CN109634746A - A kind of the utilization system and optimization method of web cluster caching - Google Patents
A kind of the utilization system and optimization method of web cluster caching Download PDFInfo
- Publication number
- CN109634746A CN109634746A CN201811481419.8A CN201811481419A CN109634746A CN 109634746 A CN109634746 A CN 109634746A CN 201811481419 A CN201811481419 A CN 201811481419A CN 109634746 A CN109634746 A CN 109634746A
- Authority
- CN
- China
- Prior art keywords
- value
- server
- request
- user
- web server
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5005—Allocation of resources, e.g. of the central processing unit [CPU] to service a request
- G06F9/5011—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resources being hardware resources other than CPUs, Servers and Terminals
- G06F9/5016—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resources being hardware resources other than CPUs, Servers and Terminals the resource being the memory
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5083—Techniques for rebalancing the load in a distributed system
Landscapes
- Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a kind of utilization systems of web cluster caching, are made of load-balanced server, data analytics server and web cluster server, the web cluster server is made of multiple destination web servers.Web cluster caching of the invention can be solved effectively existing for the system of existing web cluster caching using system when the request list of certain time period hot spot data generates variation, according to data analytics server ought analysis next time instruct the web server to carry out database table caching, it will lead to the content frequent updating of web server caching, cache hit rate reduces, and system leads to the problem of fluctuation.
Description
Technical field
The present invention relates to technical field of the computer network, in particular to a kind of the utilization system and optimization of web cluster caching
Method.
Background technique
With the further development of cloud computing and Web 2.0, many enterprises or tissue can face more demands often: big
Concurrent user's access of amount, thousands of concurrent transaction processing per second, flexible elastic and scalability, low delay etc., pass
Affairs type application of uniting starts to change to the application of limit issued transaction type.A kind of limit issued transaction of the cluster cache as key
Technology can provide the technical solution of high-throughput, low delay for affairs type application.Its postpone write mechanism can provide it is shorter
Response time, while the issued transaction load of database is greatly reduced, event-driven framework, can support to advise greatly stage by stage
The transaction request of mould, high concurrent.In addition, cluster cache manages affairs in memory and provides the consistency guarantor of data
Barrier realizes high availability using Data Replication Technology in Mobile, has preferably scalability and combining properties.
Summary of the invention
The purpose of the present invention is being based on background technique, the utilization system and optimization method of a kind of web cluster caching, solution are provided
Certainly in the prior art, with the variation of system business and increase, the hot spot data of user's request there are changed possibility, when
The request list of certain time period hot spot data generates variation, according to data analytics server ought analysis next time instruct
Web server carries out database table caching, will lead to the content frequent updating of web server caching, cache hit rate reduces, is
System generates the technical issues of fluctuation.
In order to reach above-mentioned technical effect, the present invention takes following technical scheme:
A kind of utilization system of web cluster caching, is taken by load-balanced server, data analytics server and web cluster
Business device composition, the web cluster server are made of multiple destination web servers;
The data analytics server be used for record and counting user request log information, according to historical heat record with
Node probability chained list obtains the classification standard of two fields of database table and destination server chained list comprising that need to access;Load is equal
Weighing apparatus server is used to determine the destination web server of user's request according to the classification standard that data analytics server determines, and will
User's request is forwarded to corresponding destination web server, and the chained list collection of alternative target web server is traversed by mobile pointer
It closes, realizes the equilibrium of load;User request of the destination web server for the forwarding of balancing received load server, and will have
The database table of high user request amount is cached in memory, is then directly read from memory when receiving corresponding user request
Corresponding data in database table are taken to generate result and return.
Meanwhile the invention also discloses a kind of optimization methods of Web cluster cache, it is slow by a kind of above-mentioned web cluster
The utilization system deposited realizes that the optimization method of the Web cluster cache specifically includes at data prediction process and request forwarding
Process is managed, the data prediction process includes completing the record of user's request, statistics and analysis, requests corresponding mesh in user
Load in advance needs the database table accessed in mark web server, and the request forward process process includes passing through load balancing
Server compares received user's request and classification standard, determines the destination web server of the corresponding forwarding of request and completes to request
Forwarding, then user is received by destination web server and requests and handles, finally return to result.
Further, the data prediction process specifically includes the following steps:
S1: change the attenuation process of the property settings node probability of speed according to the business of system:
S1.1 determines initial value N according to system featuresinit: Ninit=N0e-al
Wherein, NinitFor probability of final extinction initial value, the node for representing current slot is important in entire historical record
Degree, N0=1, l determine the offset of initial value, and a is attenuation coefficient, indicate as time change history is for future anticipation
The value of the size variation of influence, a and l change the characteristic self-defining of speed according to the business of system;
S1.2 is according to completion value NfinishDefined formula: Nfinish=Nkf=N0e-a(m+l), according to the parameter a and l of definition,
K value is successively increased, the completion value N of each node is calculatedfinish, NfinishFor probability of final extinction completion value, node is represented to prediction
Minimum probability with directive significance, works as Nfinish=Nkf=N0e-a((k+1)T+l)< NminWhen, stop calculating, at this point, the value of k is
For the length of probability chained list and historical heat data acquisition system chained list;Wherein, NminIt is NfinishA reference value, for indicating to go through
History node is to the lower limit of the probability right of prediction, and m indicates that time interval unit is the second, if preset time interval is T, m=
KT, k are integer.
The k value that S1.3 is determined according to step S1.2, calculates separately the corresponding node probability of integer node within the scope of 1~k, counts
Calculate formula are as follows: Nkf=N0e-a(kT+l), by node probability according to sequential storage from small to large into node probability chained list;
S2. data analytics server saves the log recording of user's requested database in preset time period, every request day
Will is both needed to record user and requests the database table information that need to be accessed;The classification for determining user's request is recorded according to history log, it will
User's request of access same database table is divided into one kind, obtains the set of initial hotspots tables of data according to sixteen principles;
S3. the chained list record for traversing k secondary hot spots data constructs key- in conjunction with the node probability chained list that step S1 is obtained
The key-value pair of value: if key value is not present, the value of node probability chained list corresponding node is taken to be assigned to value;If key value
It has existed, using the probability value for adding node on the basis of value is worth, map is subjected to descending sort according to value value,
It is no more than the database table of value value summation 80% as hot spot set according to the sum of sequencing selection value value.
Further, in the step S1.2, Nmin=0.2.
Further, the request forward process process specifically includes the following steps:
S4. according to the classification standard of the obtained hot spot data table set and history of step S3, hot spot data table set is compared
With the tables of data in the classification standard of history, the classification standard for needing to update and newly-increased hot spot data table are filtered out, will be updated
Classification standard in corresponding destination web server label put back to unappropriated set, to newly-increased hot spot data table according to list
Request amount and the transmission capacity of destination web server calculate the destination web server quantity needed, never allocated column in the time of position
The destination web server of table selection corresponding number constructs new classification standard;
S5: destination web server receives the classification request of data analytics server forwarding, and determining needs database to be loaded
The title of table, from the content of database load table into memory;
S6: load-balanced server receives user's request, and the classification standard of correlation data Analysis server determines that user asks
Affiliated classification is asked, load-balanced server is determined from the destination server set of classification standard by the node that pointer is directed toward
The number of destination server;
S7: destination web server balancing received load server forwarding user request, and to the content of request at
Reason, corresponding data are obtained from memory and are returned as a result.
Further, the step S2 specifically includes the following steps:
S2.1 data analytics server saves the log recording of user's requested database in preset time period, every request day
Will is both needed to record user and requests the database table information that need to be accessed;
The user's request for accessing same database table is divided into one kind by S2.2, counts user's request amount of each classification,
Be ranked up according to the sequence of request amount from big to small, sequence be defined in preceding 20% classification, generate comprising number with
And the classification standard of database table, that is, hotspot database table set;
The hotspot database table set that step S2.2 is obtained is added to the bidirectional circulating of hot spot data historical record by S2.3
Chained list checks the length of double-linked circular list, if chained list length is not more than k, is no longer operated;If the length of chained list is big
In k, then the next node of the double-linked circular list of hot spot data historical record is deleted.
Further, the step S3 specifically includes the following steps:
The array set for the present node that the pointer that S3.1 obtains hot spot data table historical record is directed toward, moves the pointer over
Variable i is recorded in number, and i initial value is 1, the name of first database table is taken out from array set, by database table name
As key value;Meanwhile the node probability of k-1 is designated as under taking in node probability chained list as value value, by key-value key assignments
To storage into map set, remaining database table name in array set is successively taken out by this, by corresponding key-value
It stores in map set;
S3.2 before continuing to move to pointer, first judge node probability chained list next hot spot data set whether be
Sky, if next hot spot data set is not sky, the pointer of mobile hot spot data table historical record, otherwise, COLLECTION TRAVERSALSThe is complete
Finish;
The database that S3.3 takes out any i stage respectively from hotspot database table and probability chained list show with it is corresponding general
Rate value first obtains whether current database table name has existed value in map set when carrying out the assignment operation of map set,
If had existed, it is saved in map set using value value and the sum of node probability value as new value, if do not deposited
Then storing node probability value as value into map set;
After S3.4 hotspot database table and probability chained list traverse, map set is subjected to descending row according to value value
Sequence, finally, selecting the sum of value value closest but being no more than 80% tables of data set of value value summation as caching
Hot spot data table set.
Further, the step S4 specifically includes the following steps:
S4.1 takes out database table name from the classification standard of history, inquires the database table name of taking-up in hot spot data table
It whether there is in set, correspond to key-value key-value pair if it does, removing in map set;It is corresponded to if it does not, deleting
Classification standard, and the set of destination web server is put back into unappropriated server list;
S4.2 takes out database table name from map set, according to the corresponding request amount of the database table of record, asks for classification
It asks and specifies sufficient amount of destination web server, according to the service ability of destination web server and the classification in the unit time
User's request amount calculates the quantity of destination web server, i.e. the service ability of server is greater than the service that user requests,
In, the service ability of destination web server includes the processing capacity and transmission flow of server, using the number of server as collection
Close the destination server field that classification standard is written in the form of circular linked list;
S4.3 sends each classification standard to corresponding single or more according to the definition of destination web server field
On a destination web server, with the corresponding database table of caching classification, load balancing then is sent by all classification standards
Server.
Further, the step S6 specifically includes the following steps:
S6.1 load-balanced server receives and stores all classification standards of data analytics server transmission, at each point
A node is arbitrarily selected in the circular linked list of the destination web server field of class, and pointer is made to be directed toward this node;
When receiving user's request, comparison-of-pair sorting's standard is executed as follows S6.2 with determining classification belonging to user's request
Forwarding strategy:
S6.2.1 obtains the number that pointer is directed toward node, and user's request is forwarded to the representative target web service of number
On device, later, mobile pointer to next node position;
S6.2.2 increases a field newly in classification standard, is more than default occupancy threshold for storing cpu occupancy
The set of destination web server number,
The cpu occupancy of destination web server in S6.2.3 monitoring objective web server field, once target web service
The cpu occupancy of device reaches default occupancy threshold, and the number storage of this destination web server is increased newly to step S6.2.2
In field;
The cpu occupancy of destination web server in S6.2.4 monitoring step S6.2.2 field, when destination web server
When cpu occupancy is down to the half for being no more than default occupancy threshold, again by the corresponding number of this destination web server
It is inserted into the circular linked list of destination web server.
Further, the step S7 specifically includes the following steps:
S7.1 destination web server receives the classification request that data analytics server is sent, and determines that classification request is corresponding
Database table, from load table content in database into memory;
User's request of S7.2 destination web server balancing received load server forwarding, determines the classification of user's request
The database table that need to be accessed is consistent with the cache contents in memory, and data information is loaded from memory and returns to user's request.
Compared with prior art, the present invention have it is below the utility model has the advantages that
The utilization system and optimization method of web cluster caching through the invention, it is slow can effectively to solve existing web cluster
The system deposited is existing when the request list of certain time period hot spot data generates variation, according to data analytics server when next
Secondary analysis come instruct web server carry out database table caching, will lead to web server caching content frequent updating, delay
Hit rate reduction is deposited, system leads to the problem of fluctuation.
Detailed description of the invention
Fig. 1 is the schematic diagram using system of web cluster caching of the invention.
Fig. 2 is the schematic diagram of the optimization method of Web cluster cache of the invention.
Specific embodiment
Below with reference to the embodiment of the present invention, the invention will be further elaborated.
Embodiment:
Embodiment one:
As shown in Figure 1, a kind of utilization system of web cluster caching, the system are taken by data analytics server, load balancing
Business device and web cluster server composition.Data analytics server is responsible for the log information of record and counting user request, according to going through
History hot spot record and node probability chained list obtain point of two fields of database table and destination server chained list comprising that need to access
Class standard;The classification standard that load-balanced server is sent according to data analytics server carry out user's requests classification determination and
Forward process is traversed the chained list set of alternate servers by mobile pointer, realizes the equilibrium of load;Web cluster server is negative
Duty caching has the database table of high user request amount into memory, later, after receiving user's request of this classification, from
Result is obtained in memory and is returned.
Wherein, the work of data analytics server mainly includes following aspect:
Node probability is saved to probability chained list, i.e., probability attenuation process is defined according to the setting of system, including decaying is initially
Value, die-away time length and decaying completion value, calculate the corresponding probability of each node, store into node probability chained list.Together
When, the length for defining historical heat data base table recording is k.Save the history log of user's requested database in current slot
Record, every Request Log are both needed to the type of record user's request and request corresponding database table information;K-1 heat before saving
The historical analysis of point data table records, wherein the hot spot data table that this hot spot data table analysis record is cached with web server is gone through
Records of the Historian record has differences, and is that data analytics server takes according to the web that the history log that user in current slot requests obtains
The hot spot data table set record of business device caching, referred to as initial hotspots tables of data set, and what usage history record prediction obtained
Hot spot data table set, referred to as caching hot spot data table set.Save last classification standard, including database table name and right
The destination server answered.It specifically includes:
First, change the attenuation process of the property settings node probability of speed according to the business of system:
Specifically, probability of final extinction initial value NinitFormula are as follows: Ninit=N0e-al, probability of final extinction initial value NinitRepresentative is worked as
Significance level of the node of preceding period in entire historical record, N0The offset of=1, l decision initial value.L is controlled, it can
Change the size of initial value.It is very fast in business variation, in the more system of new business, the initial value of decaying can be set
It is high, new business data are compensated based on the accounting in historical forecast;Change in less system in business, the initial value of decaying
Lower, the probability accounting of reduction new business data can be set.A indicates attenuation coefficient, indicates with time change, history for
The size variation of the influence of future anticipation.
The determination method of a and l specifically: according to the record of data analytics server and analysis, available initial hotspots
Tables of data set, it is assumed that length k.Analyzed using three hot spot data table set of arbitrary continuation, respectively correspond i-1, i and
The stage that tri- predetermined time periods of i+1 are T (i is the integer not less than 2).Compare the initial hotspots number in i stage and i+1 stage
According to set, the probability that the i stage occurs in i+1 stage initial hotspots data acquisition system, i.e. weighing factor are obtained.It is assumed that the i stage
Initial hotspots tables of data collection is combined into { student, user, order, product }, the initial hotspots tables of data set in i+1 stage
For { student, user, order, picture }, then Probability pi=3/4=0.75.In the same way, the i-1 stage is compared
With the initial hotspots data in i+1 stage, weighing factor p of available (i-1)-th stage relative to the i+1 stagei-1.Preset time
Between be divided into T, compute repeatedly to obtain n*k-2 p in n times of kT time spaniAnd pi-1, the value of n is integer, it is proposed that value range
For (10,50).If the value of n is too small, the accuracy of probability is influenced;But value is too big, will increase unnecessary calculating.It is right
(n*k-2) secondary probability is averaged respectively, obtains i stage and i-1 stage for the average weight of i+1 stage hot spot data set
With
I-th of stage:
According toWithJoint solves, and obtains the value of a and l.
Carry out probability of final extinction completion value NfinishDefinition:
Probability of final extinction completion value NfinishDefined formula are as follows: Nfinish=Nkf=N0e-a(m+l), probability of final extinction completion value
NfinishRepresent the minimum probability that node has directive significance to prediction.Wherein, m indicates time interval, and unit is the second.It is at this
In system, preset time interval is T, then m=kT, k are integer.NfinishIt needs to preset a reference value NminFor indicating history
Lower limit of the node to the probability right of prediction.If Nmin=0.2, according to the parameter a and l of definition, k value is successively increased, N is worked asfinish
=Nkf=N0e-a((k+1)T+l)When < 0.2, stop calculating.At this point, the value of k is probability chained list and historical heat data acquisition system chain
The length of table.
According to determining k value, the corresponding probability of integer node within the scope of 1~k is calculated separately, from small to large according to probability
Sequential storage is into probability chained list.
Wherein, within the scope of 1~k the corresponding probability of integer node calculation formula are as follows: Nkf=N0e-a(kT+l), specifically, with
Upper probability analysis task theoretically only needs in system initialization or there are larger adjustment when progress.
Second, it is recorded according to history log in current slot and determines initial hotspots database table set:
Firstly, determining classification and the needs of user's request according to classification standard according to the data historian log of record
The database table of access.Determine classification standard: the database table that user's request need to access is same database table.
Then, according to the data historian log of record, classification and the needs of user's request are determined according to classification standard
The database table of access.It specifically includes: counting user's request amount of each type, i.e. request number of times;Count the use of each classification
Family request amount is ranked up according to the sequence of request amount from big to small.If user's request amount is identical, the data volume of computation requests
As sort by, i.e., it is ranked up from big to small according to request data quantity to obtain hotspot database table set.Consider
Under most business scenario, 80% amount of access is all concentrated on 20% dsc data (sixteen principles), to sequence preceding
20% classification is defined (note: (k1, k2…k10) list, take (k1, k2)), generate point comprising number and database table
Class standard.The sum of classification accounting of selection can be less than 20%, but no more than 20%.
Then, the hotspot database table set that above-mentioned steps obtain is added to the bidirectional circulating of hot spot data historical record
Chained list: array set is added to chained list tail portion in mobile tail pointer.Later, the length of chained list is checked.If chained list length is not small
In k, no longer operated;If the length of chained list is greater than k, mobile pointer deletes the bidirectional circulating chain of hot spot data historical record
The next node of table.
Third records each tables of data in the historical record of prediction subsequent time period according to the chained list of k secondary hot spots data
Probability as hot spot:
Firstly, obtaining the array set for the present node that pointer is directed toward, the number moved the pointer over is recorded variable i, at the beginning of i
Initial value is 1.The name that first database table is taken out from array set, using database table name as key value;Meanwhile taking chain
The probability of k-1 is designated as under in table as value value, by the storage of key-value key-value pair into map set.Successively take out array
Remaining database table name in set, by corresponding key-value storage into map.
Then, it before continuing to move to the pointer of chained list of storage hot spot data table name set, needs first to judge probability chain
Whether the next of table is empty, the next hot spot data set of next expression direction.If i < 1, next are not empty, mobile hot spot
The pointer of tables of data historical record.Otherwise, next is sky, and COLLECTION TRAVERSALSThe finishes.
Then, taken out respectively from two tables of hotspot database table and probability chained list any i stage database table name and
Corresponding probability value.At this point, whether to need first to obtain current database table name in map when carrying out the assignment operation of map
Through existence value.If had existed, need for the sum of value value probability value corresponding with node to be saved in as new value
In map.If it does not, storing probability value as value into map.
Finally, map set is carried out descending row according to value value after hotspot database table and probability chained list traverse
Sequence.It selects the sum of value value closest but is no more than hot spot number of the 80% tables of data set of value value summation as caching
According to table set.
4th, new classification standard is generated according to hot spot data table set and history classification standard:
Firstly, take out database table name from classification standard, inquire the database table name of taking-up in hot spot set whether
In the presence of.Key-value key-value pair is corresponded to if it does, removing in map;If it does not, corresponding classification standard is deleted, it will
The set of web server is put back into unappropriated server list.
Then, database table name is taken out from map, according to the corresponding request amount of the database table of record, for classification request
Specify sufficient amount of web server.According to the service ability of server such as transmitted per unit time capacity and in the unit time
User's request amount of the classification calculates the quantity of destination web server, so that the service ability of server is requested greater than user
Service, wherein the service ability of web server includes the processing capacity and the transmission aspect of flow two of server.It will service
The destination server field of classification standard is written as set in the form of circular linked list for the number of device.
Finally, sending corresponding list for each classification standard according to the definition of destination server field in classification standard
In a or multiple web servers, with the corresponding database table of caching classification.Later, it sends all classification standards to negative
Carry equalization server.
The work of load-balanced server mainly includes following aspect:
First, all classification standards of data analytics server transmission are received and stored, in the destination service of each classification
A node is arbitrarily selected in the circular linked list of device field, and pointer is made to be directed toward this node.
Second, when receiving user's request, comparison-of-pair sorting's standard is to determine classification belonging to user's request.Execute following turn
Hair strategy:
Step 1 obtains the number that pointer is directed toward node, and user's request is forwarded to the representative web server of number
On.Later, pointer is moved to next node position.
Step 2 increases a field newly in classification standard, compiles for storing cpu occupancy more than the server of preset value
Number set.
Step 3, the cpu occupancy of web server in monitoring objective server field, once the cpu of web server is accounted for
The value for reaching setting with rate, by the number storage of this web server into the newly-increased field of previous step.
Step 4, the cpu occupancy of web server in two field of monitoring step, when the cpu occupancy of web server drops
When below to the half of preset value, it is re-inserted into the chained list of destination server by the corresponding number of this server.
The work of web cluster server mainly includes following aspect:
First, destination web server receives the classification request that data analytics server is sent, and determines that classification request is corresponding
Database table, from load table content in database into memory.
Second, user's request of destination web server balancing received load server forwarding determines the classification of user's request
The database table that need to be accessed is consistent with the cache contents in memory, and data information is loaded from memory and returns to user's request.
Embodiment two
As shown in Fig. 2, a kind of optimization method of Web cluster cache, the utilization system cached by above-mentioned web cluster is real
It is existing, data prediction process and request forward process process are specifically included, the data prediction process includes completing user to ask
Record, the statistics and analysis asked request the database that load needs access in advance on corresponding destination web server in user
Table, the request forward process process include that received user's request and classification standard are compared by load-balanced server, really
The destination web server of the fixed corresponding forwarding of request and the forwarding for completing request, then user's request is received simultaneously by destination web server
Processing, finally returns to result.
Specifically includes the following steps:
S1. change the attenuation process of the property settings node probability of speed according to the business of system:
S1.1 determines initial value N according to system featuresinit: Ninit=N0e-al
Wherein, NinitFor probability of final extinction initial value, the node for representing current slot is important in entire historical record
Degree, N0=1, l determine the offset of initial value, and a is attenuation coefficient, indicate as time change history is for future anticipation
The value of the size variation of influence, a and l change the characteristic self-defining of speed according to the business of system;
Specifically, the determination method of a and l: according to the record of data analytics server and analysis, available initial hotspots
Tables of data set, it is assumed that length k.It is analyzed now using three hot spot data table set of arbitrary continuation, respectively corresponds i-
1, tri- predetermined time periods of i and i+1 are the stage of T.The initial hotspots data acquisition system for comparing i stage and i+1 stage, obtains i
The probability that stage occurs in i+1 stage initial hotspots data acquisition system, i.e. weighing factor.It is assumed that the initial hotspots data in i stage
Table collection is combined into { student, user, order, product }, the initial hotspots tables of data collection in i+1 stage be combined into student,
User, order, picture }, then Probability pi=3/4=0.75.In the same way, i-1 stage and i+1 stage are compared
Initial hotspots data, weighing factor p of available (i-1)-th stage relative to the i+1 stagei-1.Prefixed time interval is T, in n
Times kT time span computes repeatedly to obtain n*k-2 piAnd pi-1, the value of n is integer, it is proposed that value range is (10,50).Such as
The value of fruit n is too small, influences the accuracy of probability;But value is too big, will increase unnecessary calculating.It is secondary to (n*k-2) general
Rate is averaged respectively, obtains i stage and i-1 stage for the average weight of i+1 stage hot spot data setWith
I-th of stage:
It can basisWithJoint solves, and obtains the value of a and l
S1.2 is according to completion value NfinishDefined formula: Nfinish=Nkf=N0e-a(m+l), according to the parameter a and l of definition,
K value is successively increased, the completion value N of each node is calculatedfinish, NfinishFor probability of final extinction completion value, node is represented to prediction
Minimum probability with directive significance, works as Nfinish=Nkf=N0e-a((k+1)T+l)< NminWhen, stop calculating, at this point, the value of k is
For the length of probability chained list and historical heat data acquisition system chained list;Wherein, NminIt is NfinishA reference value, for indicating to go through
For history node to the lower limit of the probability right of prediction, m indicates time interval, if preset time interval is T, m=kT, k are whole
Number, m unit are the second.
The k value that S1.3 is determined according to step S1.2, calculates separately the corresponding node probability of integer node within the scope of 1~k, counts
Calculate formula are as follows: Nkf=N0e-a(kT+l), by node probability according to sequential storage from small to large into node probability chained list.
The above probability analysis task theoretically only needs in system initialization or there are larger adjustment when progress.
S2. it is recorded according to history log in current slot and determines initial hotspots database table set:
S2.1 determines the classification of user's request according to classification standard and needs to visit according to the data historian log of record
The database table asked.Determine classification standard: the database table that user's request need to access is same database table;
S2.2 determines the classification of user's request according to classification standard and needs to visit according to the data historian log of record
The database table asked.According to the data historian log of record, classification and the needs of user's request are determined according to classification standard
The database table of access.Count user's request amount of each type, i.e. request number of times.User's request amount of each classification is counted,
It is ranked up according to the sequence of request amount from big to small.If user's request amount is identical, the data volume of computation requests is as sequence
Foundation, i.e. being ranked up from big to small according to request data quantity.Consider 80% amount of access under most business scenario
(sixteen principles) is all concentrated on 20% dsc data, sequence is defined (such as: (k in preceding 20% classification1, k2…k10)
List takes (k1, k2)), generate classification standard, that is, hotspot database table set comprising number and database table.Point chosen
The sum of class accounting can be less than 20%, but no more than 20%;
The hotspot database table set that step S2.2 is obtained is added to the bidirectional circulating of hot spot data historical record by S2.3
Chained list: array set is added to chained list tail portion in mobile tail pointer.Later, the length of chained list is checked.If chained list length is not small
In k, no longer operated;If the length of chained list is greater than k, mobile pointer deletes the bidirectional circulating chain of hot spot data historical record
The next node of table.
S3. each data telogenesis in the historical record of prediction subsequent time period is recorded according to the chained list of k secondary hot spots data
For the probability of hot spot:
S3.1 obtains the array set for the present node that pointer is directed toward, and the number moved the pointer over is recorded variable i, at the beginning of i
Initial value is 1.The name that first database table is taken out from array set, using database table name as key value;Meanwhile taking chain
The probability of k-1 is designated as under in table as value value, by the storage of key-value key-value pair into map set.Successively take out array
Remaining database table name in set, by corresponding key-value storage into map.
S3.2 first judges probability chained list before continuing to move to the pointer of chained list of storage hot spot data table name set
Whether next is empty, the next hot spot data set of next direction.If i < 1, next are not sky, mobile hot spot data table is gone through
The pointer of Records of the Historian record.Otherwise, next is sky, and COLLECTION TRAVERSALSThe finishes.
S3.3 takes the database table name and correspondence in any i stage respectively from two tables of hotspot database table and probability chained list
Probability value.At this point, needing first to obtain whether current database table name has been deposited in map when carrying out the assignment operation of map
It is being worth.If had existed, need the sum of value value probability value corresponding with node being saved in map as new value
In.If it does not, storing probability value as value into map.
After S3.4 chained list traverses, map is subjected to descending sort according to value value, selects the sum of value value closest
But it is no more than hot spot data table set of the 80% tables of data set of value value summation as caching.
S4. according to the classification standard of the obtained hot spot data table set and history of step S3, hot spot data table set is compared
With the tables of data in the classification standard of history, the classification standard for needing to update and newly-increased hot spot data table are filtered out, will be updated
Classification standard in corresponding destination web server label put back to unappropriated set, to newly-increased hot spot data table according to list
Request amount and the transmission capacity of destination web server calculate the destination web server quantity needed (at least so that total in the time of position
Destination web server transmitted per unit time capacity not less than request amount in the unit time), never distribution list selection pair
The destination web server of quantity is answered to construct new classification standard;
S4.1 takes out database table name from classification standard, and whether the database table name for inquiring taking-up is deposited in hot spot set
?.Key-value key-value pair is corresponded to if it does, removing in map;If it does not, corresponding classification standard is deleted, by web
The set of server is put back into unappropriated server list.
S4.2 takes out database table name from map, according to the corresponding request amount of the database table of record, refers to for classification request
Fixed sufficient amount of web server.According to the service ability of server and in the unit time, user's request amount of the classification is calculated
The quantity of destination web server out, the i.e. service ability of server need to be greater than the service of user's request, wherein web server
Service ability includes two aspects of processing capacity and transmission flow of server.Using the number of server as set with endless-chain
The destination server field of the form write-in classification standard of table.
S4.3 sends each classification standard to corresponding single according to the definition of destination server field in classification standard
Or in multiple web servers, with the corresponding database table of caching classification.Later, load is sent by all classification standards
Equalization server.
S5: destination web server receives the classification request of data analytics server forwarding, and determining needs database to be loaded
The title of table, from the content of database load table into memory;
S6: load-balanced server receives user's request, and the classification standard of correlation data Analysis server determines that user asks
Affiliated classification is asked, load-balanced server is determined from the destination server set of classification standard by the node that pointer is directed toward
The number of destination server;
S6.1 load-balanced server receives and stores all classification standards of data analytics server transmission, at each point
A node is arbitrarily selected in the circular linked list of the destination web server field of class, and pointer is made to be directed toward this node;
When receiving user's request, comparison-of-pair sorting's standard is executed as follows S6.2 with determining classification belonging to user's request
Forwarding strategy:
S6.2.1 obtains the number that pointer is directed toward node, and user's request is forwarded to the representative target web service of number
On device, later, mobile pointer to next node position;
S6.2.2 increases a field newly in classification standard, is more than default occupancy threshold for storing cpu occupancy
The set of destination web server number,
The cpu occupancy of destination web server in S6.2.3 monitoring objective web server field, once target web service
The cpu occupancy of device reaches default occupancy threshold, and the number storage of this destination web server is increased newly to step S6.2.2
In field;
The cpu occupancy of destination web server in S6.2.4 monitoring step S6.2.2 field, when destination web server
When cpu occupancy is down to the half for being no more than default occupancy threshold, again by the corresponding number of this destination web server
It is inserted into the circular linked list of destination web server.
S7: destination web server balancing received load server forwarding user request, and to the content of request at
Reason, corresponding data are obtained from memory and are returned as a result:
S7.1 destination web server receives the classification request that data analytics server is sent, and determines that classification request is corresponding
Database table, from load table content in database into memory;
User's request of S7.2 destination web server balancing received load server forwarding, determines the classification of user's request
The database table that need to be accessed is consistent with the cache contents in memory, and data information is loaded from memory and returns to user's request.
It is understood that the principle that embodiment of above is intended to be merely illustrative of the present and the exemplary implementation that uses
Mode, however the present invention is not limited thereto.For those skilled in the art, essence of the invention is not being departed from
In the case where mind and essence, various changes and modifications can be made therein, these variations and modifications are also considered as protection scope of the present invention.
Claims (10)
1. a kind of utilization system of web cluster caching, which is characterized in that by load-balanced server, data analytics server and
Web cluster server composition, the web cluster server are made of multiple destination web servers;
The data analytics server is used to record and the log information of counting user request, according to historical heat record and node
Probability chained list obtains the classification standard of two fields of database table and destination server chained list comprising that need to access;Load balancing clothes
Business device is used to determine the destination web server of user's request according to the classification standard that data analytics server determines, and by user
Request is forwarded to corresponding destination web server, and the chained list set of alternative target web server is traversed by mobile pointer,
Realize the equilibrium of load;User request of the destination web server for the forwarding of balancing received load server, and will have higher
The database table of user's request amount is cached in memory, and number is then directly read from memory when receiving corresponding user request
Result is generated according to data corresponding in the table of library and is returned.
2. a kind of optimization method of Web cluster cache, which is characterized in that slow by a kind of web cluster as described in claim 1
The utilization system deposited realizes that the optimization method of the Web cluster cache specifically includes at data prediction process and request forwarding
Process is managed, the data prediction process includes completing the record of user's request, statistics and analysis, requests corresponding mesh in user
Load in advance needs the database table accessed in mark web server, and the request forward process process includes passing through load balancing
Server compares received user's request and classification standard, determines the destination web server of the corresponding forwarding of request and completes to request
Forwarding, then user is received by destination web server and requests and handles, finally return to result.
3. a kind of optimization method of Web cluster cache according to claim 2, which is characterized in that the data prediction
Process specifically includes the following steps:
S1: change the attenuation process of the property settings node probability of speed according to the business of system:
S1.1 determines initial value N according to system featuresinit: Ninit=N0e-al
Wherein, NinitFor probability of final extinction initial value, significance level of the node of current slot in entire historical record is represented,
N0=1, l determine the offset of initial value, and a is attenuation coefficient, indicate the influence with time change history for future anticipation
Size variation, the value of a and l changes the characteristic self-defining of speed according to the business of system;
S1.2 is according to completion value NfinishDefined formula: Nfinish=Nkf=N0e-a(m+l), according to the parameter a and l of definition, successively
Increase k value, calculates the completion value N of each nodefinish, work as Nfinish=Nkf=N0e-a((k+1)T+l)< NminWhen, stop calculating, this
When, the value of k is the length of probability chained list and historical heat data acquisition system chained list;Wherein, NfinishFor probability of final extinction completion value,
Represent the minimum probability that node has directive significance to prediction, NminIt is NfinishA reference value, for indicating history section
Point is to the lower limit of the probability right of prediction, and m indicates that time interval unit is the second, if preset time interval is T, m=kT, k
For integer;
The k value that S1.3 is determined according to step S1.2 calculates separately the corresponding node probability of integer node within the scope of 1~k, calculates public
Formula are as follows: Nkf=N0e-a(kT+l), by node probability according to sequential storage from small to large into node probability chained list;
S2. data analytics server saves the log recording of user's requested database in preset time period, and every Request Log is equal
User need to be recorded and request the database table information that need to be accessed;The classification for determining user's request is recorded according to history log, will be accessed
User's request of same database table is divided into one kind, obtains the set of initial hotspots tables of data according to sixteen principles;
S3. the chained list record for traversing k secondary hot spots data, in conjunction with the node probability chained list that step S1 is obtained, constructs key-value's
Key-value pair: if key value is not present, the value of node probability chained list corresponding node is taken to be assigned to value;If key value has been deposited
Using the probability value for adding node on the basis of value is worth, map is being subjected to descending sort according to value value, according to row
Sequence selects the sum of value value to be no more than the database table of value value summation 80% as hot spot set.
4. a kind of optimization method of Web cluster cache according to claim 3, which is characterized in that in the step S1.2,
Nmin=0.2.
5. a kind of optimization method of Web cluster cache according to claim 3, which is characterized in that at the request forwarding
Manage process specifically includes the following steps:
S4: it according to the classification standard of the obtained hot spot data table set and history of step S3, compares hot spot data table set and goes through
Tables of data in the classification standard of history filters out the classification standard for needing to update and newly-increased hot spot data table, by point of update
Corresponding destination web server label puts back to unappropriated set in class standard, when to newly-increased hot spot data table according to unit
Interior request amount and the transmission capacity of destination web server calculate the destination web server quantity needed, and never distribution list is selected
The destination web server for selecting corresponding number constructs new classification standard;
S5: destination web server receives the classification request of data analytics server forwarding, and determining needs database table to be loaded
Title, from the content of database load table into memory;
S6: load-balanced server receives user's request, and the classification standard of correlation data Analysis server determines that user requests institute
The classification of category, load-balanced server determine target by the node that pointer is directed toward from the destination server set of classification standard
The number of server;
S7: user's request of destination web server balancing received load server forwarding, and the content of request is handled,
Corresponding data are obtained from memory to return as a result.
6. a kind of optimization method of Web cluster cache according to claim 3, which is characterized in that the step S2 is specific
The following steps are included:
S2.1 data analytics server saves the log recording of user's requested database in preset time period, and every Request Log is equal
User need to be recorded and request the database table information that need to be accessed;
The user's request for accessing same database table is divided into one kind by S2.2, counts user's request amount of each classification, according to
The sequence of request amount from big to small is ranked up, and is defined, is generated comprising number and number in preceding 20% classification to sequence
According to classification standard, that is, hotspot database table set of library table;
The hotspot database table set that step S2.2 is obtained is added to the double-linked circular list of hot spot data historical record by S2.3,
It checks the length of double-linked circular list, if chained list length is not more than k, is no longer operated;If the length of chained list is greater than k,
Then delete the next node of the double-linked circular list of hot spot data historical record.
7. a kind of optimization method of Web cluster cache according to claim 5, which is characterized in that the step S3 is specific
The following steps are included:
The array set for the present node that the pointer that S3.1 obtains hot spot data table historical record is directed toward, the number moved the pointer over
Be recorded variable i, i initial value is 1, the name of first database table is taken out from array set, using database table name as
Key value;Meanwhile the node probability that k-1 is designated as under taking in node probability chained list deposits key-value key-value pair as value value
It stores up in map set, remaining database table name in array set is successively taken out by this, corresponding key-value is stored
Into map set;
S3.2 first judges whether next hot spot data set of node probability chained list is sky, such as before continuing to move to pointer
The next hot spot data set of fruit is not sky, and the pointer of mobile hot spot data table historical record, otherwise, COLLECTION TRAVERSALSThe finishes;
S3.3 takes out the database table name in any i stage and corresponding general respectively from hotspot database table and node probability chained list
Rate value first obtains whether current database table name has existed value in map set when carrying out the assignment operation of map set,
If had existed, it is saved in map set using value value and the sum of node probability value as new value, if do not deposited
Then storing node probability value as value into map set;
After S3.4 hotspot database table and node probability chained list traverse, map set is subjected to descending row according to value value
Sequence, finally, selecting the sum of value value closest but being no more than 80% tables of data set of value value summation as caching
Hot spot data table set.
8. a kind of optimization method of Web cluster cache according to claim 7, which is characterized in that the step S4 is specific
The following steps are included:
S4.1 takes out database table name from the classification standard of history, inquires the database table name of taking-up in hot spot data table set
In whether there is, if it does, remove map set in correspond to key-value key-value pair;If it does not, deleting corresponding point
Class standard, and the set of destination web server is put back into unappropriated server list;
S4.2 takes out database table name from map set, according to the corresponding request amount of the database table of record, refers to for classification request
Fixed sufficient amount of destination web server, the user of the classification according to the service ability of destination web server and in the unit time
Request amount calculates the quantity of destination web server, so that the service ability of server is greater than the service of user's request, wherein
The service ability of destination web server includes the processing capacity and transmission flow of server, using the number of server as set
The destination server field of classification standard is written in the form of circular linked list;
S4.3 sends corresponding single or multiple mesh for each classification standard according to the definition of destination web server field
It marks in web server, with the corresponding database table of caching classification, then sends load balancing service for all classification standards
Device.
9. a kind of optimization method of Web cluster cache according to claim 7, which is characterized in that the step S6 is specific
The following steps are included:
S6.1 load-balanced server receives and stores all classification standards of data analytics server transmission, in each classification
A node is arbitrarily selected in the circular linked list of destination web server field, and pointer is made to be directed toward this node;
For S6.2 when receiving user's request, comparison-of-pair sorting's standard executes following forwarding to determine classification belonging to user's request
Strategy:
S6.2.1 obtains the number that pointer is directed toward node, and user's request is forwarded on the representative destination web server of number,
Later, pointer is moved to next node position;
S6.2.2 increases a field newly in classification standard, for storing the target that cpu occupancy is more than default occupancy threshold
The set of web server number,
The cpu occupancy of destination web server in S6.2.3 monitoring objective web server field, once destination web server
Cpu occupancy reaches default occupancy threshold, the field that the number storage of this destination web server is increased newly to step S6.2.2
In;
The cpu occupancy of destination web server in S6.2.4 monitoring step S6.2.2 field, when the cpu of destination web server is accounted for
When being down to the half for being no more than default occupancy threshold with rate, by the corresponding number of this destination web server again insertable into
Into the circular linked list of destination web server.
10. a kind of optimization method of Web cluster cache according to claim 8, which is characterized in that the step S7 is specific
The following steps are included:
S7.1 destination web server receives the classification request that data analytics server is sent, and determines that corresponding data are requested in classification
Library table, from load table content in database into memory;
User's request of S7.2 destination web server balancing received load server forwarding determines that the classification of user's request needs to visit
The database table asked is consistent with the cache contents in memory, and data information is loaded from memory and returns to user's request.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811481419.8A CN109634746B (en) | 2018-12-05 | 2018-12-05 | Web cluster cache utilization system and optimization method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811481419.8A CN109634746B (en) | 2018-12-05 | 2018-12-05 | Web cluster cache utilization system and optimization method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109634746A true CN109634746A (en) | 2019-04-16 |
CN109634746B CN109634746B (en) | 2022-03-01 |
Family
ID=66071230
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811481419.8A Active CN109634746B (en) | 2018-12-05 | 2018-12-05 | Web cluster cache utilization system and optimization method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109634746B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111049899A (en) * | 2019-12-11 | 2020-04-21 | 贝壳技术有限公司 | kafka message storage system, method, apparatus, and computer-readable storage medium |
CN113901262A (en) * | 2021-09-24 | 2022-01-07 | 北京达佳互联信息技术有限公司 | Method and device for acquiring data to be processed, server and storage medium |
CN114173159A (en) * | 2021-11-23 | 2022-03-11 | 武汉市烽视威科技有限公司 | Hot content prediction method, device, equipment and readable storage medium |
Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1758610A (en) * | 2005-11-11 | 2006-04-12 | 清华大学 | Consumption control method of parallel computing trunking power supply |
US20110066709A1 (en) * | 2009-09-17 | 2011-03-17 | Oracle International Corporation | Web server caching for performance improvement |
CN103338249A (en) * | 2013-06-26 | 2013-10-02 | 优视科技有限公司 | Cache method and device |
CN104065568A (en) * | 2014-07-07 | 2014-09-24 | 电子科技大学 | Web server cluster routing method |
CN104580393A (en) * | 2014-12-18 | 2015-04-29 | 北京蓝汛通信技术有限责任公司 | Method and device for expanding server cluster system and server cluster system |
AU2015203474A1 (en) * | 2012-07-23 | 2015-07-16 | Facebook, Inc. | Structured search queries based on social-graph information |
CN106326012A (en) * | 2016-08-25 | 2017-01-11 | 中国农业银行股份有限公司 | Web application cluster buffer utilization method and system |
CN106936623A (en) * | 2015-12-31 | 2017-07-07 | 五八同城信息技术有限公司 | The management method of distributed cache system and cache cluster |
US20180121304A1 (en) * | 2016-10-27 | 2018-05-03 | Machine Zone, Inc. | System and method for managing a cluster of cache servers (trask) |
-
2018
- 2018-12-05 CN CN201811481419.8A patent/CN109634746B/en active Active
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1758610A (en) * | 2005-11-11 | 2006-04-12 | 清华大学 | Consumption control method of parallel computing trunking power supply |
US20110066709A1 (en) * | 2009-09-17 | 2011-03-17 | Oracle International Corporation | Web server caching for performance improvement |
AU2015203474A1 (en) * | 2012-07-23 | 2015-07-16 | Facebook, Inc. | Structured search queries based on social-graph information |
CN103338249A (en) * | 2013-06-26 | 2013-10-02 | 优视科技有限公司 | Cache method and device |
CN104065568A (en) * | 2014-07-07 | 2014-09-24 | 电子科技大学 | Web server cluster routing method |
CN104580393A (en) * | 2014-12-18 | 2015-04-29 | 北京蓝汛通信技术有限责任公司 | Method and device for expanding server cluster system and server cluster system |
CN106936623A (en) * | 2015-12-31 | 2017-07-07 | 五八同城信息技术有限公司 | The management method of distributed cache system and cache cluster |
CN106326012A (en) * | 2016-08-25 | 2017-01-11 | 中国农业银行股份有限公司 | Web application cluster buffer utilization method and system |
US20180121304A1 (en) * | 2016-10-27 | 2018-05-03 | Machine Zone, Inc. | System and method for managing a cluster of cache servers (trask) |
Non-Patent Citations (6)
Title |
---|
SONGHUAN LI 等: "Redis-based web server cluster session maintaining technology", 《2017 13TH INTERNATIONAL CONFERENCE ON NATURAL COMPUTATION, FUZZY SYSTEMS AND KNOWLEDGE DISCOVERY (ICNC-FSKD)》 * |
ZHONGJU ZHANG等: "Web server load balancing: A queueing analysis", 《EUROPEAN JOURNAL OF OPERATIONAL RESEARCH》 * |
尤国华: "远程控制系统中Web服务器的请求调度算法研究", 《中国博士学位论文全文数据库 信息科技辑》 * |
程艳蕾: "无线传感网中基于概率衰减模型的节点调度研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
胡霖: "基于节点缓存的Web服务器性能优化研究", 《无线互联科技》 * |
陈海涛 等: "SmartCache:基于兴趣的协作式Web缓存", 《计算机科学》 * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111049899A (en) * | 2019-12-11 | 2020-04-21 | 贝壳技术有限公司 | kafka message storage system, method, apparatus, and computer-readable storage medium |
CN111049899B (en) * | 2019-12-11 | 2021-01-05 | 贝壳找房(北京)科技有限公司 | kafka message storage system, method, apparatus, and computer-readable storage medium |
CN113901262A (en) * | 2021-09-24 | 2022-01-07 | 北京达佳互联信息技术有限公司 | Method and device for acquiring data to be processed, server and storage medium |
CN114173159A (en) * | 2021-11-23 | 2022-03-11 | 武汉市烽视威科技有限公司 | Hot content prediction method, device, equipment and readable storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN109634746B (en) | 2022-03-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Bozer et al. | Design and performance models for end-of-aisle order picking systems | |
US9311020B2 (en) | Method and apparatus for automated migration of data among storage centers | |
CN105653591B (en) | A kind of industrial real-time data classification storage and moving method | |
CN107426332B (en) | A kind of load-balancing method and system of web server cluster | |
CN104965861B (en) | A kind of data access monitoring device | |
CN109634746A (en) | A kind of the utilization system and optimization method of web cluster caching | |
CN104702625B (en) | Access request dispatching method and device in a kind of CDN network | |
CN100399301C (en) | Mass storage device and method for dynamically managing a mass storage device | |
US10191930B2 (en) | Priority queuing for updates in a database system | |
CN103345514A (en) | Streamed data processing method in big data environment | |
CN103379156B (en) | Realize the mthods, systems and devices of memory space dynamic equalization | |
CN111124762A (en) | Dynamic copy placing method based on improved particle swarm optimization | |
CN105373347B (en) | A kind of hot spot data identification of storage system and dispatching method and system | |
CN103902260B (en) | A kind of forecasting method of object-based file system | |
CN103150245B (en) | Determine method and the storage controller of the access characteristics of data entity | |
CN104731528B (en) | A kind of construction method and system of cloud computing block storage service | |
CN102129472A (en) | Construction method for high-efficiency hybrid storage structure of semantic-orient search engine | |
CN105657064B (en) | Swift load-balancing method based on dummy node storage optimization | |
CN107122126A (en) | The moving method of data, device and system | |
CN109982104A (en) | The video of mobile awareness prefetches and caching Replacement Decision method in a kind of mobile edge calculations | |
CN109165096A (en) | The caching of web cluster utilizes system and method | |
JP2003005920A (en) | Storage system and data rearranging method and data rearranging program | |
CN104035807B (en) | Metadata cache replacement method of cloud storage system | |
CN106649687A (en) | Method and device for on-line analysis and processing of large data | |
CN106569892A (en) | Resource scheduling method and device |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |