CN101529860A

CN101529860A - Improved computer-based system comprising several nodes in a network

Info

Publication number: CN101529860A
Application number: CN 200780039790
Authority: CN
Inventors: N·本卡利发; O·科泽特; C·吉坦尼特; S·理查德
Original assignee: Seanodes
Current assignee: Seanodes
Priority date: 2006-10-26
Filing date: 2007-10-25
Publication date: 2009-09-09
Also published as: FR2907993A1; FR2907993B1

Abstract

The invention relates to an improved computer-based system comprising several nodes in a network. The computer-based system comprises several computer-based facilities termed nodes interconnected in a network, - storing nodes comprise a direct-access local memory unit (38), and a storage server (46) for managing access to this unit on the basis of a queue, - application nodes, each comprising: * an application environment with a representation of files that are accessible in the form of addresses of blocks designating a physical address on a local memory unit (33), and * a storage client (44) capable of interacting with a storage server (46), on the basis of an access request designating a block address. A storage server (46) comprises a scheduler capable of executing the access requests contained in its queue (64) in a determined order. This order is determined as a function of a set of rules forming performance criteria, and involving one or more state parameters of the queue and/or of the storing node on which it resides.

Description

The improved computer based system that comprises several nodes in the network

Technical field

The present invention relates to the computer based system, comprise the several computer facilities or the station of the interconnecting nodes that is called in the network.

Background technology

Modern network comprises some subscriber stations, and they are connected to one or more server and can local ground or remotely share application and/or memory space.

In the field of the application program use mass data of sharing or in the field of shared mass data, commonly use special-purpose storage system such as storage area network (being SAN).

Use these improved systems to have certain disadvantages, such as the cost that is associated, performance and extensibility restriction, and corresponding with it general installation inconvenience.

Moreover, with regard to modern network, use these improved systems signifying fully not use the equipment that has existed in the network.

The present invention sets about improving this situation.

Summary of the invention

For this purpose, the present invention proposes the computer based system, comprise a plurality of computer based facilities, be called the interconnecting nodes in the network.Some node that is called memory node at least in the described node comprises at least one local direct access storage unit, and is arranged to according to access request formation and manages storage server to the visit of this local memory unit.

Some node that is called the application program node at least in the described node comprises:

* application environment has been equipped with:

-file system manager is arranged to that keeping can be by the expression of virtual address form from the file of this node visit, and

-corresponding relation module can keep the corresponding relation between at least one physical address on the local memory unit of each virtual address and memory node, and

* storage client can be mutual according to access request and any described storage server of specifying physical address.

In this computer based system, at least one described storage server comprises scheduler, can carry out the described access request that comprises in its formation with the order of appointment, and described scheduler is arranged to and this order is defined as forming performance criteria and comprises described formation and/or the function of the rule group of one or more state parameters of the resident described memory node on it of this formation.

The advantage that such computer based system has is for the intrinsic storage resources at the station (they will be also referred to as node) of using described network, so that store data effectively.In these stations, described storage server can be used to optimize the use of described network and described storage resources.Therefore this system makes and can maximally utilise the capability of described network and do not need specialized system.

The invention still further relates to the data management processes that can in the network of the computer based facility that comprises a plurality of interconnection that are called node, use, may further comprise the steps:

A. according to the expression of virtual address form, send file request from the application program node of described network,

B. according to the corresponding relation between at least one physical address on the local memory unit of the memory node of each virtual address and described network, determine pairing at least one physical address of described virtual address,

C. send the access request that indicates described physical address to the storage server of managing the visit of its local memory unit, and

D. described access request is placed in the formation of described storage server, and, carry out the access request that comprises in the described formation to be defined as forming performance criteria and to comprise described local memory unit and/or the order of the function of the rule group of one or more state parameters (the occupation rate level of processor load, central memory etc.) of the memory node be concerned about.

Description of drawings

With reference to the accompanying drawings, from as displaying and the study of the following explanation of unrestriction and some embodiment of providing, further advantage of the present invention and characteristics will become apparent, wherein:

Fig. 1 has shown according to the general utility functions figure that the present invention is based on system for computer;

Fig. 2 has shown the logic embodiment of system among Fig. 1;

Fig. 3 has shown the composition example of the unit among Fig. 2;

Fig. 4 has shown the method that the file in the system of Fig. 1 is conducted interviews;

Fig. 5 has shown the function embodiment of part process among Fig. 4;

Fig. 6 has shown in the scope of the enforcement of Fig. 5, the example of scheduling in storage server and execution circulation;

Fig. 7 to Figure 10 has shown the function example of Fig. 6;

Figure 11 has shown the scheduling of storage server in alternate embodiment and has carried out the example of circulation;

Figure 12 has shown the function example of Figure 11; And

Figure 13 to Figure 15 has shown the function example of Figure 12.

Following accompanying drawing and explanation have comprised the key element with some character in fact.They not only can understand the present invention as help, and can be in due course as helping to limit it.

Embodiment

The invention belongs to the type of having utilized the key element that meets copyright protection.The owner of a right does not oppose that anyone carries out unaltered duplicating to patent document or its specification (form that occurs such as it) in official document.For other aspects, the author keeps its all rights.

Fig. 1 has shown according to the general design drawing that the present invention is based on system for computer.In this system, application environment 2 has the access entrance to file system manager 4.Visual layer 6 corresponding relation of having set up between file system manager 4 and the storage server 8.

Fig. 2 has shown that the logic of system is implemented among Fig. 1.In this enforcement, one group of station 10 that this paper is also referred to as node interconnects in network, and they form wherein physics and application resource.

In the example that this paper introduces, described network is made of 5 stations that are labeled as Ni, and wherein i changes between 1 and 5.The application layer 12 that application environment 2 is upward distributed by N1, N2 and N3, the application layer 14 on the N4 and the application layer 16 on the N5 constitute.

Should be noted that term facility or station that text uses should be construed as denoting for the computer based unit in the network generally, application program or server program or the two be operation thereon all.File system manager 4 is made of distributed file system 18 and two non-distributed file system 20 and 22.System 18 is distributed on N1, N2 and the N3 and has defined can be from all files of distributed application program layer 12 visit.File system 20 and 22 has defined respectively can be from all files of

application layer

14 and 16 visits.

File by

file system

18,20 and 22 appointments physically is stored in the virtual memory space 24, and it is distributed on all Ni, and wherein i changes between 1 and 5.This paper is divided into shared logic space 26 and two privately owned

logical spaces

28 and 30 with virtual memory space 24.

Shared logic space 26 is corresponding to the space that can utilize distributed file system 18 from 12 visit of distributed application program layer, and privately owned

logical space

28 and 30 is corresponding to utilizing file system 20 and 22 spaces from

application layer

14 and 16 visits.

Logical space 26 is distributed on N1, N2 and the N3, and privately owned logical space 28 is distributed on N3 and the N4, and privately owned logical space 30 is on N5.

Therefore, the application program of layer 12 (perhaps being respectively 14 or 16) utilizes file system 18 (perhaps being respectively 20 or 22) " having seen " to be stored in data in the logical space 26 (perhaps being respectively 28 or 30), even these data may not physically be present on one of the memory disc at the station 10 of using this application program.

Moreover,

space

26,28 and 30 is pure logics, i.e. their not direct representations amount of physical memory.Logical space utilizes the virtual address of quoting in

file system

18,20 and 22 or comprising to shine upon.

In order to visit the data in these files, be necessary to use the corresponding relation module.The corresponding relation module is comprising the mapping table between the virtual address of data in the logical space and the physical address of amount of physical memory that the indication actual storage these data.

For the corresponding relation module, numerous embodiment are possible.The example of the scope that the distribution of the amount of physical memory of this paper introduction just is intended to show that the present invention is very general.

As from shown in example visible, each station both had been used for application layer and also had been used for accumulation layer.This multifunctionality makes the free space on the whole stations can use network, rather than allows this space free time.

But, within the scope of the invention, can make some station specialization and create the node that is exclusively used in the node of storage or is exclusively used in application program.

This means that within the scope of the invention any station can be used as application program node, memory node, perhaps takes on this two kinds of roles simultaneously.

All application program, storage and file system resource can be integrated on each station locally, also can be distributed on some stations of network.

Here it is for example to the situation of the example of station N1, N2 and N3, and its resource is distribution on application program aspect and file system and storage aspect with being integrated.

Fig. 3 has shown the examples of architectures at station 10 among Fig. 2.Can represent one of station N1, N2 or N3 at the station shown in this example.

The similar that the Nx that stands has individually is in global structure shown in Figure 1.Therefore it comprises that application layer 32, file system 34, virtual level 36 and form are for directly visiting the memory space 38 of local storage.

Virtual level 36 comprises motor (motor) 40 and mapping table 42.Direct visit to memory space 38 is managed by storage client 44 and storage server 46.Hereinafter will introduce the operation role and the method for these unit.

The example of this paper introduction has shown improvement embodiment of the present invention, wherein all resources, and---application program and storage---all is distributed on the network.

For example this means that file system 34 is not all on this station, but be distributed on several in them, and to its visit relate to network on comprise communicating by letter between other nodes of desired data.

For virtual level 36, storage client 44 and storage server 46 also is like this.The distribution of these unit utilizes administration module 48 to manage.

When administration module 48 mainly is used in establishment and upgrades logical space.During creating or revising logical space, administration module 48 calls virtual level 36, so that create each virtual address of logical space and the mapping table between the physical address on the given memory node.

Then, the mapping of formulating during according to the establishment logical space, on the level of the file system of utilizing this logical space, foundation can be by this file system access the virtual address of file and the data of composition this document between corresponding relation, " physics " data are stored in in the mapping table and the physical address that virtual address is associated.

This means,, just produced the corresponding relation between virtual address and the physical address in case administration module has been created logical space.Therefore, in the file system that provides the access entrance of logical space, virtual address looks like " empty ", and their physical address corresponding have utilized mapping table to be " saved ".

In setting up this space, during linking between the virtual address of the data of file and these data, just filled physical address just.

Can carry out in many ways by the work that virtual level is finished.In one embodiment, virtual level is distributed in data on different types of storage resources, so that find best trading off between the storage resources in making full use of network stream and the memory capacity that makes full use of these resources.In European patent 1454269B1 [0046] to [0062] paragraph, introduced the example of this virtual level.

Virtual level 36 also can add the mechanism of protecting written data.This mechanism can be for example based on the mode with RAID, each that has a physical address on the amount of physical memory that is positioned on the different stations is write request optionally duplicate.

Mapping table 42 may not be simple table.It can comprise especially with logical space or it and keep the relevant configuration data in some spaces of described corresponding relation for it.In this case, it can be particularly and the mechanism of virtual level 36 (it is used to upgrade the distribution of the corresponding relation between virtual address and the physical address) mutual so that guarantee to make full use of storage resources stream in the network and make full use of optimal compromise between the memory capacity of these resources.In European patent 1454269B1 [0063] to [0075] paragraph, introduced an embodiment of these mechanism.

At the remainder of this specification from start to finish, whether the resource of being considered is distributed unimportant.

In order to help to understand the present invention, application layer and accumulation layer are distinguished in suggestion.In fact, management has many advantages to the access modes of the data of being stored in the accumulation layer with respect to existing method.

Fig. 4 has shown the process of the access file of being carried out by system.

The application program of the application layer of given node is started by file access request 50 visit of file.File access request 50 comprises:

Address in the identifier of the file of being concerned about of file system and this file,

The size of request, i.e. the figure place that will visit according to the address of indication file, and

The type of request just reads or writes.

In step 52, file system is determined one or more virtual addresses of data in this file, and produces one or more virtual access requests according to request 50 and these virtual addresses.

Each virtual address requests all comprises:

Desired virtual address,

The size of request, i.e. the figure place that will visit according to the virtual address of indication, and

The type of request, it is consistent with the type of request 50.

With reference now to the system of being introduced among Fig. 2,, step 52 comprises determines logical space and virtual address or by the number of address on specified this space of request 50, and produces one or more " virtual " request.

Between file access request and virtual access request, exist grade difference.In fact, the file access request will be the content of a large amount of virtual addresses as destination object so that content that can the reconstruct file, virtual request then with the content of this address associated data piece as destination object.

The virtual access request that obtains is sent to virtual level then, and it determines physical address and corresponding memory space in step 54.

In order to determine physical address, the mode of operation of virtual level is for using motor 40 and mapping table 42.

In the scope of read access request, Already in the memory space 38, motor 40 calls mapping table 42 with virtual address to desired file, so that determine the physical address of data in the file by corresponding relation.

In the scope of write access request, this document may not be pre-existing in memory space 38.Yet as previous finding, the corresponding relation between virtual address and the physical address is fixed, so motor 40 with the read request scope in identical mode operate so that the physical address of specified data.

In all cases, in case motor 40 has been determined physical address, it just produces the physical access request in step 56, and sends it to storage client 44.

In step 56, produce the physical access request according to request 50 and the physical address in step 54, determined.

These requests comprise:

The physical address of-target;

The size of-request, promptly according to the figure place that will visit as the physical address of the destination object of asking, and

-attempt the type of the action carried out, just read or write.

From step 54, directly obtain the size of physical address and request, and request type is inherited from the type of described virtual access request.

Begin circulation then,, when the physical access request has been sent to storage client 44, just reached stop condition 58 wherein for the whole physical addresss that obtain in the step 52.

In fact, each physical access request all is placed in the request queue of storage client 44 so that carry out in step 60.Storage client 44 can comprise many formations alternatively, and for example it carries out mutual every storage server 46 with it all formation.

For brevity, in this circulation, all physical access requests of step 56 all are shown as successively and carry out.But, carry out and also can carry out concurrently, rather than only with serial mode.

In the example of being introduced, request sends layer by layer, has arrived the physical access layer up to them.But, can only determine and send address (virtual is physics then),, recover the selected attribute of original document request so that form the physical access request then in the aspect of physical layer.

In order to carry out given physical access request, storage client 44 is mutual with the storage server 46 of storage sites, and this station is comprising the memory space 38 by the specified physical address place of described physical access request.This Fig. 5 that will utilize alternately makes an explanation.

Just as seen in Figure 5, the execution of 44 pairs of physical access requests of storage client comprises at first by described storage server 46 reception physical access requests.The form that this reception of this paper is taked is for sending head 62, it tell storage server 46 requests type, this request size and as the physical address of destination object.

This request is stored in the formation 64 of storage server 46 by its head then.Formation 64 comprises still unenforced all access request and their executing state that is sent to described storage server 46 by all storage client 44.

Storage server 46 can comprise many formations 64, for example a formation is all arranged for every storage client 44 in the network, perhaps each storage space of its visit all there is a formation, perhaps is suitable for carrying out hereinafter any other layout of the scheduling strategy that will introduce for storage server 46 management.

Therefore storage server 46 can receive a large amount of requests successively from one or more storage client, and the full level of the disk of the station that makes the order of carrying out them can help most taking to carry out it, its management and the full level of network in general.

In known layout, the relation between storage client and the storage server is called as " oriented client ".In such relation, be in the great majority from the request queue of storage client 44, have only server to respond previous request after, just allow client computer to send new access request to it.

The framework of this paper introduction has constituted the management to the visit of storage space of " service-oriented device ".Different with known layout, therefore given storage client 44 can send a large amount of access request to same storage server 46, and does not need the latter at first the result to be beamed back in the requests of storage client 44 previous transmissions.This just feasible better balance that can in the I/O visit, realize disk and offered load, and particularly useful.

With to receive request in its formation 64 parallel, the step 66 that storage server 46 is carried out in the circulation, the request that it is dispatched therein and carries out in formation to be received.Therefore head 62 pairing requests are handled in this circulation with the determined order of storage server.

In step 66, server is to the request operation dispatching in the formation 64, manages the use of memory space of its visit and the use at the station that it moves thereon so that optimize it under the situation of considering several parameters (such as the full level of the central memory at processor load, described station etc.) locally.

Scheduling and the execution carried out in step 66 utilize Fig. 6 to make an explanation.

At given time, storage server 46 has one group of request in its formation, and they can be in various states.These states can be for example " with processed " (in request when just having been received by formation), " waiting pending " (when storage server 46 has had the needed total data of the request of execution and arranged execution afterwards) or " in the process of implementation ".

As if same these requests have at least two kinds of types of separating.First type is called " network " and indication necessary exchange that takes place between storage server 46 and given storage client 44.Another kind of type is called the visit that " disk " and indication storage server 46 must be carried out on one of memory space of its management, so that read or write data.

The state parameter of the memory space of being managed according to the network state parameter of the character of these requests and state thereof, system with by storage server 46 during the scheduling of execution in step 66, and the operating position at the station of moving thereon, and considered several parameters, such as the filling extent of the central memory of handling load, described station etc.

Step 66 will be introduced about the situation of a plurality of data storage disks of stores service management wherein.As a result, the many key elements that have global nature about circulation are some tables, might be multidimensional.

Therefore, the table that uses is called Last_Sect, it comprises single file, and last accessed on the pairing disk of these row sector is being shown in its every tabulation.

Similarly be, will use matrix T m_Used, certain client computer is represented to store in wherein every provisional capital, and every row are all represented certain disk, and the element value at the place, crosspoints of the capable and y of x row represents that request that client computer x sends occupies the time of disk y.

The circular treatment data 70 of step 66.Data 70 comprise the tabulation File of request and the tabulation List_Req of request.The tabulation File of request comprises one group of request with " with processed " state, just this time be engraved in some requests of being received in the formation of storage server.

The tabulation that is called List_Req is comprising one group of request with " waiting pending " or " in the process of implementation " state.Every of these requests all is attended by the age indicating device.The age indicating device of request has been pointed out to add its cycle-index of process of tabulation List_Req to from this request.

In step 72, storage server with tabulation File and List_Req as parameter call function Init ().Function Init () will introduce in more detail with reference to figure 7.

Function Init (), calls with File and the List_Req function Add_New_Req () as parameter that tabulates in step 702 in step 700 beginning.Function Add_New_Req () has the request newly that obtains all and adds it effect of tabulation List_Req to from tabulation File.In tabulation List_Req, the age indicating device minuend Add_New_Req () of new request is reset to 0.

The back of step 702 is the double condition tests about the situation that takies of storage server, so that the optimization system function.First condition is tested in step 704, wherein test wait indicating device Stat_Wt.

When indicating device Stat_Wt equals 0, this means from previous circulation to begin without any wait.On the contrary, the wait of previous cycle period equals 1 indicated by indicating device Stat_Wt.

Second condition tested in step 706, wherein exists plural request among the storage server proof listing File.

If one of these conditions do not satisfy, just formerly there has been wait cycle period, perhaps exists plural request in tabulation File, just carries out function Init () in step 708, and wherein indicating device Stat_Wt is set to 0, is used for next circulation.

Then, in step 710, whether storage server inspection tabulation List_Req is empty.If be not empty, function Init () stops in step 712, the scheduling circulation can be continued so that handles request among the List_Req that tabulates.

If tabulation List_Req is empty, so just there is no need to continue the scheduling circulation, and storage server is waited for one millisecond in step 714 by function Wait (1), then, indicating device Stat_Wt is set to 1, be used for next circulation, and turn back to step 702, so that any new request that recovery is received by the formation of storage server.

At function Init () afterwards, storage server is at step 76 call function Run_Old ().The purpose of this function is the request of carrying out among the List_Req with very high age indicating device.

Function R un_Old () utilizes Fig. 8 introduction, and if carried out old request and equal 1 with regard to return pointer Rst, otherwise Rst equals 0.

After beginning step 800, storage server is at step 802 call function Max_Age ().Function Max_Age () will tabulate List_Req as parameter, and return the highest age indicating device of the request among the List_Req.

If this age indicating device is greater than 150, just storage server is at step 804 call function Age () so, it will tabulate List_Req and digital 120 as parameter.Function Age () determines that age indicating device among the List_Req is greater than 120 request group.These requests are stored among the request list List_Old.

Then, storage server comes call function Req_Min_Sect () with tabulation List_Old and table Last_Sect as parameter in step 806.Function R eq_Min_Sect () makes and can judge the most approaching last sector of recently being visited, request visit sector which request has among the tabulation List_Old.

The mode of accomplishing this point is for for every the request that comprises among the List_Old, calculate between the last sector accessed on this disk that is comprised among target disk sector and the Last_Sect apart from absolute value.In case determined minimum value, corresponding request just is stored among the request Req.

Then, storage server is carried out it in step 808 by asking Req to call as the parameter of function Exec ().Function Exec () carries out request Req, measures the time of implementation of this request and this time is stored among the digital T_ex.

The execution of this request utilizes Fig. 9 introduction.This execution is based on tlv triple (triplet), comprises the physical address-size-type of request 900, and this request package is contained in the head in the formation 64.

In step 902, will determine the disk that to carry out and the link (chain) of network I/O about the test of request type.

If write request, storage server just requires storage client to write data to its transmission in step 904.Storage server is waited for this data, and after receiving them, is entered in the space specified by physical address in step 906.

Storage server 46 is write receipt 908 to storage client 44 transmissions then and is write to confirm this.After this, execution is finished in step 914.

If read request, the data that storage server 46 just comprises in the step 910 visit space specified by physical address until the size of this request, and send to storage client 44 in step 912 with these data.After this, execution is finished in step 914.

In case carried out request Req, storage server just upgrades List_Req in step 810.The mode of carrying out this renewal for will tabulate List_Req and digital T_ex as parameter call function U pd ().

This function is cancelled request Req from tabulation List_Req and List_Old, and upgrade matrix T m_Used, digital T_ex is added the element that the crosspoint of the row of storage client 44 correspondences that the send request Req row corresponding with the determined target disk of request Req is located.This makes the state that can make every storage client occupy each disk keep up-to-date.At last, the row of the disk that request Req has visited among the updating form Last_Sect are to consider the last sector of actual access.

Storage server 46 checks in step 812 whether tabulation List_Old is empty then.If be empty, carried out " old " request to show just be set to 1, and function R un_Old () finishes in step 816 at step 814 indicating device Rst.If not this situation, function R un_Old () turns back to step 806 to carry out remaining other old requests.

If the age indicating device that function Max_Age () returns is less than 150, be set to 0 at step 818 indicating device Rst so and do not carry out any " old " request as yet, and function R un_Old () finishes in step 820 to show.

Scheduling and execution circulation continue to test at step 78 couple indicating device Rst then, so that determine whether to have carried out " old " request.If like this, storage server just utilizes step 72 to repeat this circulation by call function Init () once more so.

Otherwise, storage server just by with tabulation List_Req as parameter call function R un_Min_Use (), stop this scheduling and carry out circulation.

Count Run_Min_Use () with reference to Figure 10 letter of introduction.After the initialization in step 1000, storage server 46 in step 1002 with numeral 1 as parameter call function Add_Age ().To the tabulate age indicating device of the whole requests among the List_Req of function Add_Age () all increases by 1, and counter Min_t is reset to 0.

In step 1004, storage server 46 then with tabulation List_Req sum counter Min_t as parameter call function U se_Inf_Min_t ().Function U se_Inf_Min_t () travels through tabulation List_Req, and whether the element that the crosspoint of the row of every storage client 44 correspondences of asking to send among validation matrix Tm_Used all it row corresponding with the disk of its appointment is located is less than Min_t.

Exactly, this means if sent the time that the client computer of request taken its target disk to be less than Min_t that given request is just selected.All requests of choosing thus all are stored among the tabulation List_Min_Req.

In step 1006, storage server is tested to check whether List_Min_Req is sky.If like this, counter Min_t increases by 1 in step 1008 so, and repeating step 1004.

In case tabulation List_Min_Req comprises at least one request, storage server 46 is with regard to execution in

step

1010,1012 and 1014, they only are used herein with the previous step of introducing 906,908 and 910 difference is tabulation List_Min_Req, rather than tabulation List_Old.

After best request was carried out according to

step

1010,1012 and 1014, storage server 46 was at step 1016 call function Rst_Tm_Used ().

The purpose of function R st_Tm_Used () is to compare with other storage client a storage client 44 matrix T m_Used to be resetted when using disk quite a lot of.

For this situation, function R st_Tm_Used () is with whole element summations of matrix T m_Used.This has represented the summation of the holding time of the disk that 44 pairs of storage servers 46 of whole storage client are managed.

If this summation has surpassed predetermined value, whole elements of matrix T m_Used all are set to 0 so.Otherwise matrix T m_Used is constant.

After step 1016, function R un_Min_Use () stops in step 1018, and scheduling and execution circulate in step 72 and begin once more.

Therefore function R un_Min_Use () can be according to the information that comprises in the request head and the execution of dispatch request, and do not consider the existence by the specified any data of these requests.

So, can dispatch the execution, particularly write request of a large amount of requests by this way, and can not make the storage space overload because of the data that will write in these requests.

But, in other application programs, only those among the dispatch list File are carried out all available request of the needed total data of request.The mode of accomplishing this point can be by data supply circulation is provided concurrently, thereby guarantees to be filled to be the quantity optimization that makes the request that will dispatch to the space that storage distributed of request msg.

About the situation of a plurality of data storage disks of stores service management by the agency of step 66.

As a result, compare with circulation, character is multidimensional for many elements of the overall situation are tables sometimes.If storage server is only managed single disk, situation just can be simplified.So, element Last_Sect becomes simple value, and element T m_Used becomes one dimension table (storage client).

Moreover, this mode of operation dispatching for the whole requests in some formations of storage server are all put together.But, can serial or parallel ground index by the List Req that will tabulate or by to each formation operation dispatching, respectively as distinguishing these requests from the function of the formation of wherein taking out request.

Figure 11 has shown scheduling and has carried out the alternate embodiment of circulation.

In this specific embodiment, scheduling is consistent with circulation shown in Figure 6 in fact with the execution circulation, and only it also has sleep cycle 110, carries out this sleep cycle 110 before step shown in Figure 6.

Sleep cycle 110 comprises sleep management function Sleep_Mng () 112, and its result is stored among the variable Slp.

Variable Slp has shown whether storage server is placed dormant decision-making temporarily.This function will further be introduced together with Figure 13 to Figure 15.

After function S leep_Mng () 112, sleep cycle 110 comprises the test 114 about variable Slp value.

If this variable is not 0, storage server minuend Force_Slp () 116 places sleep provisionally so.Sleep cycle 110 resets at 112 places then.

Function F orce_Slp () 116 " places sleep with storage server ", and mode is for sending the request that is called " placing sleep " in formation.Place the request of sleep to have precedence over all other requests.When it is performed, can cause storage server in the time of parameterisable, to dally.This function can be considered with Fig. 7 in function Wait () be equal to.

If variable Slp is 0, the execution of scheduling and execution circulation and shown in Figure 6 identical.

Introduce function Slp_Mng () referring now to Figure 12.As from this width of cloth figure as seen, function S lp_Mng () comprises that the order of function U pd_Slp_Par () 1122, function Perf_Slp () 1124 and function Mnt_Slp () 1126 carries out, and stops in step 1128 again.

The role of function U pd_Slp_Par () 1122 upgrades the parameter that is used to judge whether server is placed sleep.To introduce this function by means of Figure 13 now.

As from Figure 13 as seen, function U pd_Slp_Par () 1122 has upgraded two parameters of Tm_psd, Nb_Rq_Slp and variable Slp.

In step 1302, utilize function Elps_Time () to upgrade parameter Tm_psd.Function Elps_Time () has calculated since last execution function U pd_Slp_Par () 1122 and how long has experienced.

Accomplish this point mode can for, for example from one be recycled to all upgrade when next circulation keeps each the execution time/the date variable, cycle period compares the current time of this variable when carrying out in the back then.

In step 1304, parameter N b_Rq_Slp is increased numerical value Tm_psd*Fq_Rq_Slp.Parameter N b_Rq_Slp indicates to place the request number of sleep.

In the implementation column of Jie Shaoing, exist the sleep condition of two kinds of main types here.First kind is the condition type of connection performance, and second kind is the type that relevant name takies degree.This degree can utilize administration module to define particularly, perhaps puts it briefly to be considered as the parameter of being fixed by the system manager.

Parameter N b_Rq_Slp belongs to this second type.It is a counter, guarantees server with frequency Fq_Rq_Slp establishment sleep request, and this frequency is the parameter that the keeper regulated by storage server.

But, as what further will become apparent, sleep request only actually just can be carried out under certain conditions.This counter will make to determine to have carried out how many sleep requests.

Then, in step 1306, variable Slp is reset to 0, is used for current sleep cycle, and function U pd_Par_Slp () stops in step 1308.

Introduce function Perf_Slp () referring now to Figure 14.This function makes and can place sleep according to the state parameter decision of memory node and formation.

For this reason, this function is based on 1402 and 1,404 two tests.First test 1402 relates to the degree that takies of local resource (being the resource of the memory node carried out of storage server) thereon.

Here in the implementation column of Jie Shaoing, tested is processor resource.In order to accomplish this point, called the function that the evaluation process device takies degree.These functions can belong to type, and can be based on for example to the reference by the global variable that operating system kept, and they also can be more special-purpose functions.

If processor has been in limit load (for example, being higher than 90%), it will determine to place sleep so that do not damage the performance of memory node storage server so.

Therefore, if this is the case, variable Slp step 1406 be set to 1 and function Perf_Slp () stop in step 1408.

Should be noted that together with or replace processor load and can use many other conditions, other conditions that can imagine such as the access load or the technical staff of local memory unit at this.

Only first test 1402 when negative, that is to say when the load of processor is not too heavy, just carry out second and test 1404.This second test relates to the request number that comprises in the assessment formation.

In fact, if this number is too low, storage server does not just utilize the scheduling circulation well, has therefore reduced performance potentially.

Therefore, if the request number that exists in the formation is too low, variable Slp step 1406 be set to 1 and function Perf_Slp () stop in step 1408.Otherwise function directly stops in step 1408.

Here should be noted that also and can use many other conditions that principle is just storage server not to be placed sleep as long as satisfy favourable performance condition as yet.

In the middle of these alternative conditions, people can note the request type that exists in for example formation.Therefore, if request is " away from ", that is to say if surpass certain fixed threshold, just can think and wait for that in formation " more close " request arrives more favourable by the physical address of each request appointment and the distance between the previous physical address of visiting of storage server.Operable another kind of type criterion is the character of request, promptly reads or writes, and perhaps relates to other criterions of asking feature.

To introduce function Mnt_Slp () by means of Figure 15 now.The feasible plan period that can cancel sleep of this function, perhaps opposite, force " maintenance type " sleep.Function Mnt_Slp () is based on 1502 and 1,508 two tests.

First test 1502 compares with parameter N b_Rq_Slp and in order to allow one of request of sleeping to be performed required sleep request minimum number.This just is summed up as judgement and whether has carried out many sleep requests recently.

If parameter N b_Rq_Slp is less than this minimum number, variable Slp step 1504 be set to 0 and this function stop in step 1506.

Only when being sure, first test 1502 just carry out second test 1508.This second test compares parameter N b _ Rq_Slp with the maximum quantity of sleep request.Whether this just is summed up as judgement and has pass by for a long time since having carried out any sleep request.

If parameter N b_Rq_Slp is less than described maximum quantity, function stops in step 1506.Under opposite situation, variable Slp is set to 1 in step 1510, and this function stops in step 1506 then.

This means:

Even-plan sth. ahead and will carry out the sleep request, but owing to carried out many other requests, therefore will not carry out it not long ago, perhaps opposite,

-when carrying out the sleep request, also can make random decision carrying out a sleep request when having experienced a period of time, even according to other criterions this and unnecessary.

Because function Mnt_Slp () is the fact of carrying out afterwards at function Perf_Slp (), can overthrow the latter's decision-making so be to be understood that it.In other words, owing to safeguard storage server, this function makes can cancel the sleep period of having planned by variable Slp is worked, and perhaps executes a unplanned sleep.

Should be noted that also function F orce_Slp () reduces by a unit with counter Nb_Rq_Slp when each the execution.

As what above will become apparent from reading, scheduling and execution circulation comprise three major parts and optional pre-circulation of handling successively:

-optional sleep cycle is used to guarantee the performance of the memory node at storage server place;

-first is used to manage new request;

-second portion is used to handle the oldest request;

-third part is used to handle the request from using the minimum storage client of storage server.

Very obvious, the circulation independently of one another and that simplify of these parts can only comprise one or several among the latter.The processing that is clear that these parts equally can walk abreast, and can carry out third part, rather than resets this circulation after having carried out " old " request.

Second and third part can be considered as the more particular instance of universal.

Therefore, at second portion, " old " processing of request is arised from the needs of management " exception " request, it is no matter because what is former thereby preferentially do not carried out by this algorithm as yet.This is the guidance design that must remember, because it is contemplated that other embodiments of avoiding these situations.

As for third part, general notion is according to the function dispatch request of the relation between storage client and the storage server as quantitative criteria.Therefore, in the example of being introduced, the quantitative criteria of local memory unit service time is used to distinguish between some requests.

But, can use other quantitative criteria, such as the average level of exchanges data, the averaging network stand-by period during these are mutual, the degree of data-bag lost etc. according to the statistics that characterizes storage client/storage server interaction characteristic.

Moreover, the embodiment of this paper introduction provides as the example of simplifying, and utilizes conventional programming technique (such as other parameters of using buffering area or consideration scheduling) can do further improvement.

In the description that is provided, scheduling is based on the strategy that helps two main shafts: carry out old request and share load between disk and client computer.

(by changing preceding method accordingly) can implement other strategies helping additive method, such as:

-maximally utilising the disk bandwidth, the request aggregation that for example will be close to or be close to vicinity is in single request, thus saving is to the visit of disk.

-maximally utilise the disk stand-by period, be that some optimization requests of target are so that reduce the stand-by period, perhaps by produce the request (promptly pointing to data by prediction request in the future) that can indicate at the storage server place for example by the center that produces at the storage server place with disk.

Other strategies and realization thereof and countless mutation will be apparent to the technical staff.

Therefore, the application program that data have been stored in visit can comprise the driver of the relation between the multiple key element of management, these key elements for example be between mutual, corresponding relation module between mutual, file system and the corresponding relation module between application program and the file system and the storage client alternately; The tactful embodiment of storage server, mode is for obtaining the result and utilize this result (the perhaps form after this results modification) to call next key element from each key element.

As an alternative, described system can be autonomous, does not rely on the application program of calling data, and these key elements can communicate with one another, and makes described information from a downward Es-region propagations of key element to a key element and then catch.

Similarly be, can guarantee the communication between these key elements in a different manner, for example utilize POSIX interface, IP, TCP and udp protocol, shared storage, RDMA (long-range direct access storage).Should be kept in mind that the advantage that the objective of the invention is to provide specialized accumulator system in the heart according to conventional network resources.

More than the system embodiment of Jie Shaoing is Network Based, and the computer that produces station wherein comprises:

* special use or general processor (for example belonging to CISC or RISC type or other types);

The memory of * one or more memory discs (hard disk that for example has serial ATA or scsi interface or other interfaces) or any other type, and

* network interface (for example Gigabit, Ethernet, Infiniband, SCI...)

* based on the application environment of operating system (such as Linux), be used for support application program and file system manager is provided;

* produce the application component of corresponding relation module, for example the Clustered Logical VolumeManager module of Exanodes (registered trade mark) application program of Seanodes (registered trade mark) company;

* produce the storage client of each NBD and the application component of server-client, for example the Exanodes Network Block EM equipment module of Exanodes (registered trade mark) application program of Seanodes (registered trade mark) company;

* the application component of key element, for example the Exanodes Clustered ServiceManager module of Exanodes (registered trade mark) application program of Seanodes (registered trade mark) company are shared in management.

Can in network, produce such system, comprise:

* be suitable for application program of using on the network and the conventional subscriber station that is used as the application program node, and

* the computer based equipment group that produces according to the above description is as the server and the memory node of network.

Other equipments and the application program that produce the alternate embodiment of the device in the scope of the invention will become apparent the technical staff.

The computer based system is contained in the present invention, and it integrally comprises application program node and memory node.It also contains each key element of this computer based system, the application program node and the memory node that especially separate, and the multiple device that is used to produce them.

Equally, data management processes should totally consider, mutual according to application program node and memory node just, but also to consider to be suitable for producing the application program node of this process and each computer based facility of memory node.

More than specific embodiments of the invention have been introduced in Shuo Ming statement.It should be considered as restrictive or describe the latter, and it has covered all combinations of characteristics of some mutation of mutual introduction especially in the mode of restriction.

The present invention also covers the software elements of being introduced as product, and they emerge with any computer-readable " medium " (support).The storage support of magnetic, optics and/or electronics is contained in statement " computer-readable medium ", and transmits to support to be medium, such as the analog or digital signal.

Such medium also comprise software elements self, just are suitable for the directly key element of operation, such as the software elements that is used to install and/or dispose, such as mounting disc or the long-range installation procedure that loads.Such installation can be carried out on client computer station and server station globally, also can carry out dividually with suitable product at every turn.

Claims

1. a computer based system comprises a plurality of computer based facilities that are called the interconnecting nodes in the network,

-some node that is called memory node comprises at least one local direct access storage unit (38) at least, and be arranged to according to the storage server (46) of access request formation (64) management to the visit of this local memory unit (38),

-some is called the node of application program node at least, comprise separately:

* application environment has with the expression of block address form from the addressable file of this node, and each address specifies at least one physical address on the local memory unit (38) of memory node, and

* storage client (44) can be mutual according to access request and any described storage server (46) of physical block address,

It is characterized in that, at least one described storage server (46) comprises scheduler, this scheduler can be carried out the access request that comprises in its formation (64) with the order of appointment, and is that described scheduler is arranged to and this order is defined as forming performance criteria and comprises described formation and/or the function of the rule group of one or more state parameters of the resident memory node on it of this formation.

2. computer based according to claim 1 system is characterized in that, described rule group also comprises one or more state parameters of described local memory unit.

3. computer based according to claim 1 and 2 system, it is characterized in that, described rule group comprises and is suitable for the request that exists in the described formation (64) is selected and based on the rule of the content of the head (62) of every request, and is that this rule does not exist under other data conditions that are associated with these requests and can move.

4. according to the described computer based of one of aforementioned claim system, it is characterized in that described rule group comprises and is suitable for the request that exists in the described formation (64) is selected and based on the rule according to the previous mutual quantitative criteria of setting up between this storage client (44) and the storage server (46) be concerned about.

5. computer based according to claim 4 system is characterized in that, described quantitative criteria is set up according to the duration of the previous load of the storage server of being concerned about (46) by described storage client.

6. according to the described computer based of one of aforementioned claim system, it is characterized in that described rule group comprises and is suitable for the request that exists in the described formation (64) is selected and the rule of the time span that existed based on described request in described formation (64).

7. according to the described computer based of one of aforementioned claim system, it is characterized in that, described rule group comprise be suitable for to the request that exists in the described formation (64) select and the physical address of before having visited based on the physical address of every request appointment and described storage server (46) between the rule of distance.

8. according to the described computer based of one of aforementioned claim system, it is characterized in that, at least the one of described storage server (46) is arranged to and carries out the access request that comprises in its formation (64) in order, and wherein said order is confirmed as the function of selected rule group in the middle of a plurality of Different Rule groups that form strategy about described performance criteria.

9. according to the described computer based of one of aforementioned claim system, it is characterized in that, at least some described storage client (44) goes through in the storage client of being concerned about optionally to send access request to this storage server (46) before the previous access request that sends of given storage server is finished.

10. according to the described computer based of one of aforementioned claim system, it is characterized in that some described storage server (46) is managed several local memory unit (38) at least.

11., it is characterized in that some described node is application program node and memory node at least according to the described computer based of one of aforementioned claim system.

12., it is characterized in that described rule group comprises that the execution that is suitable for the request that will comprise in the described formation moves the rule of selected time span according to the described computer based of one of aforementioned claim system.

13. computer based according to claim 12 system is characterized in that described rule is based on the degree that takies of the resource of described memory node.

14., it is characterized in that described rule is based on criterion selected in the middle of the group of the request type that comprises in request quantity that comprises and the described formation from comprise described formation according to claim 12 or 13 described computer based systems.

15., it is characterized in that described rule is based on the assessment to the elapsed time since described previous moving according to the described computer based of one of claim 12 to 14 system.

16. according to the described computer based of one of aforementioned claim system, it is characterized in that, it comprises corresponding relation module (36), can keep the corresponding relation between at least one physical address on the local memory unit (38) of each block address and memory node.

17. computer based according to claim 16 system, it is characterized in that, described corresponding relation module (36) comprising: mapping table (42), this mapping table (42) contain the corresponding relation between at least one physical address on the local memory unit (38) of each block address and memory node; And motor (40), be designed to given request be defined at least one physical address: with block address inquiry mapping table (42) by following manner, and have the given request of determined physical address to the storage client of being concerned about (44) transmission, so that carry out described given request.

18. a data management processes can be used in the network of the computer based facility that comprises a plurality of interconnection that are called node, may further comprise the steps:

B. according to the corresponding relation between at least one physical address on the local memory unit (38) of the memory node of each virtual address and described network, determine pairing at least one physical address of described virtual address,

C. send the access request that indicates described physical address from storage client (44) to the storage server of managing the visit of the described local memory unit (38) that is associated with described physical address (46),

It is characterized in that described process further may further comprise the steps:

D. described access request is placed in the formation (64) of described storage server (46), and, carry out the access request that comprises in the described formation (64) to be defined as forming performance criteria and to comprise described formation and/or the order of the function of the rule group of one or more state parameters of its resident described memory node on it.

19. process according to claim 18 is characterized in that, described rule group also comprises one or more state parameters of described local memory unit.

20. according to claim 18 or 19 described processes, it is characterized in that, described rule group comprises and is suitable for the request that exists in the described formation (64) is selected and based on the rule of the content of the head (62) of every request, and is that this rule comes into effect not existing under other data cases that are associated with these requests.

21. according to the described process of one of claim 18 to 20, it is characterized in that described rule group comprises and is suitable for the request that exists in the described formation (64) is selected and based on according to the rule of the described storage client (44) of sending every request with the previous mutual quantitative criteria of setting up of the storage server of being concerned about (46).

22. process according to claim 21 is characterized in that, described quantitative criteria is set up according to the duration of the previous load of the storage server of being concerned about (46) by described storage client.

23., it is characterized in that described rule group comprises and is suitable for rule that the request that has had length seclected time in the described formation (64) is selected according to the described process of one of claim 18 to 22.

24. according to the described process of one of claim 18 to 23, it is characterized in that, described rule group comprise be suitable for to the request that exists in the described formation (64) select and the physical address of before having visited based on the physical address of every request appointment and described storage server (46) between the rule of distance.

25., it is characterized in that described rule group comprises selected many rules in the middle of a plurality of Different Rule groups that form strategy about described performance criteria according to the described process of one of claim 18 to 24.

26. according to the described process of one of claim 16 to 23, it is characterized in that, send the steps d of access request in advance to this storage server (46) same access request sender (44). before finishing, the access request sender who is concerned about (44) can reproduce step c with another physical address.

27., it is characterized in that some described storage server (46) comprises formation (64) for each access request sender (44) at least according to the described process of one of claim 18 to 26.

28., it is characterized in that some described storage server (46) is being managed the visit to a plurality of local memory unit (38) at least according to the described process of one of claim 18 to 27.

29., it is characterized in that described rule group comprises that the execution that is suitable for the request that will comprise in the described formation moves the rule of selected time span according to the described process of one of claim 18 to 28.

30. process according to claim 29 is characterized in that, described rule is based on the degree that takies of the resource of described memory node.

31., it is characterized in that described rule is based on criterion selected in the middle of the group of the request type that comprises in request quantity that comprises and the described formation according to claim 29 or 30 described processes from comprise described formation.

32., it is characterized in that described rule is based on the assessment to the elapsed time since described previous moving according to the described process of one of claim 29 to 31.

33. a computer based equipment that forms memory node is arranged to the steps d of carrying out described process according to one of claim 18 to 32..

34. a computer program comprises some program code devices, is suitable for when moving on computers carrying out according to the described process of one of claim 18 to 33.