CN102341779A

CN102341779A - Method, system and computer program product for managing the placement of storage data in a multi tier virtualized storage infrastructure

Info

Publication number: CN102341779A
Application number: CN2010800102363A
Authority: CN
Inventors: P.萨布罗尼尔
Original assignee: International Business Machines Corp
Current assignee: International Business Machines Corp
Priority date: 2009-03-02
Filing date: 2010-01-12
Publication date: 2012-02-01
Also published as: WO2010099992A1; EP2404231A1

Abstract

A storage management method for use in SAN based virtualized multi-tier storage infrastructure in a loosely defined and changing environment. Each physical storage media is assigned a tier level based on its Read I/O rate access density. The method comprises a top down method based on data collected from the virtualization engine compared to Read I/O capability and space capacity of each discrete virtual storage pool to determine whether re-tiering situations exist, a drill-in analysis algorithm based on relative Read I/O access density to identify which data workload should right-tiered among the composite workload hosted in the discrete virtual storage pool.

Description

Method, system and the computer program of the layout of managed storage data in the virtual storage organization of multilayer level

Technical field

The present invention relates to data processing field, and relate to storage administration and data placement optimization in the virtual storage organization of multilayer level particularly.

Background technology

Enterprise faces the increase complexity of the quick growth that needed by their storage, managed storage and the main challenge that the demand of the high availability of storage is caused.Storage area network (SAN) technology makes it possible to construct storage system through storage pool (pooling) individually from host computer, causes improved efficient.

Can also be used for storage management technique, Storage Virtualization that the user shields the complexity of physical storage.Piece virtual (being also referred to as piece sometimes assembles) provides the logical view of the physical storage (such as disk drive, solid-state disk and tape drive) of actual storage data to server.Logical view can comprise divides free memory many virtual storage regions that (or gathering) forms, and need not consider the physical layout of actual storage.Server is no longer checked the physical target, but checks the amount of logic (volume) that can aim at them and use.Server similarly is that virtual storage region is their direct attached belongings to their data of virtual storage region transmission.

Virtual can perhaps in the indicating panel driver, the generation on the grade of the piece of certain location in grade, the individual files of amount.Can in main frame (server) and/or memory device (intelligent disk array), assemble by execution block.

In data storage, the problem of (placement) of arranging of the precise information in a cover storage level is one of the most insoluble problem.Hierarchical storage be specify different classes of (category) data allocations to dissimilar storage mediums so that reduce total storage overhead.Classification can be based on grade, performance requirement, frequency of utilization, capacity and other consideration of required protection.

The customer requirements that is used for user's layout is often specified perhaps based on hope rather than accurate capacity plan loosely.In addition, even starting condition is sufficient, is applied in and also can experiences fierce data access change in the life cycle that runs through them.For example, the issue of the Internet application that following user's number is difficult to predict probably has in the actual data access behavior very different with the activity of initial configuration value and/or plan preset time.Time lapse, this application possibly gone up in the increased functionality that changes from cause the data access behavior and benefited.Subsequently, selected function possibly become and not be used, and causes following to change in data access patterns because the application that the function that they have is updated is taken over.Except application behavior uncertainty, the data access behavior in the single application is inhomogeneity fully.For example, highly active database journal and static parameter table will present the characteristic of very different data access patterns.All cross over these life cycles and change, the storage administrator faces loose appointment and environment that change, can not think that wherein the input of user's technology is accurate and believable, judges thereby adopt proper data to arrange.

A large amount of memory technology that combine, that in storage level (fiber channel (FC), series connection AT annex (SATA), solid state drive (SSD)), use causes in the storage level even complicated more application data is arranged judgement with their redundancy is set up (RAID 5, RAID 10 etc..), and wherein the cost of per unit memory capacity is can be in from 1 to 20 scope between SATA and the SSD.The correct level of using application data is important for enterprise reduces them when keeping application performance expense now.

In U.S. Pat 5,345, a kind of method that is used in the distribution of a plurality of memory device management data collection has been proposed in 584.It is well suitable to arrange for the independent data set of the independent memory device of under the situation that does not have local speed buffering layer, visiting based on the method for the data storage factor that is used for data set and memory device.Discarded this structure now mostly, this is because modern memory device is holding (host) data set across the breaking of multiple storage devices with the speed buffering layer that can cushion high number write-access instruction in (stripped) pattern.Use in addition total rate of people logging in (that is, read movable with write movable with) very inaccurate for the characteristic of the modern memory device of performance; For example; 300GB fiber channel driver typically can be supported a per second 100-150 random access; Can continue 15 minutes 1000 of per seconds of buffering and write instruction and write the speed buffering layer, each be 8K byte (typical database block big or small), causes total rate of people logging in to become inaccurate.This problem makes any will baffling based on the model that always reads and write movable and capacity.

Classification (hierarchical) data storing method in a kind of storage area network (SAN) has been proposed in the WO 2007/009910 from the assignee; Wherein SAN comprises a plurality of host data processors that couple with the Storage Virtualization engine, and itself and a plurality of physical storage medium couple.The designated level grade of each physical medium.The selectivity reorientation of the data block when this method surpasses level medium threshold value based on the visit behavior when data block.This method possibly cause for comprising by high demand application and hanging down uneconomic solution that demand is used the hybrid working amount of many application of forming.For such workload, this method will cause recommending or selecting two types storage resources.First kind of storage resources type will be that " high-performance Sasobit and SBR D etc. " type and second kind of storage resources will be " low performance SATA drive etc. " types, and possibly be enough and more economically for " on average " performance characteristic of the workload of supporting to assemble based on the solution of fiber channel (FC) dish.In essence, use 1,2 and 20 ratios of type price/unit of the capacity of SATA, FC and SSD will cause the FC solution than cheap five times of the SSD that merges and SATA solution.

The present invention is directed to the above-mentioned problem of quoting that solves.

Summary of the invention

The present invention proposes a kind of method that management data is arranged in the virtualized multilayer level storage organization of loose definition and environment that change that is used for.Each physical storage medium is assigned with the level grade that reads I/O speed visit density based on it.The present invention includes tightknit (top down) method; It is compared so that determine whether to exist level (re-tiering) situation of arranging again with the I/O ability that reads and the spatial content in each discrete virtual store pond based on the data from the virtualization engine collection; And the degree of depth (drill-in) analytical algorithm, it should correctly arrange level based on the relevant I/O of reading visit density so that be identified in the hybrid working amount of holding in the discrete virtual store pond which data.

Discrete storing virtual pond of this method operation and storing virtual dish grade, and be utilized in the complementary workload profile of the opportunity property that occurs in the multiple aggregation hybrid working amount.This method reduces the amount of arranging the level action again that will generate through the microanalysis in piece or storing virtual dish grade significantly, and more cost effective recommendation is provided.

Based on the behavior of tightknit tried analyzing storage resources, this method testing amount is arranged the suitable situation of level again, and provides and arrange level (up or down) again and recommend.

Again the arrangement level of being advised/correctly arrange the level action can be stored keeper's analysis is used for checking or passes to virtualization engine automatically being used for the virtual disk migration.

This method also comprise cover service quality problem write the response time assembly.This method is used the alarm based on the threshold value that is defined by the storage administrator.But this processing comprise virtualized storage architecture structuring and repeat assessment, cause workload data to arrange the treatment scheme of level action again.This processing comprises that also structured process is used for analysis and writes response time service quality alarm, judges whether to arrange again level and discerns which workload data to arrange level again to.

According to the present invention, method and systems a kind of as described in appended independent claims are provided.

The further embodiment of definition in appended dependent claims.

To with the mode of preferred embodiment and example aforesaid and other target, module and advantage of the present invention be described with reference to accompanying drawing now.

Description of drawings

Fig. 1 shows wherein the example of storage area network that can embodiment of the present invention;

Fig. 2 shows the virtualized simple view of piece;

Fig. 3 shows wherein the assembly of virtualization engine that can embodiment of the present invention;

Fig. 4 shows the assembly that level analyzer (START) is arranged in the storage that is used for correctly arranging level according to the present invention;

Fig. 5 illustrates at the correct preference data service model dimension that uses among the embodiment of level processing of arranging;

Fig. 6 illustrates the storage data service technology and the economic field of use;

Fig. 7 A, 7B, 7C and 7D show the example of actual conditions of the technical field blended data workload of the use that is used for storage pool;

What Fig. 8 illustrated three-dimensional model used in the present invention reads I/O speed density;

Fig. 9 shows by difference and reads reading I/O speed density and illustrating suitable hot analogy of data that two data workloads of I/O speed density form;

Figure 10 show remove how to revise the hybrid working amount when forming one of workload data read I/O speed density;

Figure 11 shows and supports the warning system based on threshold value of the present invention;

Figure 12 provides because it relates to the treatment scheme of the method for describing among the support the present invention who reads I/O speed density and SPACE APPLICATION; And

Figure 13 provides because it relates to the treatment scheme of embodiment of the support method of the analysis that writes I/O response time alarm.

Embodiment

The present invention proposes to use virtualization engine and analyzer assembly; This virtualization engine has the knowledge of the position of data and data, and this analyzer assembly is used to discern and should receives data to arrange level again and recommend real data to arrange the situation of level action again.

With reference to Fig. 1, show have several host application server 102 attached SAN 100.These can be many different types, the teller work station of the typically enterprise servers of some quantity, and some quantity.

What also be attached to SAN (via dish redundant array (RAID)) is the physical storage of various grades.In this example; The physical storage that has Three Estate: level 1, it can be for example such as the enterprise-level storer of

system storage DS8000; Level 2; It can be the intermediate range storer, such as the system storage DS5000 that is equipped with the FC dish; And level 3; It can be the low side storer, such as

the system storage DS4700 that is equipped with serial advanced technology attachment (SATA) driver.

Typically, each M dish (MDisk) and each RAID array 101 corresponding with single level belongs to single level.Each RAID controller 103 can be controlled and belong to various level RAID storer.Except using different levels in different physical disks types, can also different levels be applied to different RAID types; For example, can the RAID-5 array be positioned over the level higher than RAID-0 array.

The storer virtual engine 104 of the data routing of SAN through being arranged in all SAN data is by virtual, and presents virtual disk 106a to 106n to host server and workstation1 02.These virtual disks are made up of the capacity that the memory device across three levels provides.

Virtualization engine 104 comprises one of a plurality of nodes 110 (showing four), and these a plurality of nodes provide virtual, high-speed cache and copy services to main frame.Typically, node is disposed and the cluster (cluster) of configuration node in couples, and wherein every pair of node is known as I/O (I/O) group.

Because storer is attached to SAN, it is added to various storage pools, and each is by 103 controls of RAID controller.Each RAID controller presents SCSI (small computer system interface) dish to virtualization engine.The dish that is appeared can be managed through virtualization engine, and is called as by management dish or M dish.These M dishes are divided into expansion area (extent), the used capacity of fixed size block, and it begins to finishing label sequentially from each M dish.Can connect, peel off (strip) and break these expansion areas, perhaps can use any desired algorithm to produce the bigger virtual disk (VDisk) of presenting to main frame through node.

In by management dish group or MDG 108, can gather M dish M1, M2...M9, typically embody characteristic through factor such as performance, RAID grade, reliability, manufacturer etc.According to preferred embodiment, as shown in Figure 1, all the M dish carts among the MDG show the storer of same level grade.Many MDG that same level can be arranged in virtual storage architecture, each all is discrete virtual store pond.

Virtualization engine is transformed into the expansion area that V coils with the LBA (LBA) of virtual disk, and the expansion area of mapping V dish is to M dish expansion area.Fig. 2 shows the example of coiling the mapping of M dish from V.Each expansion area of V being coiled A is mapped to by the expansion area of one of management dish M1, M2 or M3.Can show that some are managed the dish expansion area and are not used from mapping table by the metadata creation of each node storage.These untapped expansion areas are available for creating new V dish, migration, expansion etc.

Typically, establishment and distribution virtual disk make enterprise-level server initially use the enterprise-level storer perhaps based on using owner's demand.This can not come to annotate fully through the actual data access characteristic.The present invention provides a kind of utilization structureization correctly to arrange level and handles the method for discerning better data placement situation.Different and the more cheap primary data that support of the present invention is used to use is arranged.For example, can in level 2 storage mediums, discharge the primary data that is used for all application and arrange, and the present invention will support this section data or whole arrangement levels again based on the actual conditions of the virtual storage organization of integral body.

For this reason, except being used to follow the tracks of by management dish expansion area to the metadata of the mapping of virtual disk, also monitor rate of people logging in to each expansion area.When any expansion area that provides being read and write data, use the access count update metadata.

To I/O stream be described with reference to Fig. 3 now.As shown in Figure 3, the virtualization engine of node 110 comprises with lower module: SCSI front end 302, storage virtualization 310, SCSI rear end 312, memory manager 314 and task manager 316.

The SCSI front end layer receives the I/O request from main frame; Carry out LUN mapping (that is, the LBA between virtual disk A and C is to logical unit number (LUN) (or expansion area)); And SCSI read the internal format that changes into node with write command.The SCSI back-end processing through top virtualization layer to its transmission to by the request of management dish, and order is addressed to the RAID controller.

The I/O storehouse can also comprise other module (not shown) such as remote copy, flash copy or high-speed cache.Usually the two appears high-speed cache in virtualization engine and RAID controller grade.

The node that shows among Fig. 3 belongs to the I/O group of distributing V dish A and B to it.This means that this node presents interface to V dish A and B and is used for main frame.Can also be corresponding by

management dish

1,2 and 3 with other virtual disk of distributing to other node.

Task manager 316 management of metadata 318, level level data and access value that it comprises the map information that is used for each expansion area and is used for the expansion area.This metadata also is available for virtualization layer 310 with memory manager 314.

Consider now the reception that writes request 350 from main frame, said write request 350 comprise the virtual disk that request is pointed ID and should be to its LBA that writes.Write when request receiving, front end will specify LBA to be converted into the expansion area ID (LUN) of virtual disk, suppose that this is the expansion area 3 (A-3) that V coils A.Virtualisation component 310 is used with the metadata shown in the form of mapping table among Fig. 2, comes the expansion area 6 (M2-6) of mapped extension district A-3 to M dish 2.Write request is passed to M dish 2 then via SCSI rear end 312 relevant controlling device, and data are written to expansion area M2-6.Virtualization layer sends message 304 to task manager, and writing to the expansion area 6 of M dish 2 asked in indication.Task manager then about expansion area M2-6 update metadata so that indicate this expansion area full now.Task manager also upgrades the access value of the metadata that is used for this expansion area.This can be through the count value in the replacement metadata, writes as the time of access value and realizes through being stored in it.Task manager is used for indication and has upgraded metadata so that reflect write operation to virtualisation component return messages 304.

To describe with reference to Fig. 4 now and allow correct storage arrangement level analyzer (START) manager component that is used for correctly arranging level of arranging the level action.START carries out the movable analysis of SAN so that identification should receive correctly to arrange the situation of level action and prepare suitable V dish migration action lists.At first, data acquisition unit 401 is through periodically being captured in the access activity of the topology data that comprises in the virtualization engine and each LUN and V dish, as the memory resource manager.This can comprise and writes and read activity count, response time and other monitor data.This can comprise rear end and the front end activity data and the internal measurement of virtualization engine, like the formation grade.Data acquisition unit periodically (preferred cycle is typically per 15 minutes) inserts these a series of data in its local thesaurus (repository), and with its storage longer time section (typically 6 months).

Data aggregators 402 is handled the SAN data that cover the longer time section (for example one day 96 sample for example through visit data collector thesaurus (using the mechanism that registers report such as one); Each is 15 minutes), and produce and to comprise minimum, maximum, average, shape factor ... cluster set be used for V dish and MDG by the virtualization engine management of SAN.

The data that can be produced by data aggregators are compared with the SAN model element data 403 that comprise for the I/O processing power of each MDG.This I/O processing power can be based on disk array manufacturer specification, disk array modeling activity characteristic (figure) (such as what produced by Disk Magic application software), or by the general industrial technology ability characteristics that receives of the dish of RAID controller control, and their number, their redundancy are set up and in high-speed cache fitting percentage (hit ratio) value of RAID controller grade.Also can use other I/O to handle the modeling ability algorithm.

Data by data aggregators produces can also be compared with the gross space capacity of each MDG, and MDG can be stored in the SAN model element data perhaps from the virtualization engine collection.

Data-analyzing machine assembly 404 is carried out these relatively and based on sent the alarm of correct arrangement level by the threshold value of memory manager setting.These alarms coverings do not have that balance is used and need consider the MDG of V dish migration action to it.

For any MDG in the alarm, data-analyzing machine provides the view that gos deep into of all V dishes that held by MDG of arranging with read access rate density.This view allows the identification immediately of " heat " V dish and " cold " V dish.According to the type of alarm, this gos deep into view and points to the V dish easily, and its migration to another level will solve this MDG alarm.Through to the correct level of arranging of these V dish, source MDG will see that the read access rate density value of the hybrid working amount of being held by this MDG becomes more near the MDG capability, make this MDG use about its use field balance better.

For all MDG, data-analyzing machine calculating clean (net) is read I/O visit density and is read the ratio of I/O processing power divided by MDG remaining space capacity as the MDG residue.Reading workload that I/O visit density will equal to read only I/O visit density will be considered as and be used for the complementary workload of this MDG in its current state.

According to the V dish migration action lists that the type of alarm is made up of " heat " or " cold " V dish, prepare and can be passed to virtual engine by the data-analyzing machine assembly and be used for perhaps automatically or after the memory manager checking implementing shown in 405 at SAN.

Can use following algorithm to confirm to arrange again the MGD target of the specific V dish of level to it.At first, its remaining space capacity or read the MDG that the I/O processing power is not enough to hold V dish vestige (footprint) (space that V dish vestige equals to be used for this V dish with read the I/O demand) and be eliminated as possible target.Then, select to read the immediate MDG that reads I/O visit density only of I/O visit density (for example, V dish workload profile is and MDG complementary workload in its current state) with the V dish.Be this operation of the dish of the V among the MDG in alarm repetition, until this alarm of accumulation associated weight solution of the V dish of arranging level again.Also repeat this operation for the MDG in all alarms.Can consider other algorithm so that solve assistance in the processing in alarm.

Fig. 5 illustrates the three-dimensional model that in specific embodiment of the present invention, uses.In embodiment based on

SAN virtualization controller (SVC), through " by management dish group " (MDG) unite a series of on storage array, hold by visit the rear end stores service is provided through the SVC layer by management dish (LUN) and at " separation mode ".As providing by the V dish through the being seen front-end stores service of data processing main frame.The hybrid working amount of many V dishes, all V dishes that for example in the MDG that provides, hold also can be described according to this three-dimensional model.

Fig. 6 illustrates two main fields such as the application of the stores service of RAID array, MDG, LUN or V dish.

First field is the functional area of stores service.It is positioned at the border of the gross space (with the M byte mode) of storage pool, is defined by the storage administrator with its maximum acceptance response time but its maximum reads I/O rate processing ability.

Second field is the economic field of the use of stores service.But this be positioned at previous field be positioned near maximum read I/O ability and storage space the border the acceptance response time restriction with the minimizing capacity.

Fig. 7 A-7D provides the examples shown of the workload situation in two use fields.

In Fig. 7 A, the whole memory capacity of data occupancy, I/O processing power are well used and are write the response time value and are out of question.There is matched well between data placement and the storage pool.

In Fig. 7 B, the I/O processing power is almost by whole uses, only extraordinary component ground memory allocated capacity and write I/O response time value and be out of question.Further capacity allocation probably causes the I/O restriction.Mobile data selected will be suitable to the storage pool of higher I/O ability.

In Fig. 7 C, the almost whole memory capacity of data occupancy, the I/O processing power is underused and is write I/O response time value and is out of question.The chance that has the storage pool that uses very possible more cost effective lower I/O processing power.

In Fig. 7 D, memory capacity is almost completely distributed, I/O processing power well balanced, however it is too high to write I/O response time value.Have and to evaluate the risk whether high response time value causes workload SLA (typically lapse of time) in batch before any action judging.

Fig. 8 has introduced the I/O speed visit density factor that reads, and it can be used for memory device (aspect capacity) or such as the part of using or the use workload data of (vitalizing many V dishes at a V holds) by assessment.Following equation provides additional detail.

● for MDG: maximum access density=I/O processing power/total memory capacity

● for application: the storage space of maximum access density=actual maximum I/O speed/distribution

● coil for V: the storage space of maximum access density=actual maximum I/O speed/distribution

Read I/O speed visit density and measure, and be that " heat " store operational amount and visit density applications are the heat of " cold " store operational amount when likening, can understand its algebraically easily when using wherein high visit density applications with the mode of IO/ second/megabyte.As illustrated in Fig. 9 and 10, the weighting caloric equation that is applied to warm water (heat+cold) is applied to " heat " and " cold " workload data.

If the gathering workload of all V dishes that in MDG, hold " near " MDG theoretical visit density and the almost all uses of this MDG capacity, then this MDG operates in its economic region.The present invention proposes to optimizing the processing that MDG uses by producing with other MDG exchange workload of different access density.The preferred embodiments of the present invention are to use and read I/O speed density so that in various levels, the MDG capacity is classified.The MDG that on level 1RAID controller, holds has the highest I/O speed density that reads in all MDG, and the MDG of the minimum I/O of reading speed visit density will belong to more low-grade level (according to the typically level 3-5 that divides into groups of the level in the virtual structure).

When giving the alarm based on the threshold value by the storage administrators may definition is the preferred embodiment through the data-analyzing machine embodiment of the present invention.Below listed three different alarms:

1. memory capacity is almost all distributed: in this case, distribute to the V dish by management dish pool-size near (with the % form) MDG memory span.

2.IO capacity almost all uses: the maximum at back end plate (by management dish group) reads I/O speed near (with the % form) maximum theoretical in this case.

3. " height " response time value: the quantity that writes instruction " important " (with the % form) that when comparing, in the SVC high-speed cache, keeps in this case with the sum that writes instruction.This phenomenon has disclosed the increase that writes the response time, and it possibly cause for the destruction of the SLA desired value of workload in batch.

Figure 11 shows this three alert thresholds, points to MDG like them and uses the field.

The driving principle that storage pool is optimized is following:

1. if " capacity of distribution " is lower significantly than " reading the I/O ability " near " max cap. " and " reading the I/O activity ", then " reading the I/O ability " does not have complete equipilibrium.So, must (that is, remove the application data of minimum rate of people logging in density so that Free up Memory holds the application data of higher rate of people logging in density in MDG) from discrete virtual store pond.Should the application data of the minimum rate of people logging in density that is removed be sent to the storage pool of lower read access rate density capability.This processing is called " arranging level downwards ".

2. if it is lower significantly than " max cap. " near " reading the I/O ability " and " capacity of distribution " " to read the I/O activity ", the performance limitations that the capacity unbalance of storage pool and the very possible generation of the more application datas of interpolation are not expected.Handle this situation and need from storage pool, remove the application data of the highest rate of people logging in density so that discharge the I/O ability that reads.This capacity will be used for holding the application data of lower rate of people logging in density afterwards.The application data that removes (the highest rate of people logging in density) can be sent to the storage pool of higher " reading the I/O density capability ".This processing is called " upwards arranging level ".

3. value increase and this can application service tier be reached an agreement on (SLA) place risk when the write cache buffer memory is filled, " to write the response time ".In this case, be necessary to carry out trend analysis so that look ahead " writing the response time " value and evaluation discharge to jeopardize and use SLA.This in this way situation, related application data (V dish) must " upwards be arranged level " to the storage pool of the higher I/O of writing ability.If SLA is not in the risk, application data is arranged in its current storage pool and must maintenance change.

4. be in wherein to have living space like storage pool and distribute fully and it reads the intermediateness that " reading the I/O ability " kept off in the I/O activity, need not consider any action.Even in MDG, present hot workload, its behavior can cause the average work load in the MDG ability through cold workload balance.This opportunity implementations reduces the imagination amount of correct arrangement level significantly, and it possibly attempted recommending inadequately by microanalysis.

5. if " read the I/O activity " near " reading the I/O ability " and " capacity of distribution " " max cap. " no better than, storage pool capacity well balanced is as long as " writing the response time " value remains on and can accept interior and two alarms of restriction and compensate one another.

6. when confirming that which V dish should correctly be arranged level, definitely read I/O speed V dish actual value and can not use " same as before ", this is the high-speed cache that occurs because of in the virtualization engine grade.I/O is read in this high-speed cache permission supply under the situation of not causing the rear end reading command to be asked to the front end data processor.The relevant I/O of the reading speed activity of each V dish that the front end gathering workload that method use of the present invention is held in MDG is compared is so that arrangement V coils and carries out reality and arrange the level judgement again between " heat " and " cold ".

It will be apparent to one skilled in the art that method of the present invention can be suitably implements in the logical unit of the parts of the step that comprises manner of execution, and such logical block can comprise the nextport hardware component NextPort of fastener components.

Can be through supporting as supporting this to optimize the realization of attempting referring now to the microprocessor of the described treatment scheme of Figure 12.

Whether the memory capacity that step 1200 check distributes is greater than by 90% of the total volume of management dish group, and wherein storage administrators may can be set up threshold value (90%) according to local policy.

Confirm the actual I/O of reading speed whether greater than 75% of the I/O ability that reads of MDG if the result, then carries out test (step 1202) for not, wherein storage administrators may can be set up threshold value (75%) according to local policy.

If-result, means the pond state that mediates for not, do not carry out further action and processing and go to step 1216.

If-test 1202 result for being; Mean that under the situation that does not consume whole spaces, assembling workload has used the high number percent that reads the I/O ability; Exist to add other workload and possibly make that to read the I/O ability saturated, cause workload SLA to bear difficult high probability.Therefore, in step 1206, recommend upwards to arrange the level operation.Next step, in step 1208, through selecting the V dish of the current the highest visit density of in MDG, holding, and to arrange upwards that level coils it to V be that another MDG of good complementary workload carries out and upwards arranges level.This V dish is correct arrange level operation after, source MDG will see that its read access rate density actual value reduces and becomes approaching with its capability, make this MDG use about its use field balance better.

Handle and go to step 1216 then.

Get back on step 1200 test of carrying out, if the result for being, then carries out and step 1202 is similarly tested.

If-result means that for being assembling workload is using the I/O ability that reads of high number percent and consuming most of space, MDG just at its economic neighborhood operation, does not carry out further action, and processing stops.

If-result means that for not reading the I/O ability underuses and consumed most of space, so, MDG reads the very possible maintenance of I/O ability and underuses.V among MDG dish will be contained on the MDG of low-level more more economically.Therefore recommend to arrange the level operation downwards in step 1212.Next step on step 1214, is carried out through the V dish of selecting the minimum visit density among the MDG and to arrange level downwards, and to arrange downwards that level coils it to V be another MDG of good complementary workload.This V dish is correct arrange level operation after, source MDG will see that read access rate density actual value increases and becomes more approaching with its capability, make this MDG use about its use field balance better.Handle and go to step 1216 then.

At last, on step 1216, available MDG storage space is assigned to other workload of complementary visit density profile, and handles and get back to the MDG that step 1200 is analyzed the back.When having analyzed all MDG, processing will be waited for until next assessment period, so that restart in 1200 for a MDG of tabulation.

Can but analysis/alarming method be integrated in the repeated storage management processing as the regular monitoring task.For example, every day, system's implementation of this method can produce the storage administration instrument panel, and it is for each MDG report, with the ability that has outstanding alarm when where applicable with capacity with write the relative actual value of response time situation.Instrument panel will be followed the view that gos deep into the behavior that the V dish that is held by each MDG is provided, this view with read I/O rate of people logging in density with maybe by storage administrator's assessment be used for arrange to the correct arrangement level motion tabulation of virtual engine transmission.

Figure 13 shows the process flow diagram that is used to look after the analysis/alarming method that writes I/O service quality aspect.In the figure, writing I/O rate indicator (indicator) replacement with another writes the I/O response time and triggers.This designator postpones I/O speed and always writes the ratio between the I/O rate value based on the front end write cache.Write cache postpone the I/O operation be in the write cache of virtual engine, keep write the I/O operation, this is because the rear end storage pool is because saturated and can not accept them.When the amount of write cache delay I/O operation arrives the remarkable number percent that always writes the I/O activity, the very possible slack-off and response time increase of front end applications.This designator is an another embodiment of the present invention as the use of permutatation level alarm.

In step 1300, carry out test and check front end write cache delay I/O speed whether to arrive threshold value, wherein this threshold value is set up according to local policy by the storage administrator.

If the result then handles and goes to step 1320 for not.

If the result is for being, so step 1302 follow the tracks of cause alarm the V dish to the application of using these V dishes.Next step, in step 1303, collection is used to use value time lapse [A] and the value of lapse of time SLA target [T] in batch in batch.Typically be responsible for through the application performance designator these values externally being provided to the present invention down IT operating personnel.Next step is in step 1304, and new test checks application SLA (typically lapse of time target) in batch whether to be in risk through A and T are compared with the secure threshold grade.

If the result means that for not A is lower than T significantly, so observed high response time value is inessential for duration in batch, on step 1306, does not carry out further action, and handles and go to step 1320.

If the result means A near T for being,, use for example TPC figure report to carry out and write I/O response time and the trend analysis that writes the I/O rate value as embodiment so in step 1308.

Continue to handle in step 1310, wherein carry out T.T. that new test checks applications wait to write the I/O operation and whether be added value (this total write time equal the V dish in all alarms all sampling periods that write I/O response time and the product that writes I/O speed with):

If-result is for denying; Mean that applications wait writes not increasing in time T.T. of I/O operation during batch processing; And therefore do not have deterioration duration SLA in batch, on step 1312, do not carry out further action so, and then handle in step 1320.

If-result is for being; Mean that applications wait writes increasing T.T. of I/O operation during batch processing; And possibly causing in batch, the duration becomes risky; Step 1314 is gone in processing, and wherein the usage trend analysis result is deduced (for example using linear modelling) following duration value in batch.

Whether processing continues to check SLA target (T) risky future in short-term in step 1316.If the result handles and goes to step 1312 for not, the result handles and goes to step 1318 so that some (or owning) V dishes are upwards arranged level for being else if, uses the SLA risk for the MDG with higher I/O ability creates.

At last, in step 1320, available MDG memory capacity is distributed to complementary other workload of visiting the density profile, and processing turns back to step 1300 so that analyze ensuing MDG.When having analyzed all MDG, processing will be waited for until next assessment period, to restart in 1300 for a MDG of tabulation.

Analysis/alarming method of describing among Figure 12 and 13 also can be used for embodying the characteristic of the unknown new workload of I/O profile.Can in " child care " MDG, hold this workload in the section (for example month) at a time and be used for its I/O behavior measure, so that gather enough behavioral datas.After this period, the Application V dish can be based on the space requirement that is provided by the data-analyzing machine assembly, read the I/O demand and read the I/O density value and come the correct level of arranging.This " child care " processing can replace will using which memory hierarchy and the optimum before required needs to complicated memory property estimation operation of which MDG in decision with low expense.To change through the future that the regular monitoring task is handled in the application behavior so, do not need expensive storage slip-stick artist's intervention with the aligning of guaranteeing application demand and storage organization.

In alternative embodiment, when the rear end disk array that is connected with virtual storage organization need remove to entrust (de-commisioning), analysis/alarming method of the present invention can be used for reorientating application data.In this case, the data-analyzing machine assembly can with data can be used to be judged as each logic storage unit and should use the discrete storage pool of which storage level and which (for example, MDG) optimum for each logic storage unit.

In another embodiment, when the disk array that is not connected with virtual storage organization need go to entrust, analysis/alarming method of the present invention can be used for reorientating application data.In this case, before the virtual logical storage unit being repositioned onto other discrete virtual store pond, dish possibly be connected with virtual storage architecture and experience the child care characterization.Alternately, this processing possibly comprise that use gathers on disk array exists performance data, and uses the data that provided by the data-analyzing machine assembly on virtual storage organization, to reinstall application.

Those skilled in the art will be appreciated that, though described the present invention about aforementioned exemplary embodiment, the invention is not restricted to this, and have many possible distortion and modifications that fall in the scope of the present invention.

Scope of the present invention comprises any novel feature of the combination of characteristic disclosed herein.The applicant provides such notice at this, during the application carries out, can form new claim in the such characteristic of any other application of deriving from the application or the combination of characteristic.Particularly, with reference to accompanying claims, can combine the characteristic of each independent claims rather than particular combination only in any suitable manner to enumerate in the claim.

For fear of doubt, the meaning that the term that uses with claim " comprises " and will not be interpreted as " only by ... composition " is described as running through at this.Those skilled in the art will be appreciated that, though the present invention describes about aforementioned exemplary embodiment through using SAN amount controller word, the invention is not restricted to this, and have the many possible wording of describing MDG or V dish.For example, MDG can be called as storage pool, virtual store pond or discrete virtual store pond, and the V dish is called as the virtual store logical block.

Claims

1. method that is used for the data storage of supervising the network; Said network comprises a plurality of host data processors that couple through storage virtualization plan engine and a plurality of physical storage medium; Said Storage Virtualization engine comprises the map unit that is used for mapping between virtual disk (V dish) arrives by management dish (M dish); Wherein a plurality of of same level grade are grouped so that form discrete virtual store pond (MDG) by the management dish, and said method comprises:

● the spatial content and the metadata that quantizes to read the I/O ability in each discrete virtual store pond described in storage;

● periodically from virtualization engine, gather about the storage of virtual disk and use, read I/O and write the movable information of I/O;

● assemble the information of gathering;

● the data of assembling are compared with the metadata in each discrete virtual store pond; And

● according to the result of comparison step, reach the tabulation of arranging the level action again that generates virtual disk based on threshold value.

2. the method for claim 1, that wherein gathers reads and writes I/O information and can be one of speed visit, response time, rear end and/or front end activity data and/or formation grade.

3. according to claim 1 or claim 2 method, wherein acquisition step is further comprising the steps of: in each period time with the information stores of gathering to local thesaurus.

4. like any described method in the claim 1 to 3, the data of wherein assembling comprise the minimum, maximum of V dish, the value of average, shape factor.

5. like any described method in the claim 1 to 4, wherein comparison step is further comprising the steps of: whether the memory capacity that check distributes is greater than predefined capacity threshold.

6. method as claimed in claim 5, wherein with predefined capacity threshold be set to discrete virtual store pond total volume 90%.

7. like claim 5 or 6 described methods, further comprising the steps of: as to check the actual I/O speed that reads whether greater than predefined capacity threshold.

8. method as claimed in claim 8 wherein is set to 75% of the I/O ability that reads with predefined capacity threshold.

9. like any described method in the claim 1 to 8, wherein comparison step is further comprising the steps of: whether the check write cache postpones I/O speed greater than the actual predefined percentage threshold that writes the I/O rate value.

10. like any described method in the claim 1 to 9, wherein set up threshold value by the storage administrator.

11. like any described method in the claim 1 to 10, the step that wherein generates the tabulation of arranging the level action again is further comprising the steps of: the storage pool instrument panel that generates the virtual store pond ability that comprises, capacity, the actual alarm of using and sending.

12. like any described method in the claim 1 to 11, the step that wherein generates the tabulation arrange the level action again is further comprising the steps of: generate the view that gos deep into that reads V dish that I/O speed density arranges with relevant.

13. system that is used for the data storage of supervising the network; Said network comprises a plurality of host data processors that couple through storage virtualization plan engine and a plurality of physical storage medium; Said Storage Virtualization engine comprises the map unit that is used for mapping between virtual disk (V dish) arrives by management dish (M dish); The a plurality of of same level grade are grouped so that form discrete virtual store pond (MDG) by the management dish, and said system comprises the parts of the step that is used to implement any one method of claim 1 to 12.

14. a computer program comprises being used for when on suitable computer equipment, carrying out said computer program, carrying out the instruction like the step of any one method of claim 1 to 12.

15. a computer-readable medium has coding computer program as claimed in claim 14 above that.