CN104899098B - A kind of vCPU dispatching method based on shared I/O virtualized environment - Google Patents
A kind of vCPU dispatching method based on shared I/O virtualized environment Download PDFInfo
- Publication number
- CN104899098B CN104899098B CN201510233215.2A CN201510233215A CN104899098B CN 104899098 B CN104899098 B CN 104899098B CN 201510233215 A CN201510233215 A CN 201510233215A CN 104899098 B CN104899098 B CN 104899098B
- Authority
- CN
- China
- Prior art keywords
- virtual machine
- request
- ring
- vcpu
- shared
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Landscapes
- Computer And Data Communications (AREA)
- Data Exchanges In Wide-Area Networks (AREA)
Abstract
The present invention provides a kind of vCPU dispatching method based on shared I/O virtualized environment, including the following steps: a) requests ring feature according to the shared I/O of real-time monitoring, identifies and be lined up responsive type virtual machine；B) the vCPU dispatching priority for the queuing responsive type virtual machine that step a) is identified is improved.The present invention can alleviate because vCPU queuing time it is elongated caused by virtual machine network performance decline, so as to improve the network performance of virtual machine.
Description
Technical field
The present invention relates to field of cloud calculation, specifically, the present invention relates to a kind of based on shared I/O virtualized environment
VCPU dispatching method.
Background technique
In recent years, cloud computing technology has obtained quick development.The companies such as Amazon, Google, Alibaba all take
Publicly-owned cloud computing platform is built, to provide public cloud computing service for user.At the same time, as building cloud computing platform
Major way, virtualization technology obtains constantly mature and perfect.Cloud computing service provider is by virtualization technology come to object
The resource of reason is virtualized, to realize the shared of resource and be isolated.To the void of computing resource, Internet resources and storage resource
Quasi-ization is to construct the basis of cloud computing platform.It wherein, is core to the virtualization of computing resource.Hypervisor is to realize to calculate
The main method of resource virtualizing, specific implementation includes: Xen, KVM, Hyper-v etc..Wherein, Xen is the open source of mainstream
Hypervisor is widely used by domestic and international cloud computing service provider.
VCPU is that virtual platform (such as Xen) is the virtual cpu that virtual machine provides.Only when the vCPU of virtual machine is adjusted
When spending on physical cpu, virtual machine could be run.VCPU scheduling refers to that the vCPU for being subordinated to virtual machine is dispatched to physics by Xen
The process run on CPU.Each physical cpu core has a corresponding vCPU scheduling queue, and vCPU task exists
Wait in line to dispatch in queue.On the other hand, Xen system belongs to the virtualized environment of shared I/O a kind of, and Fig. 1 shows Xen void
The networking model schematic diagram of quasi- machine, as shown in Figure 1, the network-driven model for using front and back end to separate in Xen system comes to be virtual
Machine provides network service.Under this model, the network device driver of virtual machine is divided into front-end driven and rear end drives two parts.
Front-end driven is located at virtual machine internal, and rear end driving is located in Domain 0 (zero domain).Front and back ends device drives are by sharing I/
The mode exchange of control information of O ring exchanges the data actually received and dispatched by shared data buffer area (i.e. shared drive), by altogether
Enjoy the event channel exchange asynchronous event message in memory.Rear end drives corresponding virtual interface equipment to pass through virtual bridge and object
The reason network equipment is connected, and provides the connectivity with physical network for virtual machine.In addition, xenbr0 indicates Domain 0 in Fig. 1
Interior virtual bridge equipment, eth0 indicate that the physical network device interface in Domain 0, vif indicate that virtual machine virtual network is set
Standby rear end equipment, NIC indicate that physics network card equipment, vm indicate that virtual machine, request indicate reception or the hair of front-end driven
Request is sent, response indicates response of the rear end driving to receiving or sending request.
Under this configuration, virtual machine needs to execute by the operation of its vCPU to the processing of network inter-related task, and
The vCPU scheduling scheme of existing Xen does not ensure that each virtual machine vCPU is being run at any time, this often leads
Cause the network inter-related task with time sensitivity that cannot be handled in time.Particularly, it is loaded in host physical cpu higher
In the case where, vCPU dispatches improper meeting so that the internetworkings such as more serious handling capacity reduction and improper delay occurs in virtual machine
The case where capable of reducing.
On the other hand, in order to optimize allocation of resources, save operation cost, cloud computing service provider is at operation data center
When can adjust distribution of the virtual machine in data center every now and then.For example, by the virtual machine (vm) migration of distribution relatively dispersion to virtual machine
On the hosts server of comparatively dense, and by the server suspend mode vacated to save electric power.Such case can also promote host
Physical cpu load rise, to aggravate negative effect of the vCPU scheduling to virtual machine network performance.
Therefore, the virtual machine network under the virtualized environment based on shared I/O can currently be improved there is an urgent need to one kind
The solution of energy.
Summary of the invention
The task of the present invention is provide a kind of vCPU scheduling solution that can overcome above-mentioned prior art defect.
The present invention provides a kind of vCPU dispatching methods based on shared I/O virtualized environment, including the following steps:
A) ring feature is requested according to the shared I/O of real-time monitoring, identifies and is lined up responsive type virtual machine；It is empty to be lined up responsive type
Quasi- machine, which refers to, to be easier to cause the void of choking phenomenon because network inter-related task cannot be handled in time in all virtual machines
Quasi- machine；
B) the vCPU dispatching priority for the queuing responsive type virtual machine that step a) is identified is improved, it is quick to be preferably queuing
Sense type virtual machine distributes computing resource.
Wherein, in the step a), the queuing responsive type virtual machine includes the upward queuing responsive type of data receiver
Virtual machine.
Wherein, in the step a), ring is requested to overflow the upward row of frequency identification data receiver according to the I/O of virtual machine
Team's responsive type virtual machine.
Wherein, in the step a), upward according to the I/O request queue of virtual machine overload frequency identification data receiver
It is lined up responsive type virtual machine.
Wherein, in the step a), ring is requested to overflow frequency and I/O request queue overload frequency according to the I/O of virtual machine
Identify the upward queuing responsive type virtual machine of data receiver.
Wherein, in the step a), the process of the upward queuing responsive type virtual machine of identification data receiver includes following
Sub-step:
A1 the I/O for) monitoring virtual machine requests ring characteristic；
A2) ring characteristic statistics I/O request ring is requested to overflow frequency and I/O request queue overload frequency according to I/O；
A3) exceed preset threshold value when I/O request ring overflows any one of frequency and I/O request queue overload frequency
When, judge the virtual machine queuing responsive type virtual machine upward for data receiver.
Wherein, the step a1) in, the I/O request ring characteristic includes: that I/O needed for current received data packet is asked
Number and I/O is asked to request the I/O request space of the corresponding shared drive current residual of ring, the step a2) in, according to working as
I/O needed for preceding received data packet requests number and the I/O of the corresponding shared drive current residual of I/O request ring to request space,
It counts I/O request ring and overflows frequency.
Wherein, the step a1) in, the I/O requests ring characteristic further include: is not located in current I/O request ring
Manage the length of I/O request queue, the step a2) in, the length of not processed I/O request queue in ring is requested according to current I/O
Frequency is overloaded in degree, statistics I/O request queue.
Wherein, the step a1) in, the I/O based on preset monitoring cycle monitoring virtual machine requests ring characteristic；
Affiliated step a2) in, based on the preset recognition cycle comprising several monitoring cycles, count in the recognition cycle
The corresponding I/O request ring of virtual machine overflows frequency and frequency is overloaded in I/O request queue.
Wherein, in the step a), the queuing responsive type virtual machine includes the upward queuing responsive type of data sender
Virtual machine.
Wherein, in the step a), according to the vCPU overload situations of virtual machine and I/O request queue length, data are identified
Queuing responsive type virtual machine on sending direction.
Wherein, in the step a), the process of the upward queuing responsive type virtual machine of identification data sender includes following
Sub-step:
A11) in each monitoring cycle, differentiate whether vCPU overloads in the monitoring cycle, and obtain in the monitoring cycle
I/O request queue length characteristic；
A12 the queuing responsive type virtual machine on biometric data sending direction) is shaken according to request queue length, it is described
The concussion of request queue length is characterized in: the sum of I/O request queue length characteristic in the monitoring cycle of vCPU overload sum_rq_
The ratio of pattern and the monitoring cycle number count_cpu_ol of CPU overload in recognition cycle.
Wherein, the process of the upward queuing responsive type virtual machine of the identification data sender further include: described in execution
Step a12) before, according to the size of the monitoring cycle number count_cpu_ol of CPU overload in recognition cycle, screen out in advance non-
It is lined up responsive type virtual machine.
Wherein, the step a12) further include: when request queue length concussion feature is near preset threshold
When in numerical intervals, differentiate that current virtual machine is to be lined up responsive type virtual machine.
Compared with prior art, the present invention has following technical effect:
The present invention can alleviate because vCPU queuing time it is elongated caused by virtual machine network performance decline, so as to improve void
The network performance of quasi- machine.
Detailed description of the invention
Hereinafter, carrying out the embodiment that the present invention will be described in detail in conjunction with attached drawing, in which:
Fig. 1 shows the networking model schematic diagram of Xen virtual machine；
Fig. 2 shows the upward queuing responsive type virtual machine recognition methods of the data receiver in one embodiment of the invention
Flow chart；
Fig. 3 shows the upward queuing responsive type virtual machine recognition methods of the data sender in one embodiment of the invention
Flow chart；
Fig. 4 shows the scene figure of a virtual machine network responsive type identification and vCPU scheduling in the present invention；
Fig. 5 is shown under test scene based on some virtual machine in one embodiment of the invention respectively in unloaded, fortune
I/O when row UDP flow amount and operation TCP flow amount requests ring feature related data.
Specific embodiment
Inventor is under the virtualized environment based on shared I/O, and handling capacity reduces and the network performances drops such as improper delay
Low situation has made intensive studies.Under the virtualized environment of shared I/O, on the one hand, the virtual i/o of each virtual machine needs
It is scheduled in shared I/O ring, to wait physics I/O resource allocation to the virtual i/o, on the other hand, when physics I/O resource point
When the dispensing virtual i/o, it is also necessary to which the vCPU of respective virtual machine executes corresponding I/O inter-related task, if vCPU is not at this time
Actual physical computing resources are assigned, i.e., the vCPU is currently not in operating status, then the net with time sensitivity
Network inter-related task cannot be handled in time, and share be also possible to will appear obstruction, subsequent virtual I/ in I/O ring at this time
O waits in line the distribution of physics I/O, so that the network performance of virtualization system entirety be caused to decline.And on the other hand, in order to
It optimizes allocation of resources, save operation cost, often adjustment is empty every now and then at operation data center for cloud computing service provider
Quasi- distribution of the machine in data center.For example, the host of virtual machine (vm) migration to the virtual machine comparatively dense of distribution relatively dispersion is taken
It is engaged on device, and by the server suspend mode vacated to save electric power, this can promote host physical cpu load rise, to aggravate
VCPU dispatches the negative effect to virtual machine network performance.
Based on above-mentioned analysis, inventors herein propose it is a kind of by identify may cause virtualization system network performance decline
It is lined up responsive type virtual machine, and then based on recognition result adjustment to the vCPU scheduling scheme of such virtual machine, to guarantee these void
The network inter-related task of quasi- machine can handle in time, so that alleviating share in I/O improper waits in line phenomenon, into
And the handling capacity of virtualization system is improved, avoid the improper delay of virtualization system.
The present invention will be further described with reference to the accompanying drawings and examples.
According to one embodiment of present invention, a kind of vCPU tune based on the identification of above-mentioned queuing responsive type virtual machine is provided
Degree method, the dispatching method include the following steps:
Step a: first according to the shared I/O of real-time monitoring request ring feature, identify respectively data receiver upwards and
The upward queuing responsive type virtual machine of data sender.Being lined up responsive type virtual machine is exactly that those in Xen platform are easy because cannot
That class virtual machine for handling network inter-related task in time and whole system being caused to be blocked.It hereinafter, will be in conjunction with attached drawing and concrete example
Son describes the recognition methods of queuing responsive type virtual machine of the data receiver upwards and on data sending direction respectively.
Step b: scheduling process is executed by improving the Credit Scheduler of Xen current default.In the present embodiment,
The dispatching priority for being lined up the vCPU of responsive type virtual machine is improved, the front end for making this part vCPU preferentially be come queue is (existing
Credit Scheduler be that each physical cpu safeguards a queue, vCPU to be scheduled is placed in fifo fashion
It is waited in queue scheduled).In this way, the vCPU for being lined up responsive type virtual machine is more frequently dispatched, to improve host
When cpu load is higher, the network performance of virtual machine.
Fig. 2 shows in one embodiment of the invention, the upward queuing responsive type virtual machine recognition methods of data receiver
Flow chart, the upward queuing responsive type virtual machine recognition methods of the data receiver specifically includes the following steps:
Step 110: setting up monitoring point at the driving of Xen virtual machine.In the present embodiment, in Xen virtual machine rear end equipment
Driving at set up monitoring point, this be it is a kind of be easier to realize solution.It should be noted that in another embodiment,
It can also set up monitoring point at the front-end driven of Xen virtual machine, it in this way can also request ring property to Xen virtual platform
It is monitored, but since front-end driven is in VME operating system, so in the embodiment, it is also necessary to correspondingly modify
The shared I/O of Xen requests ring mechanism, is set with ensuring that information detected by headend equipment monitoring point can be passed to rear end
It is standby.
Step 120: determining that monitoring cycle, recognition cycle, I/O request ring spilling discrimination threshold, request queue overload are critical
Value overloads discrimination threshold with request queue.Wherein, monitoring cycle refers to the basic cycle being monitored to request ring property, identification
Period refers to the period that the request ring property monitored according to several monitoring cycles identifies queuing responsive type virtual machine.
I/O request ring overflows the total amount of data for referring to the data packet currently just handled beyond shared drive associated by I/O request ring
Situation, request queue overload refer to the length of request queue beyond queue length preset value (i.e. request queue overload critical value)
Situation, this queue length preset value can according to need determination.
In one example, monitoring point is set up at the rear end equipment driving of virtual machine in Domain 0.If requesting I/O
Whether the monitoring cycle m_period that ring property is monitored is 10ms, be the identification for being lined up responsive type and being judged to virtual machine
Period r_period is 100ms, and it is 0.2 that I/O, which requests ring to overflow discrimination threshold ro_thresh, and critical value qo_ is overloaded in request queue
Limit is 128, and it is 0.8 that discrimination threshold qo_thresh is overloaded in request queue.
Step 130: initialization monitoring counter mc sets mc=0.It initializes I/O request ring and overflows number ro_count=
0, frequency of occurrence qo_count=0 is overloaded in initialization I/O request queue.
Step 140: monitoring point is according to monitoring cycle timing acquisition and records reception data direction I/O request ring characteristic
(otherwise referred to as I/O requests ring feature related data).In a preferred embodiment, step 140 includes that 141 to 142 this is several
Sub-steps:
Step 141: monitoring point according to it is current receive data grouping (i.e. received data packet) needed for I/O request number and current
Residue I/O requests the relationship in space to determine that I/O request ring overflows situation in ring, and records result.If being remained in current I/O request ring
Remaining I/O request insufficient space requests number to accommodate I/O needed for current reception data grouping, then is considered as I/O request ring and overflows, and
Set ro_count=ro_count+1.Ro_count indicates the number that I/O request ring overflows.
Step 142: monitoring point not processed I/O request queue length and request queue in ring being requested according to current I/O
The relationship for carrying critical value determines request queue overload situations, and records result.If current request queue length is greater than qo_limit,
Then it is considered as request queue overload, juxtaposition qo_count=qo_count+1.Qo_count indicates the number of request queue overload.It needs
Illustrate, in this step, what is differentiated is in the currently monitored period, and whether request queue overloads, and what qo_count was represented is
The number of the monitoring cycle of request queue overload occurs in entire recognition cycle.
After completing step 140, step 150 is continued to execute.
Step 150: monitoring point adds up to monitoring counter mc, that is, sets mc=mc+1.
Step 160: if the value of monitoring counter is less than the number r_period/m_ of monitoring cycle contained by recognition cycle
Period is then returned and is re-execute the steps 130.Otherwise, step 170 is executed.
Step 170: the probability that I/O request ring overflows occurs for monitoring cycle in statistics recognition cycle.I/O requests ring to overflow
Probability refers to that in a recognition cycle, overflow probability occurs for single monitoring cycle, for example, (10 in a recognition cycle
Monitoring cycle), I/O requests ring to overflow generation 2 times altogether, then it is 0.2 that overflow probability, which occurs, for single monitoring cycle.If the prison counted
Surveying the probability that period generation I/O request ring overflows is more than that I/O request ring overflows discrimination threshold ro_thresh, then knows virtual machine
Responsive type Wei be lined up and (correspond to step 190), return step 130 in Fig. 2；Otherwise, step 180 is further executed.
Step 180: the probability of request queue overload occurs for monitoring cycle in statistics recognition cycle.If the probability is more than request
Virtual machine is then identified as being lined up responsive type (corresponding to step 190) in Fig. 2 by queue overload discrimination threshold qo_thresh.It is no
Then, then virtual machine is identified as non-queuing responsive type and (corresponds to step 191) in Fig. 2.Return step 130.
The narration that data receiver direction is lined up the recognition methods of responsive type virtual machine terminates.
It should be noted that the queuing responsive type virtual machine decision rule of above-mentioned steps 170 and 180 meets any one,
I.e. it is believed that the virtual machine is to be lined up responsive type virtual machine.In another embodiment, step 180 is first carried out, such as recognition cycle
The probability that request queue overload occurs for interior counted monitoring cycle is more than that discrimination threshold qo_thresh is overloaded in request queue, then
Directly virtual machine is identified as to be lined up responsive type, it is no to then follow the steps 170, if I/O request occurs for the monitoring cycle counted
The probability that ring overflows is more than that I/O request ring overflows discrimination threshold ro_thresh, then is identified as virtual machine being lined up responsive type, no
Virtual machine is then identified as non-queuing responsive type.
Further, the identifying schemes of above-mentioned queuing responsive type virtual machine are requested by the I/O of virtual machine
Ring overflows frequency and the upward queuing responsive type virtual machine of frequency identification data receiver is overloaded in I/O request queue.Aforementioned implementation
In example, it is one of the characteristic manner that frequency is overflowed in I/O request queue, identification that the probability that I/O request ring overflows, which occurs, for monitoring cycle
One of the characteristic manner that the probability of request queue overload is then I/O request queue overload frequency occurs for monitoring cycle in period.Reason
By upper theory, the I/O request ring for relying solely on virtual machine overflows frequency or I/O request queue overload frequency also may recognize that one
Point it is lined up responsive type virtual machine, also has the effect of the network performance for the virtual machine that has some improvement based on this scheduling scheme.
On the other hand, Fig. 3 shows the upward queuing responsive type virtual machine of data sender in one embodiment of the invention
The flow chart of recognition methods, the upward queuing responsive type virtual machine recognition methods of the data sender include the following steps:
Step 210: setting up monitoring point at the driving of virtual machine (such as at the driving of rear end equipment).
Step 220: determining monitoring cycle, recognition cycle, CPU overload threshold value and request queue length characteristic threshold value.If right
Whether the monitoring cycle m_period that I/O request ring property is monitored is 10ms, is to be lined up responsive type to sentence to virtual machine
Disconnected recognition cycle r_period is 100ms, CPU overload threshold value is 0.95, request queue length concussion characteristic interval be [0.4,
0.6]。
Step 230: initialization monitoring counter mc sets mc=0.
Step 240: monitoring point is according to monitoring cycle timing acquisition and records the vCPU loading condition of virtual machine to be seen.?
In specific implementation, the API that can be provided by monitoring point by Xen Hypervisor is obtained in a nearest detection cycle wait supervise
Survey the runing time t of virtual machine.Setting cpu_load is t/m_period.
On the other hand, monitoring point according to monitoring cycle timing acquisition and records transmission data direction I/O request queue length
Characteristic information rq_pattern.In a preferred embodiment, monitoring point monitors not processed I/O in current I/O request ring and asks
Seek queue length.If request queue length is 0, setting request queue length characteristic is 0, that is, setting rq_pattern is 0；Otherwise,
Setting request queue length characteristic is 1, that is, setting rq_pattern is 1.Monitoring result is recorded, that is, records monitoring cycle and in the period
Rq_pattern value.
Step 250: add up monitoring counter mc for monitoring point.
Step 260: if mc is less than the number of monitoring cycle contained by recognition cycle, otherwise return step 240 executes step
Rapid 270.
Step 270: cpu load is greater than the number count_cpu_ol of CPU overload threshold value in statistics recognition cycle.If CPU
The number of overload is less than the half of monitoring cycle number in recognition cycle, then it is (right virtual machine to be identified as non-queuing responsive type
It should be in the step 291) in Fig. 3 and return step 230；Otherwise, step 280 is continued to execute.
Step 280: request queue length characteristic information in the detection cycle of CPU overload is added, it is long according to request queue
Spend the ratio of the monitoring cycle number count_cpu_ol of CPU overload in the sum of characteristic information sum_rq_pattern and recognition cycle
Value, which judges whether it is, is lined up responsive type virtual machine, wherein if the sum of request queue length characteristic information and CPU in recognition cycle
The ratio sum_rq_pattern/count_cpu_ol of the monitoring cycle number of overload is in preset request queue length characteristic
Within section, then virtual machine is identified as queuing responsive type and (corresponding to the step 290) in Fig. 3, otherwise, it is quick to be identified as non-queuing
Sense type (corresponding to the step 291) in Fig. 3, and return step 230.
It should be noted that the one kind proposed using inventor is special by the concussion of request queue length in step 280
It levies to identify whether as the thought of queuing responsive type virtual machine.The characteristic interval of request queue length concussion herein is one with 0.5
Centered on small range section.Numerical value (i.e. the sum of the request queue length characteristic information sum_rq_ found out in step 270
The ratio of pattern and the monitoring cycle number count_cpu_ol of CPU overload in recognition cycle) closer to 0.5, indicate request
Queue length concussion is more violent, and then thinks that corresponding virtual machine is lined up sensibility and is more obvious.And too small or mistake is not considered mostly
It is to be lined up responsive type virtual machine.
Data sender terminates to the narration for being lined up the recognition methods of responsive type virtual machine.
After identifying queuing responsive type virtual machine, use the information to distinguish scheduling to the vCPU of virtual machine,
So that the vCPU for being lined up responsive type virtual machine is more frequently dispatched on physical cpu and is run, that is, whole system can be improved
Network performance.
The scheduling scheme of the present embodiment can significantly improve virtual machine network performance when host cpu load is higher.Fig. 4
The realization scene based on the vCPU dispatching method for being lined up the identification of responsive type virtual machine is shown, the zero of host in the scene
Dispose the rear end equipment of virtual bridge and the virtual machines virtual network device such as access vm01~vm06 in domain, virtual bridge again with
The physical network hardware (such as network interface card) of Xen platform connects, to keep the virtual machines such as vm01~vm06 outer by the access of Xen platform
Boundary's network.Under the scene, using previous embodiment based on the scheduling scheme for being lined up the identification of responsive type virtual machine, adjusted
Degree.Fig. 5 is shown under test scene based on some virtual machine in one embodiment of the invention respectively in unloaded (idle), fortune
I/O when row UDP flow amount and operation TCP flow amount requests ring feature related data, and I/O request ring feature related data includes:
(each of domid, Xen virtualization system virtual machine has a Domain ID to Domain ID.Domain ID is 0
Domain is a privilege Domain；ID non-zero Domain is also DomainU, i.e. General Virtual Machine), overflow number
(Overflow), request queue length (RQlen) etc..In actual test, the network throughput of virtual machine system is averagely improved
31%, delay decline 27%.As can be seen that the scheduling scheme can alleviate because vCPU queuing time it is elongated caused by virtual machine
Network performance decline, improves the network performance of virtual machine significantly.
Finally it should be noted that above embodiments are only to describe technical solution of the present invention rather than to this technology method
It is limited, the present invention can above extend to other modifications, variation, application and embodiment, and therefore, it is considered that institute in application
There are such modification, variation, application, embodiment all within the scope of spirit or teaching of the invention.
Claims (10)
1. a kind of vCPU dispatching method based on shared I/O virtualized environment, including the following steps:
A) it requests ring to overflow frequency according to the I/O of the virtual machine of real-time monitoring, identifies that the upward queuing of data receiver is sensitive
Type virtual machine；
B) the vCPU dispatching priority for the queuing responsive type virtual machine that step a) is identified is improved.
2. the vCPU dispatching method according to claim 1 based on shared I/O virtualized environment, which is characterized in that described
In step a), request ring spilling frequency and I/O request queue overload frequency identification data receiver upward according to the I/O of virtual machine
Queuing responsive type virtual machine.
3. the vCPU dispatching method according to claim 2 based on shared I/O virtualized environment, which is characterized in that described
In step a), the process of the upward queuing responsive type virtual machine of identification data receiver includes substep:
A1 the I/O for) monitoring virtual machine requests ring characteristic；
A2) ring characteristic statistics I/O request ring is requested to overflow frequency and I/O request queue overload frequency according to I/O；
A3) when I/O request ring, which overflows any one of frequency and I/O request queue overload frequency, exceeds preset threshold value, sentence
Break the virtual machine queuing responsive type virtual machine upward for data receiver.
4. the vCPU dispatching method according to claim 3 based on shared I/O virtualized environment, which is characterized in that described
Step a1) in, the I/O request ring characteristic includes: the request number of I/O needed for current received data packet and I/O request
The I/O of the corresponding shared drive current residual of ring requests space, the step a2) in, according to I/O needed for current received data packet
The I/O of number and the corresponding shared drive current residual of I/O request ring is requested to request space, statistics I/O request ring overflows frequency
Rate.
5. the vCPU dispatching method according to claim 3 based on shared I/O virtualized environment, which is characterized in that described
Step a1) in, the I/O requests ring characteristic further include: the length of not processed I/O request queue in current I/O request ring
Degree, the step a2) in, the length of not processed I/O request queue in ring is requested according to current I/O, counts I/O request queue
Overload frequency.
6. the vCPU dispatching method according to claim 3 based on shared I/O virtualized environment, which is characterized in that described
Step a1) in, the I/O based on preset monitoring cycle monitoring virtual machine requests ring characteristic；
Affiliated step a2) in, based on the preset recognition cycle comprising several monitoring cycles, count virtual in the recognition cycle
The corresponding I/O request ring of machine overflows frequency and frequency is overloaded in I/O request queue.
7. the vCPU dispatching method according to claim 1 based on shared I/O virtualized environment, which is characterized in that described
Step a) further comprises, according to the vCPU overload situations of virtual machine and I/O request queue length, identifies that data sender is upward
Queuing responsive type virtual machine.
8. the vCPU dispatching method according to claim 7 based on shared I/O virtualized environment, which is characterized in that identification
The process of the upward queuing responsive type virtual machine of data sender includes substep:
A11) in each monitoring cycle, differentiate whether vCPU overloads in the monitoring cycle, and obtain the I/O in the monitoring cycle
Request queue length characteristic；
A12 the queuing responsive type virtual machine on biometric data sending direction, the request) are shaken according to request queue length
Queue length concussion is characterized in: the sum of I/O request queue length characteristic in the monitoring cycle of vCPU overload sum_rq_
The ratio of pattern and the monitoring cycle number count_cpu_ol of CPU overload in recognition cycle.
9. the vCPU dispatching method according to claim 8 based on shared I/O virtualized environment, which is characterized in that described
Identify the process of the upward queuing responsive type virtual machine of data sender further include: executing the step a12) before, according to
The size of the monitoring cycle number count_cpu_ol of CPU overload in recognition cycle, screens out non-queuing responsive type virtual machine in advance.
10. the vCPU dispatching method according to claim 8 based on shared I/O virtualized environment, which is characterized in that described
Step a12) further include: when request queue length concussion feature is in the numerical intervals near preset threshold, differentiate
Current virtual machine is to be lined up responsive type virtual machine.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510233215.2A CN104899098B (en) | 2015-05-08 | 2015-05-08 | A kind of vCPU dispatching method based on shared I/O virtualized environment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510233215.2A CN104899098B (en) | 2015-05-08 | 2015-05-08 | A kind of vCPU dispatching method based on shared I/O virtualized environment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104899098A CN104899098A (en) | 2015-09-09 |
| CN104899098B true CN104899098B (en) | 2019-02-01 |
Family
ID=54031775
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510233215.2A Active CN104899098B (en) | 2015-05-08 | 2015-05-08 | A kind of vCPU dispatching method based on shared I/O virtualized environment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104899098B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108241522B (en) * | 2016-12-27 | 2022-05-17 | 阿里巴巴集团控股有限公司 | Sleep state switching method and device in virtualization environment and electronic equipment |
| CN107220128A (en) * | 2017-06-19 | 2017-09-29 | 郑州云海信息技术有限公司 | CPU distribution method and device in a kind of group system |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101458631A (en) * | 2007-12-14 | 2009-06-17 | 联想(北京)有限公司 | Method for scheduling self-adapting virtual machine and computer |
| CN103984536A (en) * | 2014-02-14 | 2014-08-13 | 中国科学院计算技术研究所 | I/O (input/output) request counting system and method for cloud computing platform |
| CN104123174A (en) * | 2014-08-06 | 2014-10-29 | 华中科技大学 | Dynamic real-time CPU scheduling system under virtualization environment |
-
2015
- 2015-05-08 CN CN201510233215.2A patent/CN104899098B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101458631A (en) * | 2007-12-14 | 2009-06-17 | 联想(北京)有限公司 | Method for scheduling self-adapting virtual machine and computer |
| CN103984536A (en) * | 2014-02-14 | 2014-08-13 | 中国科学院计算技术研究所 | I/O (input/output) request counting system and method for cloud computing platform |
| CN104123174A (en) * | 2014-08-06 | 2014-10-29 | 华中科技大学 | Dynamic real-time CPU scheduling system under virtualization environment |
Non-Patent Citations (1)
| Title |
|---|
| Task-aware virtual machine scheduling for I/O performance;Hwanju Kim;《Proceedings of the 2009 ACM SIGPLAN/SIGOPS international conference on Virtual execution environments》;20090313;101-110页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104899098A (en) | 2015-09-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20200167258A1 (en) | Resource allocation based on applicable service level agreement | |
| US10628216B2 (en) | I/O request scheduling method and apparatus by adjusting queue depth associated with storage device based on hige or low priority status | |
| CN107924341B (en) | Real-time local and global datacenter network optimization based on platform telemetry data | |
| KR102506605B1 (en) | Rack-level scheduling for reducing the long tail latency using high performance ssds | |
| US9037717B2 (en) | Virtual machine demand estimation | |
| US9378032B2 (en) | Information processing method, information processing apparatus, recording medium, and system | |
| US8782322B2 (en) | Ranking of target server partitions for virtual server mobility operations | |
| US7882216B2 (en) | Process and methodology for generic analysis of metrics related to resource utilization and performance | |
| US11403220B2 (en) | Method and apparatus for adaptive cache load balancing for SSD-based cloud computing storage system | |
| US7685251B2 (en) | Method and apparatus for management of virtualized process collections | |
| US8417812B1 (en) | Methods and systems for detecting anomalies during IO accesses | |
| US8782671B2 (en) | Systems and methods for flexibly controlling resource usage by a driver domain on behalf of a virtual machine | |
| Ahmad et al. | {vIC}: Interrupt Coalescing for Virtual Machine Storage Device {IO} | |
| US9509621B2 (en) | Decentralized input/output resource management | |
| US11734204B2 (en) | Adaptive processor resource utilization | |
| US9836298B2 (en) | Deployment rule system | |
| EP3428801B1 (en) | Scheduling method and electronic device | |
| CN108337188A (en) | The traffic and the management of Load-aware dynamic queue | |
| EP2772854B1 (en) | Regulation method and regulation device for i/o channels in virtualization platform | |
| WO2017010922A1 (en) | Allocation of cloud computing resources | |
| US10733022B2 (en) | Method of managing dedicated processing resources, server system and computer program product | |
| CN105045667B (en) | A kind of resource pool management method for virtual machine vCPU scheduling | |
| CN104899098B (en) | A kind of vCPU dispatching method based on shared I/O virtualized environment | |
| US9769022B2 (en) | Timeout value adaptation | |
| CN104866370A (en) | Dynamic time slice dispatching method and system for parallel application under cloud computing environment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |