CN102480502B - I/O load equilibrium method and I/O server - Google Patents
I/O load equilibrium method and I/O server Download PDFInfo
- Publication number
- CN102480502B CN102480502B CN201010562558.0A CN201010562558A CN102480502B CN 102480502 B CN102480502 B CN 102480502B CN 201010562558 A CN201010562558 A CN 201010562558A CN 102480502 B CN102480502 B CN 102480502B
- Authority
- CN
- China
- Prior art keywords
- load
- server
- clock
- value
- load value
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention provides an I/O load equilibrium method and an I/O server. The I/O load equilibrium method is applied in the I/O server. A storage system is formed by at least two I/O servers. The method comprises the following steps: calculating a second load value of a second moment according to a prediction algorithm in a first moment; acquiring a load average of the storage system in the second moment; when the second load value is greater than the load average, migrating a data file to be migrated to the target I/O server, wherein the load value of the target I/O server in the second moment is less than the second load value. By using the method in the embodiment of the invention, a load balancing strategy is determined based on the load value in the next moment so that the equilibrium of the system load in the next moment can be realized. Therefore, even if there is network communication time delay right now, the influence of the network communication time delay on the system load equilibrium in the next moment is small. Frequent load migration caused by the network communication time delay can be reduced.
Description
Technical field
The present invention relates to technical field of data storage, relate in particular to a kind of I/O load-balancing method and I/O server.
Background technology
Along with the rapid growth of data message amount, cluster file system, cluster are stored in high-performance calculation, the Internet, video field and are widely used.Cluster storage is data to be distributed to the storage mode of each node in cluster, can provide the storage solution that possesses high I/O bandwidth and flexible extensibility for unstructured data.
In cluster storage, along with rolling up of I/O number of servers, can cause serious network delay and load imbalance phenomenon, reduce the I/O service quality of cluster storage system, reduce the utilance of storage resources.In order to make full use of cluster, store existing resource, further improve parallel I/O access ability, between I/O server, realize load balancing and just seem and be even more important.
In prior art, changing load and irregular in most of cluster storages, conventionally adopt dynamic load balancing method, concrete, by a centralized servers Real-Time Monitoring and collect the load data of every I/O server, then according to the data of collecting, by algorithm, come decision-making how to carry out the migration of load, to realize the load balancing between each I/O server of current time.But when storage system is on a grand scale, the traffic between server can be very large, if whole system load is very high, more there will be obvious network service time delay phenomenon, and network service time delay meeting has influence on centralized servers collection load data, and then can postpone making of load-balancing decision.For example, if current time has A, two I/O server loads of B high, dynamic load leveling determines the load of migration A, B, but because network service time delay has caused the delay of load balancing, when pending this load balancing carries out load migration, the load of A, B may reduce, and when upper once load balancing, may again the load of just moving away from A, B be moved again.Therefore, dynamic load balancing method of the prior art, when there is network service time delay, can produce the frequent migration of load between I/O server.
Summary of the invention
The embodiment of the present invention provides a kind of I/O load-balancing method and I/O server, can, when there is network service time delay, reduce the frequent migration of load between I/O server.
In order to solve the problems of the technologies described above, the technical scheme of the embodiment of the present invention is as follows:
The embodiment of the present invention provides a kind of I/O load-balancing method, and described method is applied in an I/O server, with at least two described I/O servers, forms a storage system, comprising:
At first o'clock, inscribe the second load value that calculated for second moment according to a prediction algorithm;
Obtain the load average value that described storage system was inscribed at described second o'clock;
When described the second load value is greater than described load average value, data file to be migrated is migrated to target I/O server, described target I/O server is less than described second load value at the described load value of inscribing at second o'clock.
Further, also comprise:
Obtain the first load value of inscribing at described first o'clock;
When described the second load value is greater than described the first load value, then described in carrying out, obtain the step of the load average value that described storage system inscribed at described second o'clock.
Further, also comprise:
Obtain the load balancing factor that described storage system was inscribed at described second o'clock;
When the described load balancing factor is less than load balancing factor threshold value, then carry out the described step that data file to be migrated is migrated to target I/O server.
Further, described I/O server obtains the load average value that place storage system was inscribed at described second o'clock, comprising:
Other I/O server to described storage system sends the request of collecting the described load value of inscribing at second o'clock;
Receive described other I/O server feedback at the described load value of inscribing for second o'clock;
According to described other I/O server, at the described load value of inscribing and described the second load value, calculate the load average value that described storage system was inscribed at described second o'clock at second o'clock.
Further, described data file to be migrated is migrated to target I/O server, comprising:
Select data file to be migrated;
To described target I/O server, send migration request;
Receive the migration response that described target I/O server returns;
Described data file to be migrated is sent to described target I/O server, and deletes local data file described to be migrated.
Further, the data file that described selection is to be migrated, comprising:
Calculate the load value of each data file and the ratio of data volume of storage;
The data file of the predetermined quantity of selection ratio maximum is as data file to be migrated.
Further, also comprise:
According to a predetermined policy other I/O servers of described storage system determine one described in target I/O server;
Described predetermined policy is that described target I/O server is not more than described I/O server at the described load value second o'clock inscribed in the load value sum of the described load value of inscribing and described data file to be migrated at second o'clock.
The embodiment of the present invention also provides a kind of I/O server, is applied to, in a storage system of at least two described I/O servers formations, comprising:
Load estimation unit, for inscribing the second load value that calculated for second moment according to a prediction algorithm at first o'clock;
Load acquiring unit, the load average value of inscribing at described second o'clock for obtaining described storage system;
Load migration unit, for when described the second load value is greater than described load average value, migrates to target I/O server by data file to be migrated, and described target I/O server is less than described second load value at the described load value of inscribing at second o'clock.
Further, also comprise:
Load calculation unit, for calculating the first load value of inscribing at first o'clock;
Described load acquiring unit, also for when described the second load value is greater than described the first load value, then obtains the load average value that described storage system was inscribed at described second o'clock.
Further, also comprise:
Balance factor acquiring unit, the load balancing factor of inscribing at described second o'clock for obtaining described storage system;
Described load migration unit, migrates to target I/O server also for when the described load balancing factor is less than load balancing factor threshold value, then by data file to be migrated.
Further, described load acquiring unit comprises:
Load requests subelement, sends the request of collecting the described load value of inscribing at second o'clock for other I/O server to described storage system;
Load receives subelement, for receive described other I/O server feedback at the described load value of inscribing for second o'clock;
Computation subunit, for calculating at the described load value of inscribing and described the second load value the load average value that described storage system was inscribed at described second o'clock according to described other I/O server at second o'clock.
Further, described load migration unit comprises:
Chooser unit, for selecting data file to be migrated;
Migration request subelement, for sending migration request to described target I/O server;
Response receives subelement, the migration response sending for receiving described target I/O server;
Migration subelement, for described data file to be migrated is sent to described target I/O server, and deletes local data file described to be migrated.
Further, described chooser unit comprises:
Ratio calculation subelement, for calculating the load value of each data file and the ratio of data volume of storage;
File chooser unit, for selecting the data file of predetermined quantity of ratio maximum as data file to be migrated.
Further, described load migration unit also comprises:
Determine subelement, for according to a predetermined policy other I/O servers of described storage system determine one described in target I/O server; Described predetermined policy is that described target I/O server is not more than described I/O server at the described load value second o'clock inscribed in the load value sum of the described load value of inscribing and described data file to be migrated at second o'clock.
Next load value constantly of I/O server prediction in the embodiment of the present invention, and obtain the load average value of system, and then carry out the comparison of load value, when comparative result satisfies condition, then initiates load migration, the load balancing of etching system when meeting next.Embodiment of the present invention method is by determining load balancing based on next load value constantly, realized the equilibrium to next system load constantly, even so there is network service time delay in current time, in the time of can be to next, the load balancing of etching system produce considerable influence, thereby reduced the load frequent causing because of network service time delay, moves yet.
Accompanying drawing explanation
In order to be illustrated more clearly in the embodiment of the present invention or technical scheme of the prior art, to the accompanying drawing of required use in embodiment or description of the Prior Art be briefly described below, apparently, accompanying drawing in the following describes is only some embodiments of the present invention, for those of ordinary skills, do not paying under the prerequisite of creative work, can also obtain according to these accompanying drawings other accompanying drawing.
Fig. 1 is the flow chart of a kind of I/O load-balancing method of the embodiment of the present invention;
Fig. 2 is the flow chart of the another kind of I/O load-balancing method of the embodiment of the present invention;
Fig. 3 is the method flow diagram of load migration in the another kind of I/O load-balancing method of the embodiment of the present invention;
Fig. 4 is the structural representation of a kind of I/O server of the embodiment of the present invention;
Fig. 5 is the structural representation of the another kind of I/O server of the embodiment of the present invention;
Fig. 6 is the structural representation of a kind of storage system of the embodiment of the present invention.
Embodiment
In order to make those skilled in the art can further understand feature of the present invention and technology contents, refer to following about detailed description of the present invention and accompanying drawing, accompanying drawing only provide with reference to and explanation, be not used for limiting the present invention.
Dynamic load balancing method of the prior art is due to the load data of the I/O server that need to collect according to current time, load between each I/O server of current time is carried out to equilibrium, so, once network communication delay will cause obtaining delaying of load data, and then can postpone making of load balancing, dynamic change due to load, this load balancing may not be suitable for the loading condition of each I/O server after time delay, carry out again this load balancing and cause possibly inaccurate load migration, load imbalance that even can heavy system, and then can produce the frequent migration of load.
Based on this, the embodiment of the present invention has proposed a kind of I/O load-balancing method and I/O server, the load value in next moment of I/O server prediction, and obtain the load average value of system, and then carry out the comparison of each load value, when comparative result satisfies condition, then initiate load migration, the load balancing of etching system when meeting next.Embodiment of the present invention method is by determining load balancing based on next load value constantly, realized the equilibrium to next system load constantly, even so there is network service time delay in current time, in the time of can be to next, the load balancing of etching system produce considerable influence, thereby reduced the load frequent causing because of network service time delay, moves yet.
Below in conjunction with drawings and Examples, technical scheme of the present invention is described.
Referring to Fig. 1, it is the flow chart of a kind of I/O load-balancing method of the embodiment of the present invention.
The method is applied in an I/O server, and at least two I/O servers form a storage system, and the method can comprise:
Step 101, I/O server was inscribed according to prediction algorithm and is calculated the second load value of inscribing at second o'clock at first o'clock.
In the present embodiment, the prediction algorithm that I/O server obtains according to the forecast model based on certain is predicted at next moment (second constantly) possible load value own, also the prediction load value that this second load value is this I/O server, the actual value of not inscribing at second o'clock.Wherein, first is only two time points distinguishing front and back with second constantly constantly, not refers in particular to specifically point sometime.Two time intervals between constantly can be set as required, and very fast if for example system load situation changes, setting-up time interval is shorter, slower if loading condition changes, can setting-up time interval long etc.Prediction algorithm can be that such as linear prediction auto-correlation (Levinson-Durbin) algorithm, red pond information criterion, autoregression AR forecast model etc. repeats no more herein according to existing model and algorithm.Load value can refer to the IO throughput of I/O server etc., specifically can set according to the parameter that will realize load balancing.
I/O server can be real-time when carrying out this step, can be also periodic awakening also periodically carry out this step, this cycle also can be set as required, repeats no more herein.
Step 102, obtains the load average value that storage system was inscribed at second o'clock.
I/O server was inscribed and is obtained after the load value of inscribing for second o'clock at first o'clock, further obtain the load average value of inscribing this I/O server place storage system at second o'clock, concrete acquisition methods can be to calculate and obtain after the load value inscribed during to other I/O server request the second of this I/O server, also can directly be sent by other I/O server, specifically refer to the description of subsequent embodiment.This load average value of inscribing for second o'clock is also predicted value.
Step 103, when the second load value is greater than load average value, migrates to target I/O server by data file to be migrated.
When the second load value of I/O server is greater than the load average value of storage system, illustrate this I/O server at the second loading condition constantly with respect to other I/O server in storage system, will be in high load condition, to in second moment, reach the load balancing of storage system, the partial document data in this I/O server need to be migrated in the relatively low I/O server of load.Now, this I/O server migrates to this locality data file to be migrated in target I/O server, and the load value that this target I/O server was inscribed at second o'clock is less than the second load value.
Certainly in this step, can also increase other qualifications, with realizing under the prerequisite of the second moment storage system load balancing, reduce load migration as far as possible.Specifically refer to subsequent embodiment.
In the present embodiment, every I/O server in storage system all moves above-mentioned equalization methods, and the structure of this independent distribution formula can be avoided the problem of bottleneck and the Single Point of Faliure of traditional centralized management method; This load-balancing algorithm can reduce the mutual dependence between I/O server, allows as far as possible workload on book server, only have meet certain condition server just needs further collect necessary information.The embodiment of the present invention is by determining load balancing based on next load value constantly, realized the equilibrium to next system load constantly, even so there is network service time delay in current time, in the time of can be to next, the load balancing of etching system produce considerable influence, thereby reduced the load frequent causing because of network service time delay, moves yet.
Referring to Fig. 2, it is the flow chart of the another kind of I/O load-balancing method of the embodiment of the present invention.
The method can comprise:
Step 201, I/O server calculates the first load value a inscribing at first o'clock.
I/O server can adopt the computational methods of prior art, calculates the load value a inscribing at first o'clock, and this first moment is also current time.
Step 202, I/O server was inscribed at first o'clock according to prediction algorithm and is calculated the second load value b inscribing at second o'clock.
Then according to prediction algorithm, predict the load value b inscribing at second o'clock.This step is similar with the step 101 in previous embodiment, repeats no more herein.
Step 203, when b is greater than a, obtains the load average value c that I/O server place storage system was inscribed at second o'clock.
During load value a that the load value b inscribing second time when I/O server inscribes while being greater than first, the load that this I/O server is described will increase, and then obtain the load average value c inscribing of this I/O server place storage system at second o'clock, to determine whether need to carry out load migration to realize the load balancing of storage system.
If b is less than or equal to c, illustrate that interior other I/O server of storage system also exists the situation of load rise, and it is more than this I/O server load, to raise, and the loading condition of this I/O server is still in average level or subaverage, and this I/O server is without carrying out load migration.If b is greater than c, illustrate that this I/O server load raises comparatively serious, proceed to next step and further judge whether to carry out load migration.
Wherein, the load average value c inscribing for second of storage system o'clock can obtain by following steps:
This I/O server sends the request of collecting the load value of inscribing at second o'clock to other I/O server of place storage system;
This I/O server receives the load value of inscribing at second o'clock that other I/O server sends;
The load value that this I/O server is inscribed at second o'clock according to other I/O server and the second load value of this I/O server, calculate the load average value that storage system was inscribed at second o'clock.
In other embodiments, can be also that this I/O server is sent to other a certain I/O server by the own load value of inscribing at second o'clock, by other I/O server, calculate and obtain after mean value, be sent to this I/O server.
Step 204, I/O server obtains the load balancing factor that storage system was inscribed at second o'clock.
In the present embodiment, I/O server can also further obtain the load balancing factor that storage system was inscribed at second o'clock, and this load balancing factor can embody storage system in the degree of whole system load balancing sometime.If this load balancing factor is less than default load balancing factor threshold value, illustrate that the load of storage system is unbalance, need to carry out load migration to realize load balancing.
Wherein, this load balancing factor threshold value, for weighing the lower limit of this storage system load balancing, is less than this lower limit explanation load imbalance, and concrete numerical value can be set as required, is not construed as limiting herein.
The method that I/O server obtains the storage system load balancing factor can be the load average value c calculating acquisition based on storage system, the load average value of storage system that concrete can be accounts for the percentage of storage system maximum load value, and this maximum load value is to set according to the memory property of storage system.This load balancing factor is larger, illustrates that the load of storage system is more balanced.The computational methods of load balancing require effectively simple, do not increase as far as possible the extra load of I/O server under high capacity.Add after the load balancing factor after the reference conditions as migration, can improve the availability of load migration decision-making, avoid the action of some invalid load migrations.
Step 205, when b is greater than c, and the load balancing factor of storage system is while being less than load balancing factor threshold value, and I/O server migrates to target I/O server by data file to be migrated.
When this I/O server second load value is constantly greater than the load average value of storage system, and the load balancing factor of storage system is less than load balancing factor threshold value, illustrate that this I/O server need to carry out load migration, to reduce the load value of self, with the load of balanced storage system.Obviously, the load value that this I/O server carries out the target I/O server of load migration at least needs to be less than the load value of this I/O server.
Wherein, concrete load migration process can as shown in Figure 3, comprise:
Step 301, I/O server is selected data file to be migrated.
In the present embodiment, I/O server determines to carry out, after load migration, need to selecting data file to be migrated, and this selection course can be: first, calculate the load value of each data file and the ratio of data volume of storage.Then, the data file of the some of selection ratio maximum is as data file to be migrated.
Wherein, data file load value and data volume ratio are larger, illustrate that the load value that this data file accounts in this I/O server is larger, can preferentially select to move this data file, can realize the load value that a small amount of this class data file of migration can this I/O of fast reducing server.The data file of the some of selection ratio maximum is as data file to be migrated, and wherein, this some can be made an appointment, and in the present embodiment, can select front ten data files of ratio maximum as data file to be migrated.
Step 302, I/O server is determined a target I/O server according to a predetermined policy at other I/O servers of storage system.
In the present embodiment, this predetermined policy can be:
The load value sum of the load value that target I/O server was inscribed at second o'clock and data file to be migrated is not more than the load value that this I/O server was inscribed at second o'clock; Or the load value sum of the load value that target I/O server was inscribed at second o'clock and data file to be migrated is not more than load value poor of load value that this I/O server inscribed at second o'clock and data file to be migrated.Certainly, in other embodiments, can also according to the loading condition of storage system, set the condition of other select target I/O server, will not enumerate herein.
Adopting such strategy is in order to judge when certain load is from the I/O server migration of high capacity during to low load target I/O server, the load of target I/O server can not be higher than I/O server second load value constantly, could eliminate like this high capacity I/O server, the load Distribution situation of the whole storage system of efficient balance.
This step can be carried out with previous step or adjustment order simultaneously.
Step 303, I/O server sends migration request to target I/O server.
After this I/O server selected data file and target I/O server to be migrated, obtain the attribute of data file to be migrated, and send migration request to target I/O server, in this request, can comprise the attribute of this data file to be migrated.
Target I/O server receives after migration request, can in this locality, set up new file attribute according to the attribute of the data file to be migrated comprising in this request, then to I/O server, sends migration response.
Step 304, the migration response of I/O server receiving target I/O server feedback.
Step 305, I/O server is sent to target I/O server by data file to be migrated, and deletes local data file to be migrated.
I/O server receives after migration response, data file to be migrated can be sent to target I/O server, then this data file of this locality is deleted.After having moved, I/O server can also further notify meta data server to upgrade the distributed intelligence of this data file.
Embodiment of the present invention method has realized the equilibrium to next system load constantly, and adds after the load balancing factor, can improve the availability of load migration decision-making, avoids the action of some invalid load migrations.So even current time exists network service time delay, in the time of can be to next, the load balancing of etching system produce considerable influence yet, thereby reduced the load frequent causing because of network service time delay, moves; On the other hand, by each load value relatively, to whether carrying out the situation of load balancing, screen, more reduced the frequent migration of load.
Referring to Fig. 4, it is the structural representation of a kind of I/O server of the embodiment of the present invention.
In the present embodiment, in the storage system that this I/O server is applied at least consist of two I/O servers, this I/O server can comprise:
Load estimation unit 401, for inscribing the second load value that calculated for second moment according to a prediction algorithm at first o'clock.
Load acquiring unit 402, the load average value of inscribing at second o'clock for obtaining storage system.
Load migration unit 403, for when the second load value is greater than load average value, migrates to target I/O server by data file to be migrated, and the load value that target I/O server was inscribed at second o'clock is less than the second load value.
The prediction algorithm of load estimation unit 401 based on certain predicted at next possible load value of the moment own.Prediction algorithm can be that such as linear prediction auto-correlation (Levinson-Durbin) algorithm, red pond information criterion, autoregression AR forecast model etc. repeats no more herein according to existing model and/or algorithm.Load acquiring unit 402 further obtains the load average value of inscribing this I/O server place storage system at second o'clock, concrete acquisition methods can be to calculate and obtain after the load value inscribed during to other I/O server request the second of this I/O server, also can directly be sent by other I/O server.When the second load value of I/O server is greater than the load average value of storage system, illustrate this I/O server at the second loading condition constantly with respect to other I/O server in storage system, in high load condition, to in second moment, reach the load balancing of storage system, the partial document data in this I/O server need to be migrated in the relatively low I/O server of load.Now, load migration unit 403 migrates to this locality data file to be migrated in target I/O server, realizes the equilibrium of storage system load.
The embodiment of the present invention has realized the equilibrium to next system load constantly by said units, even so there is network service time delay in current time, in the time of can be to next, the load balancing of etching system produce considerable influence, thereby reduced the load frequent causing because of network service time delay, moves yet.
Referring to Fig. 5, it is the structural representation of the another kind of I/O server of the embodiment of the present invention.
This I/O server, except can comprising load estimation unit 501, load acquiring unit 502, load migration unit 503, can also comprise load calculation unit 504 and balance factor acquiring unit 505.
Wherein, load estimation unit 501 is similar with the load estimation unit 401 in previous embodiment, repeats no more herein.
Load calculation unit 504, for calculating the first load value of inscribing at first o'clock.
Load acquiring unit 502, can also be for when the second load value be greater than the first load value, then obtains the load average value that storage system was inscribed at second o'clock.
Balance factor acquiring unit 505, the load balancing factor of inscribing at second o'clock for obtaining storage system.
Load migration unit 503, can also be when being less than load balancing factor threshold value when the load balancing factor, then data file to be migrated is migrated to target I/O server.
Concrete, load acquiring unit 502 can comprise again:
Load requests subelement 5021, sends the request of collecting the load value of inscribing at second o'clock for other I/O server to storage system.
Load receives subelement 5022, for receive other I/O server feedback at the described load value of inscribing for second o'clock.
Computation subunit 5023, calculates for load value and the second load value of inscribing at second o'clock according to other I/O server the load average value that storage system was inscribed at second o'clock.
Load migration unit 503 can comprise again:
Chooser unit 5031, for selecting data file to be migrated.Chooser unit 5031 can also comprise ratio calculation subelement and file chooser unit.Wherein, ratio calculation subelement is for calculating the load value of each data file and the ratio of data volume of storage; File chooser unit is for selecting the data file of predetermined quantity of ratio maximum as data file to be migrated.
Determine subelement 5032, for determining a target I/O server according to a predetermined policy at other I/O servers of storage system; The load value sum of the load value that predetermined policy can be inscribed at second o'clock for target I/O server and data file to be migrated is not more than the load value that I/O server was inscribed at second o'clock.
Migration request subelement 5033, for sending migration request to target I/O server.
Response receives subelement 5034, the migration response sending for receiving target I/O server.
Migration subelement 5035, for data file to be migrated being sent to target I/O server, and deletes local data file to be migrated.
501 predictions of load estimation unit obtain the second load value that I/O server was inscribed at second o'clock, load calculation unit 504 calculates and obtains I/O server at first o'clock after the first load value of inscribing, by comparing, if the second load value is greater than the first load value, load acquiring unit 502 receives subelement 5022 and computation subunit 5023 by load requests subelement 5021, load, obtain the load average value that the storage system at this I/O server place was inscribed at second o'clock, balance factor acquiring unit 505 obtains the load balancing factor that storage system was inscribed at second o'clock.When the second load value of this I/O server is greater than load average value, and when the load balancing factor of storage system is less than default threshold value, load migration unit 503 by chooser unit 5031, determine that subelement 5032, migration request subelement 5033, response receive subelement 5034 and migration subelement 5035 migrates to target I/O server by data file to be migrated.The load value sum of the load value that this target I/O server was inscribed at second o'clock and data file to be migrated is not more than the load value that this server of I/O was inscribed at second o'clock.
The embodiment of the present invention is determined load balancing by said units based on next load value constantly, realized the equilibrium to next system load constantly, even so there is network service time delay in current time, in the time of can be to next, the load balancing of etching system produce considerable influence, thereby reduced the load frequent causing because of network service time delay, moves yet; On the other hand, by a plurality of unit relatively each load value to whether carrying out the situation of load balancing, screen, more reduced the frequent migration of load.
Referring to Fig. 6, it is the structural representation of a kind of storage system of the embodiment of the present invention.
This storage system comprises at least two I/O servers, and it is example that the present embodiment be take three I/O servers 601,602,603.
Every I/O server is all for inscribing the second load value that calculated for second moment according to a prediction algorithm at first o'clock; Obtain the load average value that storage system was inscribed at second o'clock; When the second load value is greater than load average value, data file to be migrated is migrated to target I/O server, the load value that target I/O server was inscribed at second o'clock is less than the second load value.
First every I/O server can also obtain the first load value of inscribing at first o'clock, when the second load value is greater than the first load value, then obtains the step of the load average value that storage system inscribed at second o'clock.First every I/O server also can also obtain the load balancing factor that storage system was inscribed at second o'clock, when the load balancing factor is less than load balancing factor threshold value, then data file to be migrated is migrated to the step of target I/O server.
In the present embodiment, I/O server 601,602,603 calculates respectively the load value under own current time (being also first moment), and according to prediction algorithm, calculate the load value of inscribing at second o'clock, then ask respectively other I/O server second load value constantly, calculate the load average value that obtains storage system, and calculate the load balancing factor that storage system is inscribed in acquisition at second o'clock.Suppose that the load value that I/O server is inscribed is greater than the load value of inscribing at first o'clock at 601 second o'clock, and be greater than load average value, and the load balancing factor of inscribing storage system at second o'clock is less than load balancing factor threshold value, I/O server 601 selects the data file of ratio maximum of load value and data volume as data file to be migrated in local data file, at I/O server 602, the load value sum of the load value of selecting inscribe in 603 at second o'clock and data file to be migrated is not more than of load value that I/O server 601 inscribed at second o'clock, as target I/O server, it is for example I/O server 602.I/O server 601 sends migration request to I/O server 602, after receiving the migration response that I/O server 602 returns, data file to be migrated is sent to I/O server 602, and the local data file to be migrated of deletion, complete the migration of data file from the I/O server 601 of high capacity to the I/O server 602 of low load, thereby realize the load balancing in storage system.
In the present embodiment, storage system by all moving above-mentioned equalization methods on every I/O server, and the structure of this independent distribution formula can be avoided the problem of bottleneck and the Single Point of Faliure of traditional centralized management method; This load-balancing algorithm can reduce the mutual trust between I/O server, allows as far as possible workload on book server, only have meet certain condition server just needs further collect necessary information.The embodiment of the present invention is determined load balancing by every I/O server based on next load value constantly, realized the equilibrium to next system load constantly, even so there is network service time delay in current time, in the time of can be to next, the load balancing of etching system produce considerable influence, thereby reduced the load frequent causing because of network service time delay, moves yet.
More than in device, the specific implementation process of each unit, subelement and system please refer to the corresponding description in preceding method embodiment, repeats no more herein.
Above-described embodiment of the present invention, does not form limiting the scope of the present invention.Any modification of doing within the spirit and principles in the present invention, be equal to and replace and improvement etc., within all should being included in claim protection range of the present invention.
Claims (14)
1. an I/O load-balancing method, described method is applied in an I/O server, with at least two described I/O servers, forms a storage system, it is characterized in that, comprising:
At first o'clock, inscribe the second load value that calculated for second moment according to a prediction algorithm;
Obtain the load average value that described storage system was inscribed at described second o'clock, described load average value calculates for the load value inscribed at second o'clock according to other I/O server of described storage system and the second load value of described I/O server;
When described the second load value is greater than described load average value, data file to be migrated is migrated to target I/O server, described target I/O server is less than described second load value at the described load value of inscribing at second o'clock.
2. method according to claim 1, is characterized in that, also comprises:
Obtain the first load value of inscribing at described first o'clock;
When described the second load value is greater than described the first load value, then described in carrying out, obtain the step of the load average value that described storage system inscribed at described second o'clock.
3. method according to claim 1, is characterized in that, also comprises:
Obtain the load balancing factor that described storage system was inscribed at described second o'clock;
When the described load balancing factor is less than load balancing factor threshold value, then carry out the described step that data file to be migrated is migrated to target I/O server.
4. method according to claim 1, is characterized in that, described I/O server obtains the load average value that place storage system was inscribed at described second o'clock, comprising:
Other I/O server to described storage system sends the request of collecting the described load value of inscribing at second o'clock;
Receive described other I/O server feedback at the described load value of inscribing for second o'clock;
According to described other I/O server, at the described load value of inscribing and described the second load value, calculate the load average value that described storage system was inscribed at described second o'clock at second o'clock.
5. method according to claim 1, is characterized in that, described data file to be migrated is migrated to target I/O server, comprising:
Select data file to be migrated;
To described target I/O server, send migration request;
Receive the migration response that described target I/O server returns;
Described data file to be migrated is sent to described target I/O server, and deletes local data file described to be migrated.
6. method according to claim 5, is characterized in that, the data file that described selection is to be migrated, comprising:
Calculate the load value of each data file and the ratio of data volume of storage;
The data file of the predetermined quantity of selection ratio maximum is as data file to be migrated.
7. method according to claim 6, is characterized in that, also comprises:
According to a predetermined policy other I/O servers of described storage system determine one described in target I/O server;
Described predetermined policy is that described target I/O server is not more than described I/O server at the described load value second o'clock inscribed in the load value sum of the described load value of inscribing and described data file to be migrated at second o'clock.
8. an I/O server, is applied to, in a storage system of at least two described I/O servers formations, it is characterized in that, comprising:
Load estimation unit, for inscribing the second load value that calculated for second moment according to a prediction algorithm at first o'clock;
Load acquiring unit, for obtaining the load average value that described storage system was inscribed at described second o'clock, described load average value calculates for the load value inscribed at second o'clock according to other I/O server of described storage system and the second load value of described I/O server;
Load migration unit, for when described the second load value is greater than described load average value, migrates to target I/O server by data file to be migrated, and described target I/O server is less than described second load value at the described load value of inscribing at second o'clock.
9. I/O server according to claim 8, is characterized in that, also comprises:
Load calculation unit, for calculating the first load value of inscribing at first o'clock;
Described load acquiring unit, also for when described the second load value is greater than described the first load value, then obtains the load average value that described storage system was inscribed at described second o'clock.
10. I/O server according to claim 8, is characterized in that, also comprises:
Balance factor acquiring unit, the load balancing factor of inscribing at described second o'clock for obtaining described storage system;
Described load migration unit, migrates to target I/O server also for when the described load balancing factor is less than load balancing factor threshold value, then by data file to be migrated.
11. I/O servers according to claim 8, is characterized in that, described load acquiring unit comprises:
Load requests subelement, sends the request of collecting the described load value of inscribing at second o'clock for other I/O server to described storage system;
Load receives subelement, for receive described other I/O server feedback at the described load value of inscribing for second o'clock;
Computation subunit, for calculating at the described load value of inscribing and described the second load value the load average value that described storage system was inscribed at described second o'clock according to described other I/O server at second o'clock.
12. I/O servers according to claim 8, is characterized in that, described load migration unit comprises:
Chooser unit, for selecting data file to be migrated;
Migration request subelement, for sending migration request to described target I/O server;
Response receives subelement, the migration response sending for receiving described target I/O server;
Migration subelement, for described data file to be migrated is sent to described target I/O server, and deletes local data file described to be migrated.
13. I/O servers according to claim 12, is characterized in that, described chooser unit comprises:
Ratio calculation subelement, for calculating the load value of each data file and the ratio of data volume of storage;
File chooser unit, for selecting the data file of predetermined quantity of ratio maximum as data file to be migrated.
14. I/O servers according to claim 13, is characterized in that, described load migration unit also comprises:
Determine subelement, for according to a predetermined policy other I/O servers of described storage system determine one described in target I/O server; Described predetermined policy is that described target I/O server is not more than described I/O server at the described load value second o'clock inscribed in the load value sum of the described load value of inscribing and described data file to be migrated at second o'clock.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201010562558.0A CN102480502B (en) | 2010-11-26 | 2010-11-26 | I/O load equilibrium method and I/O server |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201010562558.0A CN102480502B (en) | 2010-11-26 | 2010-11-26 | I/O load equilibrium method and I/O server |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102480502A CN102480502A (en) | 2012-05-30 |
| CN102480502B true CN102480502B (en) | 2014-10-01 |
Family
ID=46092982
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201010562558.0A Active CN102480502B (en) | 2010-11-26 | 2010-11-26 | I/O load equilibrium method and I/O server |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102480502B (en) |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2014047902A1 (en) * | 2012-09-28 | 2014-04-03 | 华为技术有限公司 | Load balancing method, device, system and computer readable medium |
| CN102968281A (en) * | 2012-11-26 | 2013-03-13 | 华为技术有限公司 | Data migration method and device |
| CN104468674B (en) * | 2013-09-25 | 2020-01-14 | 南京中兴新软件有限责任公司 | Data migration method and device |
| CN104573020B (en) * | 2015-01-12 | 2017-11-10 | 浪潮电子信息产业股份有限公司 | Data Autonomic Migration Framework optimization method in a kind of hierarchical stor |
| CN107783720A (en) * | 2016-08-24 | 2018-03-09 | 深圳市深信服电子科技有限公司 | A kind of data balancing method and storage device |
| CN114143326A (en) * | 2021-12-08 | 2022-03-04 | 深圳前海微众银行股份有限公司 | Load adjustment method, management node, and storage medium |
| CN114448897B (en) * | 2021-12-29 | 2024-01-02 | 天翼云科技有限公司 | Target migration method and device |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1512729A (en) * | 2002-12-31 | 2004-07-14 | 联想(北京)有限公司 | Method for network equipment self adaption load equalization |
| CN101593133A (en) * | 2009-06-29 | 2009-12-02 | 北京航空航天大学 | Load balancing of resources of virtual machine method and device |
| CN101695050A (en) * | 2009-10-19 | 2010-04-14 | 浪潮电子信息产业股份有限公司 | Dynamic load balancing method based on self-adapting prediction of network flow |
| CN101820384A (en) * | 2010-02-05 | 2010-09-01 | 浪潮(北京)电子信息产业有限公司 | Method and device for dynamically distributing cluster services |
-
2010
- 2010-11-26 CN CN201010562558.0A patent/CN102480502B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1512729A (en) * | 2002-12-31 | 2004-07-14 | 联想(北京)有限公司 | Method for network equipment self adaption load equalization |
| CN101593133A (en) * | 2009-06-29 | 2009-12-02 | 北京航空航天大学 | Load balancing of resources of virtual machine method and device |
| CN101695050A (en) * | 2009-10-19 | 2010-04-14 | 浪潮电子信息产业股份有限公司 | Dynamic load balancing method based on self-adapting prediction of network flow |
| CN101820384A (en) * | 2010-02-05 | 2010-09-01 | 浪潮(北京)电子信息产业有限公司 | Method and device for dynamically distributing cluster services |
Also Published As
| Publication number | Publication date |
|---|---|
| CN102480502A (en) | 2012-05-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102480502B (en) | I/O load equilibrium method and I/O server | |
| CN108182105B (en) | Local dynamic migration method and control system based on Docker container technology | |
| CN112416554B (en) | Task migration method and device, electronic equipment and storage medium | |
| US20190319881A1 (en) | Traffic management based on past traffic arrival patterns | |
| US10078639B1 (en) | Cluster file system comprising data mover modules having associated quota manager for managing back-end user quotas | |
| US8914501B2 (en) | Method, apparatus, and system for scheduling distributed buffer resources | |
| CN106648456B (en) | Dynamic copies file access method based on user's amount of access and forecasting mechanism | |
| CN110933139A (en) | System and method for solving high concurrency of Web server | |
| CN103595805A (en) | Data placement method based on distributed cluster | |
| KR102245341B1 (en) | Method for apply predictive model for workload distribution within the cloud edge | |
| CN108255427B (en) | Data storage and dynamic migration method and device | |
| CN107729570B (en) | Data migration method and device for server | |
| WO2020042612A1 (en) | Method and device for storing and reading a message, server, and storage medium | |
| CN104539744A (en) | Two-stage media edge cloud scheduling method and two-stage media edge cloud scheduling device | |
| Tu et al. | An optimized cluster storage method for real-time big data in Internet of Things | |
| CN110597598B (en) | Control method for virtual machine migration in cloud environment | |
| JP7192645B2 (en) | Information processing device, distributed processing system and distributed processing program | |
| CN113301126B (en) | Edge computing method suitable for heterogeneous networking gateway | |
| CN109815204A (en) | A kind of metadata request distribution method and equipment based on congestion aware | |
| WO2023089350A1 (en) | An architecture for a self-adaptive computation management in edge cloud | |
| CN101483668A (en) | Network storage and access method, device and system for hot spot data | |
| CN103442087A (en) | Web service system access volume control device and method based on response time trend analysis | |
| CN112637286A (en) | Network architecture adjusting method, device, system and computer readable storage medium | |
| WO2015055502A2 (en) | Method of partitioning storage in a distributed data storage system and corresponding device | |
| Guo | Ant colony optimization computing resource allocation algorithm based on cloud computing environment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant |