CN109861850A - A method of the stateless cloud workflow load balance scheduling based on SLA - Google Patents
A method of the stateless cloud workflow load balance scheduling based on SLA Download PDFInfo
- Publication number
- CN109861850A CN109861850A CN201910028641.0A CN201910028641A CN109861850A CN 109861850 A CN109861850 A CN 109861850A CN 201910028641 A CN201910028641 A CN 201910028641A CN 109861850 A CN109861850 A CN 109861850A
- Authority
- CN
- China
- Prior art keywords
- request
- queue
- engine
- rtl
- service
- 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
Abstract
The method of the invention discloses a kind of stateless cloud workflow load balance scheduling based on SLA, the present invention is according to different tenants to the needs of its business scenario and procedural model, select different SLA ranks, pass through different SLA ranks, cloud Workflow system provides tenant different request throughput services, and class-of-service is carried out to different process requests, in conjunction with the engine load real time monitoring and procedural model distribution situation on engine of Sharing Memory Realization, the overall memory expense of engine cluster is also reduced while the request wave crest for having cut down engine service, to improve the ability of cloud workflow load balancing under multi-tenant architecture, so that flow services provider is on the basis of meeting parsing execution performance demand of the different tenants to request handling capacity and different flow definitions, service can be provided for more tenants.
Description
Technical field
The present invention relates to workflows and field of cloud computer technology, more particularly to a kind of stateless Yun work based on SLA
Make the method for current load balance dispatching.
Background technique
With the development of distributed computing especially grid, cloud computing as a kind of novel service computation model and
It generates.Cloud computing is a kind of resource delivery and use pattern, refers to obtain by network and applies required resource, including hardware, flat
Platform, software etc., the network for providing resource are referred to as " cloud ".In cloud computing, anything is all service, generally can be divided into three
Level: infrastructure services (IaaS), platform i.e. service (PaaS) and software services (SaaS).
Cloud workflow is serviced as a kind of PaaS grades, is referred to and is provided workflow service with the cloud computing mode that platform services
Distributed system.Compared to conventional operation streaming system, the main advantage of cloud workflow is: cloud workflow provides to be made on demand
With, the mode paid, this mode can effectively reduce the input cost that enterprise uses Workflow Management software according to quantity, reduce starting
Difficulty；Cloud workflow has the advantages that resource high usage and service are high performance, and centralized management mode can make full use of calculating
Power, flexible resource distribution can also cope with the request load of different periods.
Conventional operation stream engine is often based on what stateful scheme was realized, and under cloud environment, in order to preferably play
The flexibility of cloud resource and the reliability for improving cloud Workflow system, the workflow engine realized based on stateless scheme can be more
Meet the demand of cloud workflow；For the cloud Workflow system realized based on stateless workflow engine, on the one hand due to work
The characteristic of stream service itself, process of analysis model and storing and resolving result are still essential, and need to occupy certain calculating
Resource and storage resource；On the other hand, under cloud environment, the operation flow that cloud workflow needs support multi-tenant is executed, load
Scene is more much more complex than conventional operation stream engine.Based on the cloud Workflow system that stateless workflow engine is realized, in face of more
When the request of tenant, multipaths model, multipaths example load, if only considering that the statelessness of service is scheduled,
It will lead to the characteristic for being unable to fully utilize and play workflow service itself, to cannot achieve preferably request load balancing effect
Fruit and user experience.
Currently, the system architecture of existing workflow group system and management structure usually only serve in workflow field
The same user or the same tissue.In cloud service business action mode, flow services provider wishes in same hardware
Process analysis service is provided to more tenants under resource situation, different tenants often takes engine according to its business scenario
The request handling capacity of business has different requirements, and same tenant is often also different to the parsing execution performance of its different flow definition
Requirement, therefore tenant and flow services provider need to sign SLA contract, and system is provided by SLA agreement to tenant corresponding
Service level.
Summary of the invention
The present invention in order to solve based on to tenants different under cloud workflow to varying service level demand and process take
Business provider, to the problem of needs of tenant's quantity is improved, provides a kind of based on the stateless of SLA under same hardware resource
The method of cloud workflow load balance scheduling, while guaranteeing cloud tenant service experience now in fact, optimization cloud workflow request
Load balancing effect and execution performance, so that cloud Workflow system be allowed to provide stream under normal service state for more tenants
Journey analysis service.
To realize aforementioned present invention purpose, the technical solution adopted is as follows: a kind of stateless cloud workflow based on SLA is negative
The method for carrying balance dispatching, when receiving tenant and uploading the request of flow instance corresponding to procedural model, cloud workflow is by process
Example request is dispatched in the stateless workflow engine in cluster, execute the following steps are included:
It is smooth that access layer loads waveform:
S101: access layer receives the request of tenant's flow instance, access layer according to tenant ID or flow instance solicited message from
The warehouse tenant SLA obtains the service request arrival rate RAR index of the tenant and the request of flow instance request is rung
Answer the other RTL of time stage；
S102: according to system limited current algorithm, judge whether tenant's service request rate meets RAR index, if it exceeds RAR
The specified service request rate of index, then direct filter request, and fed back to tenant, it prompts to buy higher RAR rank, otherwise
It performs the next step；
S103: judging that RTL level is other, according to different RTL level not Zhi Hang dispatch layer request is balanced assigns, or obtain current vertical
It executes queue and delays the number of request of queue, usage history load variation historySize, according to the request for delaying queue
Number calculates the scoring that current flow instance request is directed to each delay queue, and the request is put in the highest delay of scoring
In queue；
Dispatch layer request is balanced to assign:
S201: dispatch layer receives the request that queue is immediately performed from access layer, and dispatch layer obtains process from shared drive
The load information set E=[e for each flow engine service that service layer sends1,…,em], ei=(cpui,rami), cpuiTable
Show that flow engine services eiCurrent cpu occupancy, ramiIndicate that flow engine services eiCurrent ram occupancy；
S202: the corresponding flow instance of procedural model of dispatch layer acquisition request from flow instance warehouse is in flow engine
The distribution situation set D=[d of service1,d2,…,dm], di∈ [0,1], works as diIndicate that the procedural model was not run when=0
In eiIt is otherwise on the contrary on engine；
S203: according to distribution situation set D, flow engine service is divided into two groups of E1And E2, E1In house all processes
The engine namely d that model executedi=1；E2House remaining engine；
S204: it is directed to E1And E2Element carry out engine busy degree calculating, respectively obtain E1、E2Busy degree the smallest draw
The service of holding upJudge inequalityIt is whether true, if inequality is set up,
Flow instance request is assigned toOtherwise it is assigned toAnd the distribution situation set in modification process example warehouse；It completes
Flow instance request scheduling；
Wherein, β is to be assigned to the cost parameter of new engine as by flow instance request, can be special according to particular hardware resource
Property is configured.
Preferably, step S101, the service request arrival rate RAR, for measuring flow instance request handling capacity, table
Show the transmissible flow instance number of request of tenant's highest per second；
The RAR index is divided into three-level, defines v0, v1, v2, wherein v0、v1、v2For integer, and there is v0>v1>v2, then three
A rank is described as follows:
RAR 0: refer to service request arrival rate up to equal to v0；
RAR 1: refer to service request arrival rate up to equal to v1；
RAR 2: refer to service request arrival rate up to equal to v2；
The RAR of different stage corresponds to different chargings.
Preferably, step S101, the request response time rank RTL, for measuring different process request treatability
Can, the proposition of RTL is the diversity of the execution time range based on workflow；
The RTL level is not divided into three-level, and defined parameters a, b, t, wherein a, b, t are integer, and has a to be less than b, and t is indicated
Engine handles the time of a flow instance request demand, and indicates the length of a timeslice, can be by taking to flow engine
Business test obtains, then each rank of the RTL is as follows:
RTL 0: flow instance request responds in 1 timeslice, as t；
RTL 1: request is the latest in (a+1) a timeslice response, i.e. (a+1) t；
RLT 2: request is the latest in (b+1) a timeslice response, i.e. (b+1) t.
Further, the system limited current algorithm guarantees the RAR index of tenant using sliding window algorithm；For RTL
Rank is realized that specific mode is as follows using the mode of request caching:
Access layer maintains b+1 for storing the queue of flow instance request, when each queue corresponds to a delay
Long variable, represent the flow instance request in the queue postpones duration, and value is respectively 0t, 1t, 2t ..., bt, in which: t
For timeslice defined in RTL, postponing the queue that duration variable is 0t is to be immediately performed queue, and delay duration variable is 1t,
The queue of 2t ..., bt are delay queue；Access layer is also needed to update delay queue after the time of every 1 timeslice of mistake and be gone through
History load variation, the historic load variable are used to measure the number of request situation of each timeslice in the past；Access layer needs root
According to the flow instance number of requests situation that tenant does not store the RTL level that flow tasks are arranged with current individual queue, by new stream
Journey example request is put into corresponding delay queue.
Still further, step S103, specifically, judging that RTL level is other；
If RTL level is not RTL 0, directly the request of execution dispatch layer is balanced assigns；
If RTL level is not RTL 1, illustrates that delay time is less than or equal to a timeslice, then obtain and be currently immediately performed team
Column and the first two delay the number of request of queue, and collection is combined into N=[n0,n1,…,na]；
If RTL level is not RTL 2, illustrates that delay time is less than or equal to b timeslice, then obtain and be currently immediately performed team
Column and whole number of requests for delaying queue, collection are combined into N=[n0,n1,…,nb]。
Still further, step S103, the calculation method for calculating current request and being directed to the scoring of each delay queue
It is as follows:
In formula: ni∈ N, i indicate position of the current delay queue in set N.
Still further, the access layer needs to update delay queue after the time of every 1 timeslice of mistake and history is negative
Carry variable, comprising the following steps:
H1: the delay duration variable of all delay queues all subtracts 1, and judges to postpone whether duration variable is equal to 0, if so,
Request in queue is all added to and is immediately performed queue, and resets a length of bt when delay；
H2: it is immediately performed queue and needs to run always a thread to judge whether queue has request, and taken according to engine
The rate of business processing request, successively submits to dispatch layer for request and makes requests assignment, and record and submit in each timeslice
The number of request of flow engine；
H3: acquisition is immediately performed queue in the 1 timeslice time in the past and submits to the request that dispatch layer makes requests assignment
Number requestSize updates historic load variable historySize according to following formula:
HistorySize=α * historySize+ (1- α) * requestSize
Wherein: α indicates weight factor, represents the attenuation degree of previous second historySize value.
Still further, the busy degree calculation formula of the engine is as follows:
busynessi=w1*cpui+w2*rami,w1+w2=1
Wherein: w1And w2Two parameters respectively indicate the significance level of both load parameters of cpu and ram, need according to hard
Part resource characteristics are configured.
The present invention was requested in the dispatch layer in balanced assign, and dispatch layer is according to the flow engine service of flow services layer
Load state, and according to the characteristic of stateless workflow engine, realize that the corresponding request of the same procedural model is assigned to minority
Engine on, thus reduce the same procedural model it is multiple parsing with result storage bring calculate and memory consumption.
Beneficial effects of the present invention are as follows: the present invention according to different tenants to the needs of its business scenario and procedural model,
Different SLA ranks is selected, by different SLA ranks, cloud Workflow system provides different request handling capacities to tenant and takes
Business, and class-of-service is carried out to different process requests, in conjunction with the engine load real time monitoring and process mould of Sharing Memory Realization
Type distribution situation on engine also reduces the overall memory of engine cluster while the request wave crest for having cut down engine service
Expense, so that the ability of cloud workflow load balancing under multi-tenant architecture is improved, so that flow services provider is meeting not
With tenant to the parsing execution performance demand of request handling capacity and different flow definitions on the basis of, can mention for more tenants
For service.
Detailed description of the invention
Fig. 1 is cloud workflow core component figure.
Specific embodiment
The present invention will be described in detail with reference to the accompanying drawings and detailed description.
Embodiment 1
As shown in Figure 1, a kind of method of the stateless cloud workflow load balance scheduling based on SLA, this method includes standard
Enter that layer load waveform be smooth, dispatch layer requests balanced assignment process step；Before this two step is specifically introduced, need
It first defines and characterizes different request handling capacities in cloud workflow service SLA and the quantization of different processes request process performance is referred to
Mark indicates that the transmissible process of tenant's highest per second is real wherein request handling capacity is measured using service request arrival rate RAR
Example number of request；Different process request process performances are measured using request response time rank RTL, request response time rank
The it is proposed of RTL is the diversity of the execution time range based on workflow, and time range is from several microseconds to some months etc..
For the present embodiment according to tenant to the needs of its business scenario, RAR index can be divided into three-level, define v0, v1, v2,
In, v0、v1、v2For integer, and there is v0>v1>v2, then three ranks are described as follows:
RAR 0: it is used for high concurrent business scenario, refers to service request arrival rate up to equal to v0；
RAR 1: it is used for general voice and packet data concurrent service scene, refers to service request arrival rate up to equal to v1；
RAR 2: it is used for low voice and packet data concurrent service scene, refers to service request arrival rate up to equal to v2；
The RAR of different stage corresponds to different chargings, has the rank of higher service request arrival rate accordingly also to have more
High charging.
The present embodiment requests the different demands to processing real-time according to different processes, and RTL level can not be divided into three-level, detailed
Before thin statement, defined parameters a, b, t, wherein a, b, t are integer, and have a to be less than b, and t indicates that engine handles a flow instance
The time of request demand also illustrates that the length of a timeslice, can be by testing to obtain to engine service, each rank of SLA
It is presented below:
RTL 0: requesting the higher flow instance of requirement of real-time, and flow instance request responds in 1 timeslice, i.e.,
For t；Mostly automatic flow；
RTL 1: requesting the general flow instance of requirement of real-time, and request is the latest in (a+1) a timeslice response, i.e. (a
+1)t；
RLT 2: requesting the lower flow instance of requirement of real-time, and request is the latest in (b+1) a timeslice response, i.e. (b
+1)t。
Tenant can be directed to the actual use situation of model when uploading procedural model, be task different in procedural model
Different SLA indexs is selected, to obtain different chargings, the higher SLA index of process performance, charging is accordingly higher.
The method of stateless cloud workflow load balance scheduling described in the present embodiment based on SLA, when on reception tenant
When spreading the corresponding flow instance request of journey model, stateless work of the cloud workflow by flow instance request scheduling into cluster
Make in stream engine, execute the following steps are included:
It is smooth that access layer loads waveform:
S101: access layer receives the request of tenant's flow instance, and access layer is according to tenant ID and process solicited message from tenant
The RTL level that the warehouse SLA obtains the RAR index of the tenant and requests for the flow instance is other；
S102: according to time window algorithm, judge whether tenant's service request rate meets RAR index, if it exceeds RAR
The specified service request rate of index, then direct filter request, and fed back to tenant, it prompts to buy higher RAR rank；Otherwise
It performs the next step；
S103: judging that RTL level is other, if RTL level is not RTL 0, directly the request of execution dispatch layer is balanced assigns progress
Scheduling；If RTL level is not RTL 1, illustrate at most postpone a timeslice, then obtains and be currently immediately performed queue with before
Two are delayed the number of request of queue, and collection is combined into N=[n0,n1,…,na]；If RTL level is not RTL 2, illustrate at most postpone b
A timeslice, then obtain and be currently immediately performed queue and whole number of requests for delaying queue, and collection is combined into N=[n0,n1,…,
nb]；
S104: usage history load variation historySize, for the element of each set N, calculated currently using following
Request is directed to the scoring of each delay queue:
In formula, ni∈ N, i indicate position of the current delay queue in set N.
By the scoring of the calculated each delay queue of above step, request is put in the highest delay queue of scoring.
Dispatch layer request is balanced assign the following steps are included:
S201: dispatch layer receives the request that queue is immediately performed from access layer, and dispatch layer obtains process from shared drive
The load information set E=[e for each flow engine service that service layer sends1,…,em], ei=(cpui,rami), cpuiTable
Show that flow engine services eiCurrent cpu occupancy, ramiIndicate that flow engine services eiCurrent ram occupancy；
S202: the corresponding flow instance of procedural model of dispatch layer acquisition request from flow instance warehouse is in flow engine
The distribution situation set D=[d of service1,d2,…,dm], di∈ [0,1], works as diIndicate that the procedural model was not run when=0
In eiIt is otherwise on the contrary on engine；
S203: according to distribution situation set D, flow engine service is divided into two groups of E1And E2, E1In house all processes
The engine namely d that model executedi=1；E2House remaining engine；
S204: it is directed to E1And E2Element carry out engine busy degree calculating, formula is as follows:
busynessi=w1*cpui+w2*rami,w1+w2=1
Wherein: w1And w2Two parameters respectively indicate the significance level of both load parameters of cpu and ram, need according to hard
Part resource characteristics are configured；E is respectively obtained by the busy degree formula of engine1、E2The smallest engine service of busy degree
Judge inequalityIt is whether true；If inequality is set up, will request
It is assigned toOtherwise it is assigned toAnd the distribution situation set in modification process example warehouse；
Wherein: β is to be assigned to the cost parameter of new engine as by flow instance request, can be special according to particular hardware resource
Property is configured；
S205: request scheduling is completed.
The present embodiment was requested in the dispatch layer in balanced assign, and dispatch layer is according to the flow engine service of flow services layer
Load state it is few to realize that the corresponding request of the same procedural model is assigned to and according to the characteristic of stateless workflow engine
On several engines, so that the multiple parsing for reducing the same procedural model is calculated with result storage bring and memory consumption.
System limited current algorithm described in the present embodiment guarantees the RAR index of tenant using sliding window algorithm；For RTL level
Do not realized that specific mode is as follows using the mode of request caching:
Access layer maintains b+1 for storing the queue of flow instance request, when each queue corresponds to a delay
Long variable, represent the flow instance request in the queue postpones duration, and value is respectively 0t, 1t, 2t ..., bt, in which: t
For timeslice defined in RTL, postponing the queue that duration variable is 0t is to be immediately performed queue, and delay duration variable is 1t,
The queue of 2t ..., bt are delay queue；Access layer is also needed to update delay queue after the time of every 1 timeslice of mistake and be gone through
History load variation, the historic load variable are used to measure the number of request situation of each timeslice in the past；Access layer needs root
According to the flow instance number of requests situation that tenant does not store the RTL level that flow tasks are arranged with current individual queue, by new stream
Journey example request is put into corresponding delay queue.
The access layer needs to update delay queue and historic load variable after the time of every 1 timeslice of mistake, including
Following steps:
H1: the delay duration variable of all delay queues all subtracts 1, and judges to postpone whether duration variable is equal to 0, if so,
Request in queue is all added to and is immediately performed queue, and resets a length of bt when delay；
H2: it is immediately performed queue and needs to run always a thread to judge whether queue has request, and taken according to engine
The rate of business processing request, successively submits to dispatch layer for request and makes requests assignment, and record and submit in each timeslice
The number of request of flow engine；
H3: acquisition is immediately performed queue in the 1 timeslice time in the past and submits to the request that dispatch layer makes requests assignment
Number requestSize updates historic load variable historySize according to following formula:
HistorySize=α * historySize+ (1- α) * requestSize
Wherein: α indicates weight factor, represents the rate of decay of previous second historySize value.
The characteristics of the present embodiment is the business scenario based on different tenants to request handling capacity and different flow definition solutions
The different demands for analysing execution performance, using the diversity of the execution time range in workflow, especially partial task and are not required to
The delay for wanting the characteristic of real-time response to make requests, the request load realized to the current limliting of tenant and under current limliting constraint are equal
Weighing apparatus optimization；In conjunction with the characteristic of stateless workflow engine service, the load of the minimum engine number distribution of implementation process model request is equal
Weighing apparatus strategy；Utilize the access of Sharing Memory Realization engine service load information；Method set forth above may be implemented guaranteeing cloud
While tenant's service experience, the load balancing effect and execution performance of optimization cloud workflow request, to allow cloud workflow system
System provides process analysis service under normal service state for more tenants.
Obviously, the above embodiment of the present invention be only to clearly illustrate example of the present invention, and not be pair
The restriction of embodiments of the present invention.Any modification done within the spirit and principles of the present invention and changes equivalent replacement
Into etc., it should all be included in the scope of protection of the claims of the present invention.
Claims (8)
1. a kind of method of the stateless cloud workflow load balance scheduling based on SLA, it is characterised in that: uploaded when receiving tenant
When flow instance corresponding to procedural model is requested, stateless work of the cloud workflow by flow instance request scheduling into cluster
Flow engine in, execute the following steps are included:
It is smooth that access layer loads waveform:
S101: access layer receives the request of tenant's flow instance, and access layer is according to tenant ID or flow instance solicited message from tenant
When the warehouse SLA obtains the service request arrival rate RAR index of the tenant and the request of flow instance request is responded
Between rank RTL；
S102: according to system limited current algorithm, judging whether tenant's service request rate meets RAR index, if it exceeds RAR index
Specified service request rate, then direct filter request, and fed back to tenant, it prompts to buy higher RAR rank, otherwise execute
In next step；
S103: judging that RTL level is other, according to different RTL level not Zhi Hang dispatch layer request is balanced assigns, or obtain and currently hold immediately
Row queue and the number of request for delaying queue, usage history load variation historySize, according to the number of request for delaying queue, meter
Current flow instance request is directed to the scoring of each delay queue, and flow instance request is put in highest prolong that score
In slow queue；
Dispatch layer request is balanced to assign:
S201: dispatch layer receives the request that queue is immediately performed from access layer, and dispatch layer obtains flow services from shared drive
The load information set E=[e for each flow engine service that layer is sent1..., em], ei=(cpui, rami), cpuiIndicate stream
Journey engine service eiCurrent cpu occupancy, ramiIndicate that flow engine services eiCurrent ram occupancy；
S202: the corresponding flow instance of procedural model of dispatch layer acquisition request from flow instance warehouse is in flow engine service
Distribution situation set D=[d1, d2..., dm], di∈ [0,1], works as diIndicate that the procedural model was not run in e when=0i
It is otherwise on the contrary on engine；
S203: according to distribution situation set D, flow engine service is divided into two groups of E1And E2, E1In house all procedural models
The engine namely d executedi=1；E2House remaining engine；
S204: it is directed to E1And E2Element carry out engine busy degree calculating, respectively obtain E1、E2The smallest engine service of busy degreeJudge inequalityIt is whether true, if inequality is set up, by process reality
Example request is assigned toOtherwise it is assigned toAnd the distribution situation set in modification process example warehouse；Complete flow instance
Request scheduling；
Wherein, β is to be assigned to the cost parameter of new engine as by flow instance request, can according to particular hardware resource characteristics into
Row setting.
2. the method for the stateless cloud workflow load balance scheduling according to claim 1 based on SLA, feature exist
In: step S101, the service request arrival rate RAR indicate that tenant is per second most for measuring flow instance request handling capacity
High transmissible flow instance number of request；
The RAR index is divided into three-level, defines v0, v1, v2, wherein v0、v1、v2For integer, and there is v0> v1> v2, then three
Rank is described as follows:
RAR 0: refer to service request arrival rate up to equal to v0；
RAR 1: refer to service request arrival rate up to equal to v1；
RAR 2: refer to service request arrival rate up to equal to v2；
The RAR of different stage corresponds to different chargings.
3. the method for the stateless cloud workflow load balance scheduling according to claim 1 based on SLA, feature exist
In: step S101, the request response time rank RTL, for measuring different process request process performances, the proposition of RTL
It is the diversity of the execution time range based on workflow；
The RTL level is not divided into three-level, defined parameters a, b, t, wherein a, b, t are integer, and have a to be less than b, and t indicates engine
The time of a flow instance request demand is handled, and indicates the length of a timeslice, then each rank of the RTL is as follows:
RTL 0: flow instance request responds in 1 timeslice, as t；
RTL 1: flow instance request is the latest in (a+1) a timeslice response, i.e. (a+1) t；
RLT 2: flow instance request is the latest in (b+1) a timeslice response, i.e. (b+1) t.
4. the method for the stateless cloud workflow load balance scheduling according to claim 3 based on SLA, feature exist
In: the system limited current algorithm guarantees the RAR index of tenant using sliding window algorithm；It Shi Yong not requested for RTL level slow
The mode deposited is realized that specific mode is as follows:
Access layer maintains b+1 for storing the queue of flow instance request, and the corresponding delay duration of each queue becomes
Amount, represent the flow instance request in the queue postpones duration, and value is respectively 0t, 1t, 2t ..., bt, in which: t is
Timeslice defined in RTL, the queue that delay duration variable is 0t are to be immediately performed queue, and delay duration variable is 1t,
The queue of 2t ..., bt is delay queue；Access layer is also needed to update delay queue after the time of every 1 timeslice of mistake and be gone through
History load variation, the historic load variable are used to measure the number of request situation of each timeslice in the past；Access layer needs root
According to the flow instance number of requests situation that tenant does not store the RTL level that flow tasks are arranged with current individual queue, by new stream
Journey example request is put into corresponding delay queue.
5. the method for the stateless cloud workflow load balance scheduling according to claim 4 based on SLA, feature exist
In: step S103, specifically, judging that RTL level is other；
If RTL level is not RTL 0, directly the request of execution dispatch layer is balanced assigns；
If RTL level is not RTL 1, illustrate delay time be less than or equal to a timeslice, then obtain currently be immediately performed queue with
The first two delays the number of request of queue, and collection is combined into N=[n0, n1..., na]；
If RTL level is not RTL 2, illustrate delay time be less than or equal to b timeslice, then obtain currently be immediately performed queue with
Whole number of requests for delaying queue, collection are combined into N=[n0, n1..., nb]。
6. the method for the stateless cloud workflow load balance scheduling according to claim 5 based on SLA, feature exist
In: step S103, the calculation method for calculating the scoring that current request is directed to each delay queue are as follows:
In formula, ni∈ N, i indicate position of the current delay queue in set N.
7. the method for the stateless cloud workflow load balance scheduling according to claim 6 based on SLA, feature exist
In: the access layer needs to update delay queue and historic load variable, including following step after the time of every 1 timeslice of mistake
It is rapid:
H1: the delay duration variable of all delay queues all subtracts 1, and judges to postpone whether duration variable is equal to 0, if so, will prolong
Request in slow queue, which is all added to, is immediately performed queue, and resets a length of bt when delay；
H2: being immediately performed queue and need to run always a thread to judge whether queue has request, and according to engine service at
Request is successively submitted to dispatch layer and makes requests assignment, and recorded in each timeslice and submit to process by the rate for managing request
The number of request of engine；
H3: acquisition is immediately performed queue in the 1 timeslice time in the past and submits to the number of request that dispatch layer makes requests assignment
RequestSize updates historic load variable historySize according to following formula:
HistorySize=α * historySize+ (1- α) * requestSize
Wherein: α indicates weight factor, represents the attenuation degree of previous second historySize value.
8. the method for the stateless cloud workflow load balance scheduling according to claim 7 based on SLA, feature exist
In: the busy degree calculation formula of the engine is as follows:
busynessi=w1*cpui+w2*rami, w1+w2=1
Wherein: w1And w2Two parameters respectively indicate the significance level of both load parameters of cpu and ram, need to be provided according to hardware
Source characteristic is configured.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910028641.0A CN109861850B (en) | 2019-01-11 | 2019-01-11 | SLA-based stateless cloud workflow load balancing scheduling method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910028641.0A CN109861850B (en) | 2019-01-11 | 2019-01-11 | SLA-based stateless cloud workflow load balancing scheduling method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109861850A true CN109861850A (en) | 2019-06-07 |
| CN109861850B CN109861850B (en) | 2021-04-02 |
Family
ID=66894503
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910028641.0A Active CN109861850B (en) | 2019-01-11 | 2019-01-11 | SLA-based stateless cloud workflow load balancing scheduling method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109861850B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110659463A (en) * | 2019-08-23 | 2020-01-07 | 苏州浪潮智能科技有限公司 | Distributed operation method and device of stateless system |
| CN110737485A (en) * | 2019-09-29 | 2020-01-31 | 武汉海昌信息技术有限公司 | workflow configuration system and method based on cloud architecture |
| CN110941681A (en) * | 2019-12-11 | 2020-03-31 | 南方电网数字电网研究院有限公司 | Multi-tenant data processing system, method and device of power system |
| WO2021099903A1 (en) * | 2019-11-18 | 2021-05-27 | International Business Machines Corporation | Multi-tenant extract transform load resource sharing |
| CN115237573A (en) * | 2022-08-05 | 2022-10-25 | 中国铁塔股份有限公司 | Data processing method and device, electronic equipment and readable storage medium |
| US11841871B2 (en) | 2021-06-29 | 2023-12-12 | International Business Machines Corporation | Managing extract, transform and load systems |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101105843A (en) * | 2006-07-14 | 2008-01-16 | 上海移动通信有限责任公司 | Electronic complaint processing system and method in communication field |
| CN101694709A (en) * | 2009-09-27 | 2010-04-14 | 华中科技大学 | Service-oriented distributed work flow management system |
| US20110161952A1 (en) * | 2009-12-31 | 2011-06-30 | International Business Machines Corporation | Porting Virtual Images Between Platforms |
| US20140358624A1 (en) * | 2011-07-19 | 2014-12-04 | HCL America Inc. | Method and apparatus for sla profiling in process model implementation |
| CN104778076A (en) * | 2015-04-27 | 2015-07-15 | 东南大学 | Scheduling method for cloud service workflow |
| CN106095569A (en) * | 2016-06-01 | 2016-11-09 | 中山大学 | A kind of cloud workflow engine scheduling of resource based on SLA and control method |
| CN108665157A (en) * | 2018-05-02 | 2018-10-16 | 中山大学 | A method of realizing cloud Workflow system flow instance balance dispatching |
-
2019
- 2019-01-11 CN CN201910028641.0A patent/CN109861850B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101105843A (en) * | 2006-07-14 | 2008-01-16 | 上海移动通信有限责任公司 | Electronic complaint processing system and method in communication field |
| CN101694709A (en) * | 2009-09-27 | 2010-04-14 | 华中科技大学 | Service-oriented distributed work flow management system |
| US20110161952A1 (en) * | 2009-12-31 | 2011-06-30 | International Business Machines Corporation | Porting Virtual Images Between Platforms |
| US20140358624A1 (en) * | 2011-07-19 | 2014-12-04 | HCL America Inc. | Method and apparatus for sla profiling in process model implementation |
| CN104778076A (en) * | 2015-04-27 | 2015-07-15 | 东南大学 | Scheduling method for cloud service workflow |
| CN106095569A (en) * | 2016-06-01 | 2016-11-09 | 中山大学 | A kind of cloud workflow engine scheduling of resource based on SLA and control method |
| CN108665157A (en) * | 2018-05-02 | 2018-10-16 | 中山大学 | A method of realizing cloud Workflow system flow instance balance dispatching |
Non-Patent Citations (2)
| Title |
|---|
| YINJUAN ZHAGN ET AL: "An Improved Adaptive Workflow Scheduling Algorithm in Cloud Environments", 《IEEE》 * |
| 刘雨潇等: "多目标云工作流调度的协同进化多群体优化", 《计算机工程与设计》 * |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110659463A (en) * | 2019-08-23 | 2020-01-07 | 苏州浪潮智能科技有限公司 | Distributed operation method and device of stateless system |
| CN110659463B (en) * | 2019-08-23 | 2021-11-12 | 苏州浪潮智能科技有限公司 | Distributed operation method and device of stateless system |
| CN110737485A (en) * | 2019-09-29 | 2020-01-31 | 武汉海昌信息技术有限公司 | workflow configuration system and method based on cloud architecture |
| WO2021099903A1 (en) * | 2019-11-18 | 2021-05-27 | International Business Machines Corporation | Multi-tenant extract transform load resource sharing |
| GB2603098A (en) * | 2019-11-18 | 2022-07-27 | Ibm | Multi-tenant extract transform load resource sharing |
| GB2603098B (en) * | 2019-11-18 | 2022-12-14 | Ibm | Multi-tenant extract transform load resource sharing |
| CN110941681A (en) * | 2019-12-11 | 2020-03-31 | 南方电网数字电网研究院有限公司 | Multi-tenant data processing system, method and device of power system |
| US11841871B2 (en) | 2021-06-29 | 2023-12-12 | International Business Machines Corporation | Managing extract, transform and load systems |
| CN115237573A (en) * | 2022-08-05 | 2022-10-25 | 中国铁塔股份有限公司 | Data processing method and device, electronic equipment and readable storage medium |
| CN115237573B (en) * | 2022-08-05 | 2023-08-18 | 中国铁塔股份有限公司 | Data processing method, device, electronic equipment and readable storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109861850B (en) | 2021-04-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109861850A (en) | A method of the stateless cloud workflow load balance scheduling based on SLA | |
| CN103605567B (en) | Cloud computing task scheduling method facing real-time demand change | |
| Ge et al. | GA-based task scheduler for the cloud computing systems | |
| Salot | A survey of various scheduling algorithm in cloud computing environment | |
| CN108845874B (en) | Dynamic resource allocation method and server | |
| CN110297699B (en) | Scheduling method, scheduler, storage medium and system | |
| US8843929B1 (en) | Scheduling in computer clusters | |
| CN103927225A (en) | Multi-core framework Internet information processing and optimizing method | |
| Bi et al. | SLA-based optimisation of virtualised resource for multi-tier web applications in cloud data centres | |
| CN106233276A (en) | The coordination access control of network-accessible block storage device | |
| US8219716B2 (en) | Methods for accounting seek time of disk accesses | |
| CN108762687B (en) | I/O service quality control method, device, equipment and storage medium | |
| CN107038071A (en) | A kind of flexible dispatching algorithm of Storm tasks predicted based on data flow | |
| CN109491761A (en) | Cloud computing multiple target method for scheduling task based on EDA-GA hybrid algorithm | |
| Xin et al. | A load balance oriented cost efficient scheduling method for parallel tasks | |
| Ashouraei et al. | A new SLA-aware load balancing method in the cloud using an improved parallel task scheduling algorithm | |
| CN106095581B (en) | Network storage virtualization scheduling method under private cloud condition | |
| CN108563495A (en) | The cloud resource queue graded dispatching system and method for data center's total management system | |
| CN102917014A (en) | Resource scheduling method and device | |
| Shu-Jun et al. | Optimization and research of hadoop platform based on fifo scheduler | |
| CN117112199A (en) | Multi-tenant resource scheduling method, device and storage medium | |
| CN112363827A (en) | Multi-resource index Kubernetes scheduling method based on delay factors | |
| CN113656150A (en) | Deep learning computing power virtualization system | |
| CN110096364B (en) | Cloud server computing set control method and system | |
| Han et al. | Elastic allocator: An adaptive task scheduler for streaming query in the cloud |
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 |