator arrives (step 702). When the set time is reached,
The address information and the path group are mapped (step 703). After mapping the address information and the path group, the information is stored in the mapping table 15.
1 (step 704). After this, it will wait until the time set by the administrator comes again.

The difference from the first embodiment is that the mapping management unit 150b periodically sets the mapping table 15
1 is the point of updating with reference to the load balancing information table 131. For example, the mapping management unit 1 of this embodiment
50b has a form in which the mapping table is updated at 12:00 am every day by referring to the usage status of the pass of the day.

By periodically updating the mapping table 151 by referring to the load information, it becomes possible to map the path group and I / O address information in accordance with the load status of the path. After all, more appropriate multipath load balancing becomes possible.

As described above, according to the present embodiment, in addition to the effect of the first embodiment, the load status of the path is updated by periodically referring to the load information and updating the mapping table 151. It becomes possible to map the address information of the path group and the I / O along. Mapping according to the load status of the paths realizes more appropriate multipath load distribution.

Next, the fourth embodiment of the present invention will be described in detail.

FIG. 8 is a block diagram showing the configuration of the load balancing system according to the fourth embodiment of the present invention. Figure 8
Referring to FIG. 3, the difference between the configuration of the load balancing system of this embodiment and that of the first embodiment is that the host computer 100c does not include the path group management unit 140 and the path group table 141. The feature of the present embodiment is that the concept of a path group is not handled by changing the function of the mapping management unit 150c.

Mapping management unit 150c of the present embodiment
Directly maps the I / O address information and the path. Then, the load distribution control unit 130c uses the mapping information to perform sequential read I / O.
, And the random I / O performs multi-path load balancing with no such limitation as much as possible.

FIG. 9 is a flow chart showing the operation of the load distribution control unit 130c according to the fourth embodiment of this invention. The load distribution control unit 130c is first started up by a script or the like when the host computer boots. Then, it waits until an I / O request comes from the file system 120 (step 901).

When an I / O request comes from the file system 120, the path information to be used is selected by referring to the address information of the I / O request, the mapping table 151 and the load balancing information table 131 (step 902). Mapping information is described in the mapping table 151 so that a path can be uniquely selected from the LUN or LBA that is the address information of the I / O request. The load distribution information table 131 describes the usage status of each path and which load distribution algorithm is used in each path group.

When the path to be used is determined, the I / O request is sent to the appropriate device driver (step 90).
3). Finally, the usage status of the path that has changed due to the transmission of the new I / O request is described in the load balancing information table 131 (step 904). After this, it waits until an I / O request comes from the file system 120 again.

FIG. 10 is a flowchart showing the operation of the mapping management unit 150c according to the fourth embodiment of this invention. The mapping management unit 150c is first started up by a script or the like when the host computer boots.

First, the range of the address information of the I / O request which the file system 120 is in charge of is specified (step 1001). Specifically, it is to specify what range the LUN or LBA is used in.

Next, the address information and the path are mapped (step 1002). The purpose of the mapping in step 1002 is to guide sequential read I / O to a single path, while avoiding random I / O with such restrictions as much as possible, using a load balancing algorithm similar to the conventional one. It is to be able to perform path load distribution. Specifically, the unique path is derived from the LUN and LBA that are the address information of the I / O request.

After mapping the address information and the path, the information is described in the mapping table 151 (step 1003).

As described above, according to the present embodiment, in addition to the effects of the first embodiment, the concept of a path group is not required, so that the system can be simply constructed.

Next, the fifth embodiment of the present invention will be described in detail.

FIG. 11 is a block diagram showing the configuration of the load balancing system according to the fifth embodiment of the present invention. Referring to FIG. 11, the fifth embodiment of the present invention includes a host computer 100, a non-shared cache type host control device 400, a data storage device 200d, and a network 300.

Referring to FIG. 11, the difference between the configuration of the load balancing system of this embodiment and that of the first embodiment is as follows.
Host controller 400 having host controllers 410 and 411 for performing virtual volume function and caching function
The data storage device 20 including the disk array 240
It is independent from 0d. In the present embodiment, in the host controller 400, the data storage device 200d
It provides virtual volume function and caching function.
Therefore, the data storage device 200d itself does not need to have a host controller.

Host computer 100 of the present embodiment
The process of is similar to that of the first embodiment. However, the physical path to be mapped in this embodiment is
It is a physical path between the host computer 100 and the host control device 400.

That is, a plurality of physical paths between the host computer 100 and the host control device 400 are
The caches in the host control device 400 are grouped and managed as a path group, and path group information indicating the information of the path group is generated. Then, the address information of the input / output request and the path group are mapped,
Sequential read requests are mapped to a single path group, and mapping information indicating the contents of mapping is generated. Then, referring to the generated mapping information, as described in the first embodiment, the input / output request is appropriately distributed to the plurality of physical paths.

In the example of FIG. 11, only one host control device 400 is shown for the sake of simplicity, but it is also conceivable that a plurality of host control devices operate in cooperation.

Only two host control units 410 and 411 are shown for the sake of simplicity, but there may be a case in which about 2 to 20 are provided. Further, each host control unit 410, 411 has three communication ports, but it may be provided with about 1 to 10 communication ports.
There may be cases in which there are different numbers of communication ports, or there are mixed communication ports with different performance. Also, this figure 1
In No. 1, one host control unit has one cache, but a plurality of host control units may share one cache.

The host controller 400 may also have a data storage device such as a disk array, hard disk, semiconductor memory or the like.

Further, for the data storage device 200d, only the disk array is shown for the sake of simplicity. In addition, the data storage device 2
00d shows only one communication port for simplicity.
It may be possible to provide about 2 to 50.

According to the present embodiment, in addition to the effects of the first embodiment, the host controller having the strike controller and the data storage device having the disk array can be independent devices, and Since it is easy to change the host controller, such as adding or deleting the host controller without changing the configuration of the storage device, the performance of the virtual volume function and caching function realized by the host controller can be adjusted more flexibly. It becomes possible to do.

Further, the load balancing systems of the above-mentioned respective embodiments can be freely combined and implemented. For example, in the fifth embodiment, the second, third, fourth
It is possible to implement by using the host computer of the embodiment.

For example, in the fifth embodiment, the fourth
When the host computer 100c according to the embodiment is used, the address information of the input / output request is mapped to a plurality of physical paths between the host computer 100c and the host control device 400, and sequential read requests are made. Mapping is performed for a single path and mapping information indicating the contents of mapping is generated. And
By referring to the generated mapping information, I / O requests are distributed appropriately among a plurality of physical paths.

The load balancing system of each of the above-described embodiments is implemented by the host computers 100, 100a, 100.
b, 100c load balancing control units 130, 130
a, 130c, the path group management unit 140, the mapping management units 150, 150b, 150c, and other functions as well as hardware, of course, the load balancing programs 90, 90a, which are computer programs having the respective functions, 90b, 90c, 90d can be realized by being loaded in the memory of the computer processing device. This load balancing program 90, 90a, 9
0b, 90c, and 90d are stored in a magnetic disk, a semiconductor memory, or another recording medium. Then, it is loaded from the recording medium to the computer processing device and controls the operation of the computer processing device to realize the above-described functions.

Although the present invention has been described with reference to the preferred embodiments and examples, the present invention is not necessarily limited to the above-described embodiments and examples, and various modifications can be made within the scope of the technical idea thereof. It can be modified and implemented.

[0141]

As described above, according to the load balancing system, the host computer of the load balancing system, and the load balancing program of the present invention, the following effects are achieved.

First, according to the present invention, the host control unit to which each path is connected confirms which cache of the data storage device is used, and groups the paths for each cache used. I / O
By the mapping management unit that maps the address information of the O request (input / output request) and the path group, it is possible to perform multipath load balancing after limiting the sending of sequential read I / O to a single cache. it can. That is, there is no problem that the accuracy of the host control unit selecting the sequential read I / O is reduced, or that multiple prefetches from a plurality of host control units to the same address area are performed in a multiple manner.

Secondly, according to the second embodiment of the present invention, the administrator previously sets the path group information, the address information of the I / O request, and the mapping information of the path group in the path group table and the mapping information. By giving it to the host computer as a table, the system can be configured simply.

Thirdly, according to the third embodiment of the present invention, the load information of the path is updated by periodically referring to the load information and updating the mapping between the address information of the I / O request and the path group. Address information of the path group and I / O can be mapped according to the situation. Mapping according to the load status of the paths realizes more appropriate multipath load distribution.

Fourthly, according to the fourth embodiment of the present invention, the system is simplified by directly mapping the address information of the I / O request and the path without using the concept of the path group. can do.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a load distribution system according to a first embodiment of the present invention.

FIG. 2 is a flowchart illustrating an operation of the load distribution control unit according to the first embodiment of this invention.

FIG. 3 is a flowchart illustrating an operation of a path group management unit according to the first embodiment of this invention.

FIG. 4 is a flowchart illustrating an operation of a mapping management unit according to the first embodiment of this invention.

FIG. 5 is a block diagram showing a configuration of a load balancing system according to a second embodiment of the present invention.

FIG. 6 is a block diagram showing a configuration of a load distribution system according to a third embodiment of the present invention.

FIG. 7 is a flowchart illustrating an operation of a mapping management unit according to the third embodiment of this invention.

FIG. 8 is a block diagram showing a configuration of a load distribution system according to a fourth embodiment of the present invention.

FIG. 9 is a flowchart illustrating an operation of a load distribution control unit according to the fourth embodiment of this invention.

FIG. 10 is a flowchart illustrating an operation of a mapping management unit according to the fourth embodiment of this invention.

FIG. 11 is a block diagram showing a configuration of a load distribution system according to a fifth embodiment of the present invention.

FIG. 12 is a block diagram showing a configuration of a conventional load balancing system.

FIG. 13 is a block diagram showing a configuration of a conventional shared cache type and non-shared cache type data storage device, respectively.

[Explanation of symbols]

90, 90a, 90b, 90c, 90d Load balancing program 100, 100a, 100b, 100c Host computer 110 Application 120 File system 130, 130a, 130c Load balancing control unit 131 Load balancing information table 140 Path group management unit 141, 141a Path group Tables 150, 150b, 150c Mapping management units 151, 151a Mapping tables 160-165 Device drivers 170-175 Communication ports 200, 200d Data storage devices 210, 211, 410, 411 Host control units 220, 221, 220, 221 Cache 230- 235, 430-435 Communication port 240 Disk array 300 Network 400 Host controller

─────────────────────────────────────────────────── ─── Continuation of front page (51) Int.Cl. ⁷ Identification code FI theme code (reference) G06F 12/08 551 G06F 12/08 551Z 557 557

Claims (22)

[Claims] 1. A load balancing system in which a host computer and a non-shared cache type data storage device are connected via a network, address information of an input / output request, and a plurality of physical paths between the data storage device. Mapping with
Mapping sequential read requests to a single path, generating mapping information indicating the contents of the mapping, referring to the mapping information, and appropriately distributing the input / output requests to a plurality of physical paths. A load balancing system characterized in that 2. In a load balancing system in which a host computer and a non-shared cache type data storage device are connected via a network, a plurality of physical paths between the data storage device and the data storage device are provided in the data storage device. Each cache is grouped and managed as a path group, path group information indicating the information of the path group is generated, the address information of the input / output request and the path group are mapped, and a sequential read request is made into a single Mapping to the path group, generating mapping information indicating the contents of the mapping, referring to the path group information and the mapping information, and appropriately distributing the input / output requests to a plurality of physical paths. A characteristic load balancing system. 3. A load balancing system in which a host computer and a non-shared cache data storage device are connected via a network, wherein the data storage device relays communication between the host computer and the data storage device. The host controller that performs the caching of the I / O request is performed, and the address information of the I / O request is mapped to a plurality of physical paths between the host computer and the host controller to issue a sequential read request as a single A load distribution system characterized by performing mapping on the path, generating mapping information indicating the contents of the mapping, and referring to the mapping information to appropriately distribute the input / output requests to a plurality of physical paths. . 4. A load balancing system in which a host computer and a non-shared cache data storage device are connected via a network, wherein the data storage device relays communication between the data storage device and the host computer. A plurality of physical paths between the host computer and the host control device are grouped for each cache in the host control device and managed as a path group. The path group information indicating the group information is generated, the address information of the input / output request is mapped to the path group, the sequential read request is mapped to a single path group, and the mapping indicating the contents of the mapping is performed. Generate information, the mapping information Referring to, load distribution system characterized in that to properly distribute the output request into a plurality of physical paths. 5. The logical volume number of the input / output request is used as the address information of the input / output request in the generation of the mapping information.
To the load balancing system according to claim 4. 6. The logical address number of the input / output request is partitioned into appropriate widths, and in the generation of the mapping information, the logical address number of the input / output request corresponds to the address information of the input / output request. The load balancing system according to claim 1, wherein a partition is used. 7. The load balancing system according to claim 6, further comprising means for obtaining information on an optimum partition width used for partitioning a logical address number of the input / output request via the network. 8. A load balancing system in which a host computer and a non-shared cache type data storage device are connected via a network, wherein the host computer has a plurality of physical paths to and from the data storage device. A path group table that records path group information that is grouped for each cache in the data storage device and managed as a path group; address information of an input / output request that maps sequential read requests to a single path group; and A mapping table for recording mapping information indicating mapping with a path group, wherein the input / output request is appropriately distributed to a plurality of physical paths by referring to the path group information and the mapping information. Load balancing system. 9. A host computer of a load balancing system connected to a non-shared cache type data storage device via a network, wherein address information of an input / output request, and a plurality of physical paths to and from the data storage device. Mapping
Mapping a sequential read request to a single path and mapping management means for generating mapping information indicating the contents of the mapping, and referring to the mapping information, the input / output request to a plurality of physical paths. A host computer comprising a load balancing control means for appropriately balancing. 10. A host computer of a load balancing system, which is connected to a non-shared cache type data storage device via a network, wherein a plurality of physical paths to and from the data storage device are cached in the data storage device. A path group management unit that manages each path group as a path group and generates path group information indicating the information of the path group, and performs sequential read requests by mapping the address information of the input / output request and the path group. Mapping to a single path group, and mapping management means for generating mapping information indicating the contents of the mapping; and referring to the path group information and the mapping information, the input / output request is made into a plurality of physical Equipped with load balancing control means to properly distribute to the paths A host computer characterized in that. 11. The host computer according to claim 10, wherein the path group management means includes means for acquiring configuration information of data stored in the data storage device from the data storage device. 12. The host according to claim 10, wherein the path group management means includes means for acquiring configuration information of data stored in the data storage device from a designated external information device. Computer. 13. The mapping management means performs the mapping by using the logical volume number of the input / output request as the address information of the input / output request. One of the host computers. 14. The mapping management means partitions the logical address number of the input / output request into appropriate widths, and the partition to which the logical address number of the input / output request corresponds as the address information of the input / output request. 13. The mapping is performed by using.
The host computer according to any one of 1. 15. The mapping management means comprises means for obtaining, from the data storage device, information on an optimum partition width used for partitioning a logical address number of the input / output request. Host computer. 16. The mapping management means comprises means for obtaining, from a designated external information device, information on an optimum partition width used for partitioning a logical address number of the input / output request. Item 14. The host computer according to Item 14. 17. The host computer according to claim 12, wherein the external information device is a name server. 18. The mapping management means updates the mapping information by periodically referencing the load information of each path. Host computer. 19. In a host computer of a load balancing system, which is connected to a non-shared cache type data storage device via a network, a plurality of physical paths to and from the data storage device are cached in the data storage device. A path group table that records path group information that is grouped for each and managed as a path group, and a sequential read request is mapped to a single path group. It is characterized by further comprising: a mapping table that records the mapping information shown; and a load distribution control unit that appropriately distributes the input / output requests to a plurality of physical paths by referring to the path group information and the mapping information. Host computer. 20. A load balancing program for performing load balancing by controlling a host computer connected to a non-shared cache type data storage device via a network, the address information of an input / output request, and the data storage device. Mapping with multiple physical paths between
A mapping management function for mapping a sequential read request to a single path and generating mapping information indicating the contents of the mapping, and referring to the mapping information, the input / output request to a plurality of physical paths. A load balancing program having a load balancing control function for appropriately balancing. 21. A load balancing program for performing load balancing by controlling a host computer connected to a non-shared cache data storage device via a network, comprising: A path group management function that groups paths for each cache in the data storage device and manages them as a path group, and generates path group information indicating the information of the path group; address information of input / output requests and the path group A mapping management function for performing mapping with, mapping a sequential read request to the single path group, and generating mapping information indicating the contents of the mapping, referring to the path group information and the mapping information, The I / O requests are sent to multiple physical paths. A load balancing program having a load balancing control function for appropriately balancing the above. 22. A load balancing program for performing load balancing by controlling a host computer connected to a non-shared cache type data storage device via a network, wherein the host computer communicates with the data storage device. A plurality of physical paths are grouped for each cache in the data storage device, and a path group table for recording path group information for managing as a path group, and a sequential read request is mapped to the single path group. A mapping table that records mapping information indicating a mapping between the address information of the input / output request and the path group, and refers to the path group information and the mapping information to output the input / output request to a plurality of physical paths. Load balancing system to properly distribute Load distribution program characterized by a function.