CN103530191B - Focus identifying processing method and device - Google Patents
Focus identifying processing method and device Download PDFInfo
- Publication number
- CN103530191B CN103530191B CN201310493598.8A CN201310493598A CN103530191B CN 103530191 B CN103530191 B CN 103530191B CN 201310493598 A CN201310493598 A CN 201310493598A CN 103530191 B CN103530191 B CN 103530191B
- Authority
- CN
- China
- Prior art keywords
- cpu
- numa
- focus
- passage
- information
- 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
Abstract
The embodiment of the present invention provides a kind of focus identifying processing method and device, and this method includes：According to the topology information of the physical property information of CPU in NUMA system and the physical property information of Node Controller, and the NUMA system, the channel performance counting statistics information of each passage in the NUMA system is obtained；According to the channel performance counting statistics information of each passage in the NUMA system, the focus section in the NUMA system is recognized；According to the focus section identified, identification causes at least one the first CPU of the focus section；At least one process switching run in first CPU is gone out to be currently located NUMA domains.In the embodiment of the present invention, hot issue is identified from physical layer, and targetedly solves hot issue, that is, realizes and physically solves hot issue.
Description
Technical field
The present invention relates to the communication technology, more particularly to a kind of focus identifying processing method and device.
Background technology
Accessed in Non Uniform Memory Access（Non Uniform Memory Access, abbreviation NUMA）In system, different centers
Processor（Central Processing Unit, abbreviation CPU）The different distance between internal memory, CPU accesses different distance
Internal memory is that the delay needed is different, for example access the internal memory of distal end necessarily than access that local internal memory spends when extend.In reality
In the database application scene on border, some memory sections or data can turn into focus, for example：Some Memory Controller Hub（Memory
Controller, abbreviation MC）Focus MC, some passages, which can be turned into, can turn into focus passage etc..In one NUMA system, if
CPU accesses focus memory sections and is related to remote access, can substantially reduce the performance of NUMA system.
In the prior art, it can only recognize that focus is optimized from application level, or, pass through operating system
（Operating System, abbreviation OS）Diagnostic tool identification focus optimize.
But, hot issue that all can only be on recognition logic using prior art, and None- identified goes out hardware link or logical
Hot issue on road, i.e., can not recognize and solve hot issue from physical layer.
The content of the invention
The embodiment of the present invention provides a kind of focus identifying processing method and device, for realizing how to know from physical layer
Not with solution hot issue.
First aspect of the embodiment of the present invention provides a kind of focus identifying processing method, including：
The physical property information and Node Controller of central processor CPU in NUMA system are accessed according to Non Uniform Memory Access
Physical property information, and the NUMA system topology information, obtain the channel performance of each passage in the NUMA system
Counting statistics information；
According to the channel performance counting statistics information of each passage in the NUMA system, the heat in the NUMA system is recognized
Point section；
According to the focus section identified, identification causes at least one the first CPU of the focus section；
At least one process switching run in first CPU is gone out to be currently located NUMA domains.
It is described according to Non Uniform Memory Access with reference in a first aspect, in the first possible embodiment of first aspect
CPU physical property information and the physical property information of Node Controller in access NUMA system, and the NUMA system
Topology information, is obtained in the NUMA system before the channel performance counting statistics information of each passage, in addition to：
In the acquisition units time in NUMA system CPU physical property information and node register physical property information,
And read the topology information of the NUMA system.
With reference in a first aspect, in second of possible embodiment of first aspect, it is described by the first CPU extremely
A few process switching run goes out to be currently located NUMA domains, including：
Calculate the weight for each process run in the first CPU；
According to the weight, at least one process switching run in the first CPU is gone out to be currently located NUMA domains.
With reference to second of possible embodiment of first aspect, in the third possible embodiment of first aspect
In, the weight for calculating each process run in the first CPU, including：
Obtain the CPU usage, CPU holding times, interprocess communication time for each process run in the first CPU
Number, cache hit rate and input and output readwrite bandwidth；
According to the CPU usage of each process, CPU holding times, interprocess communication number of times, cache hit rate and defeated
Enter and export readwrite bandwidth, calculate the weight of each process.
With reference to the third possible embodiment of first aspect, in the 4th kind of possible embodiment of first aspect
In, it is described that at least one process switching run in first CPU is gone out to be currently located NUMA domains, including：
The weight of each process is ranked up from high to low；
By the weight be more than predetermined threshold value process switching to the first NUMA domains, wherein, the first NUMA domains to institute
The distance for stating focus section is currently located NUMA domains to the distance of the focus section less than described.
Any one of the 4th kind of possible embodiment with reference to first aspect to first aspect, the 5th of first aspect the
Plant in possible embodiment, the channel performance counting statistics information includes following any or combination：Passage takes ratio, led to
The bandwidth in road, the utilization rate of passage.
Second aspect of the embodiment of the present invention provides a kind of focus recognition process unit, including：
Acquisition module, the physical property for accessing central processor CPU in NUMA system according to Non Uniform Memory Access is believed
The physical property information of breath and Node Controller, and the NUMA system topology information, obtain each in the NUMA system
The channel performance counting statistics information of passage；
First identification module, for the channel performance counting statistics information according to each passage in the NUMA system, identification
Focus section in the NUMA system；
Second identification module, for the focus section identified according to, identification causes the focus section at least
One the first CPU；
Handover module, for going out to be currently located NUMA domains by least one process switching run in the first CPU.
With reference to second aspect, in the first possible embodiment of second aspect, described device also includes：
Acquisition module, physical property information and node register for CPU in NUMA system in the acquisition units time
Physical property information, and read the topology information of the NUMA system.
With reference to second aspect, in second of possible embodiment of second aspect, the handover module, including：
Computing unit, the weight for calculating each process run in the first CPU；
Switch unit, for according to the weight, the process switching that at least one in the first CPU is run to go out to work as
Preceding place NUMA domains.
With reference to second of possible embodiment of second aspect, in the third possible embodiment of second aspect
In, the computing unit, when being taken specifically for obtaining the CPU usage for each process run in the first CPU, CPU
Between, interprocess communication number of times, cache hit rate and input and output readwrite bandwidth；According to CPU usage, the CPU of each process
Holding time, interprocess communication number of times, cache hit rate and input and output readwrite bandwidth, calculate the weight of each process.
With reference to the third possible embodiment of second aspect, in the 4th kind of possible embodiment of second aspect
In, the switch unit, specifically for the weight of each process is ranked up from high to low；The weight is more than default
The process switching of threshold value to the first NUMA domains, wherein, the distance in the first NUMA domains to the focus section is less than described work as
Distance of the preceding place NUMA domains to the focus section.
Any one of the 4th kind of possible embodiment with reference to second aspect to second aspect, the 5th of second aspect the
Plant in possible embodiment, the channel performance counting statistics information includes following any or combination：Passage takes ratio, led to
The bandwidth in road, the utilization rate of passage.
In the embodiment of the present invention, by the physical property information and NC physical property information of CPU in NUMA system, and
The topology information of NUMA system, obtains the channel performance counting statistics information of each passage in above-mentioned NUMA system, according to each passage
Channel performance counting statistics information recognize the focus section in NUMA system, and then recognize and to cause at least the one of focus section
Individual first CPU, the process switching that at least one in the first CPU is run goes out the NUMA domains being currently located, targetedly solved
Certainly hot issue, so also achieves and solves hot issue on hardware link and passage.
Brief description of the drawings
In order to illustrate more clearly about the embodiment of the present invention or technical scheme of the prior art, below will be to embodiment or existing
There is the accompanying drawing used required in technology description to be briefly described, it should be apparent that, drawings in the following description are this hairs
Some bright embodiments, for those of ordinary skill in the art, without having to pay creative labor, can be with
Other accompanying drawings are obtained according to these accompanying drawings.
The schematic flow sheet for the focus identifying processing embodiment of the method one that Fig. 1 provides for the present invention；
The application scenarios schematic diagram for the focus identifying processing embodiment of the method two that Fig. 2 provides for the present invention；
The structural representation for the focus recognition process unit embodiment one that Fig. 3 provides for the present invention；
The structural representation for the focus recognition process unit embodiment two that Fig. 4 provides for the present invention；
The structural representation for the focus recognition process unit embodiment three that Fig. 5 provides for the present invention.
Embodiment
To make the purpose, technical scheme and advantage of the embodiment of the present invention clearer, below in conjunction with the embodiment of the present invention
In accompanying drawing, the technical scheme in the embodiment of the present invention is clearly and completely described, it is clear that described embodiment is
A part of embodiment of the present invention, rather than whole embodiments.Based on the embodiment in the present invention, those of ordinary skill in the art
The every other embodiment obtained under the premise of creative work is not made, belongs to the scope of protection of the invention.
The schematic flow sheet for the focus identifying processing embodiment of the method one that Fig. 1 provides for the present invention, as shown in figure 1, the party
Method includes：
S101, physical property information and Node Controller according to CPU in NUMA system（Node Controller, referred to as
NC）Physical property information, and above-mentioned NUMA system topology information, obtain the channel performance of each passage in the NUMA system
Counting statistics information.
Channel performance counting statistics information can include the multiple performance parameter of each passage got.
Multiple CPU and multiple NC are included in one NUMA system, their physical property information is obtained respectively.Wherein,
CPU physical property information can be obtained by CPU register, and the physical property information of the CPU can include：The CPU's
Quick Path Interconnect（Quick Path Interconnect, abbreviation QPI）Input and output（incoming/outgoing）Stream
Amount, cache hit rate at different levels（cache hit rate）, each cycle instruction number（Instructions per Cycle, referred to as
IPC）Deng.NC physical property information can be obtained by NC register, and the physical property information of the NC can include：The NC
I/O traffic, cache hit rate, all kinds of counting messages information etc..
In addition, the topology information by obtaining NUMA system, can know that the connection in NUMA system between each equipment is closed
The information such as the hop count needed for system, and communication.Wherein, the annexation between each equipment include identical equipment between or difference set
The passage communicated between standby, for example：In NUMA system, QPI passages, CPU and the MC communicated between CPU and CPU it
Between the network interface that is communicated between the MC passages, NC and the NC that are communicated（Network Interface, abbreviation NI）It is logical
Road, CPU and input and output multiport transponder（Input Output Hub, abbreviation IOH）The input and output of communication（Input/
Output, abbreviation I/O）The passages such as passage.
S102, the channel performance counting statistics information according to each passage in above-mentioned NUMA system, recognize above-mentioned NUMA system
In focus section.
The channel performance counting statistics information got and default threshold value are compared, it is possible to identify hot zone
Section, the focus section can be focus MC, focus passage etc., and focus MC is exactly to be accessed to measure very big MC, and focus passage is exactly
CPU flows in NUMA system continue higher passage.
Above-mentioned channel performance counting statistics information can include：Passage takes ratio, the bandwidth of passage, the utilization rate of passage etc.
Information, when recognizing focus section, focus section is recognized according to the need for actual scene from suitable information.For example, knowing
It can typically be recognized during other focus MC according to the actual occupied bandwidth of passage.
S103, according to the above-mentioned focus section identified, identification causes at least one the first CPU of the focus section.I.e.
Identify after focus section, it is possible to further identify first CPU more comprising the focus section process is accessed, it is such
First CPU can have one or more.At the same time it can also identify the data flow for causing the focus section, you can to identify
CPU accesses the focus section by the direction of which passage, and data flow.
S104, the process switching for running at least one in above-mentioned first CPU go out the NUMA domains being currently located.
The weight for the process run in the first CPU can be calculated, then according to the different by some of processes of weight
It is switched in different NUMA domains, for example can be by the higher process switching of weight to NUMA domains nearer the distance MC to be accessed.
In the present embodiment, pass through the physical property information and NC physical property information of CPU in NUMA system, and NUMA
The topology information of system, obtains the channel performance counting statistics information of each passage in above-mentioned NUMA system, according to the logical of each passage
Road performance count statistical information recognizes the focus section in NUMA system, so recognize cause focus section at least one the
One CPU, the process switching that at least one in the first CPU is run goes out the NUMA domains being currently located, and targetedly solves heat
Point problem, so also achieves and solves hot issue from link and passage.
It should be noted that before above-mentioned S101, in addition to：CPU property in NUMA system in the acquisition units time
The physical property information of energy information and NC, and read the topology information of above-mentioned NUMA system.Specifically, can be with preset time
Interval is acquired, and recognizes focus according to the data gathered in a period of time, and solve hot issue.
It should be noted that the division in domain can be carried out to NUMA system according to the topology of NUMA system, it is general according to not
Hop count needed for the MC on particular CPU is accessed with CPU or apart from dividing, for example, can will access same MC hop counts or apart from phase
Same or close CPU is divided into same domain.
The application scenarios schematic diagram for the focus identifying processing embodiment of the method two that Fig. 2 provides for the present invention, as shown in Fig. 2
Assuming that Fig. 2 show the part of a NUMA system, wherein MC is hung on CPU, exemplified by only showing a focus MC in figure.
Specifically, it is assumed that the part of NUMA system shown in Fig. 2 includes 8 CPU, 4 NC, a focus MC.Assuming that the NUMA divided
Domain is：First NUMA domains include CPU1；2nd NUMA domains include CPU2, CPU3, CPU4；3rd NUMA domains include CPU5, CPU6,
CPU7、CPU8。
Further, on the basis of above-described embodiment, the above-mentioned process switching for being run at least one in the first CPU
Go out to be currently located NUMA domains, be specially：The weight for each process run in above-mentioned first CPU is calculated, will be upper according to the weight
State the process switching that at least one in the first CPU run and go out to be currently located NUMA domains.
It should be noted that the weight for each process run in above-mentioned above-mentioned first CPU of calculating, can be specifically included：
Obtain the CPU usage of each process run in above-mentioned first, CPU holding times, interprocess communication number of times, cache hit
Rate and input and output readwrite bandwidth；According to the CPU usage of each process, CPU holding times, interprocess communication number of times, caching life
Middle rate and input and output readwrite bandwidth, calculate the weight of above-mentioned each process.It can specifically calculate each using step analysis algorithm
The weight of process, illustrates the weight for calculating some process, it is assumed that D1、D2、D3、D4、D55 factors represent the process respectively
CPU usage, CPU holding times, interprocess communication number of times, cache hit rate and input and output readwrite bandwidth are to whole business
Influence degree, it is necessary to set up the comparator matrix compared two-by-two to 5 factors, taken successively from above-mentioned 5 factors two because
Sub- DiAnd Dj, wherein i={ 1,2,3,4,5 }, j={ 1,2,3,4,5 }, aijRepresent DiAnd DjThe ratio between influence degree to whole business,
Whole comparative results matrix A={ aij}5×5Represent, A meets following condition：（1）aij>0；（2）aij=1/aji.Then square is calculated
Battle array A obtains characteristic vector corresponding to eigenvalue of maximum, is phase of the corresponding factor of same level for whole business after normalization
To the weights { X of importance1,X2,X3,X4,X5}.In addition, in order to ensure the validity of above-mentioned calculating, it is necessary to enter to above-mentioned matrix A
Row consistency check, specifically method be：（1）Calculate coincident indicator：CI=(λmax- n)/(n-1), wherein, λmaxFor matrix A
Corresponding eigenvalue of maximum, n=5.（2）Search corresponding Aver-age Random Consistency Index RI.（3）Calculate consistency ration：CR=
CI/RI.Judge, if CR is less than 0.1, then it is assumed that the uniformity of matrix A is acceptable, otherwise, it is necessary to be repaiied to matrix A
Just.Afterwards, it is assumed that to D1、D2、D3、D4、D55 factors have been normalized, then can take the aggregate resource of the process
Amount, that is, the weight of the process are expressed as D=X1*D1+X2*D2+X3*D3+X4*D4+X5*D5。
Closer, the above-mentioned process switching for being run at least one in the first CPU goes out the NUMA domains being currently located,
It is specifically as follows：The weight of above-mentioned each process is ranked up from high in the end, then the process that weight is more than predetermined threshold value is cut
The first NUMA domains are changed to, wherein, the distance in above-mentioned first NUMA domains to above-mentioned focus section, which is less than, is currently located NUMA domains to upper
State the distance of focus section.By taking Fig. 2 as an example, it is assumed that CPU5 is to influence the process run in focus MC the first CPU, CPU5 to include
P1, P2, P3, by the calculating using preceding method, the weight of this 3 processes can be got respectively, according to the power calculated
Judge whether to need handover process again, the weight if there is process is more than predetermined threshold value, then switches over, by fractional weight
Higher cutting out is currently located NUMA domains.Assuming that P2 weight is maximum, then P2 can be switched to CPU1 operations, CPU1 distances
Focus MC is nearest；Assuming that P1 weight time is high, P1 can be switched to any of CPU2, CPU3, CPU4, this 3 CPU distances
Focus MC is nearer, and needs not move through NC, can reduce the time delay for accessing focus MC.
It should be noted that can be switched over during handover process using Lightweight Process switching method.After the completion of switching
Execution preceding method can be circulated, constantly detection focus section, until having solved hot issue.
During implementing, above-mentioned channel performance counting statistics information be passage take than when, passage is taken than can be with
Including channel data（data）Take ratio and/or multi-channel interrogation（snoop）Take ratio, in some application scenarios, an equipment
With another equipment communication, it is assumed that the first equipment sends request message to the second equipment, is set to the second device request data, second
Standby receive may directly return to the data of the first device request to the first equipment after the request message（data）, it is also possible to exist and ask
Topic, inquiry message is returned to the first equipment.So obtain passage data in passage and take ratio and/or passage snoop occupancy ratios, come
Further recognize focus section, the passage data of general focus section is taken than larger, and passage snoop take it is smaller, can
To set predetermined threshold value according to concrete scene.
The structural representation for the focus recognition process unit embodiment one that Fig. 3 provides for the present invention, as shown in figure 3, the dress
Put including：Acquisition module 301, the first identification module 302, the second identification module 303 and handover module 304, wherein：
Acquisition module 301, the physical property for accessing central processor CPU in NUMA system according to Non Uniform Memory Access
The physical property information of information and Node Controller, and the NUMA system topology information, obtain in the NUMA system
The channel performance counting statistics information of each passage.
First identification module 302, for the channel performance counting statistics information according to each passage in the NUMA system, knows
Focus section in not described NUMA system.
Second identification module 303, for the focus section identified according to, identification causes the focus section extremely
Few first CPU.
Handover module 304, for going out to be currently located NUMA at least one process switching run in the first CPU
Domain.
Said apparatus is used to perform preceding method embodiment, and its implementing principle and technical effect is similar, will not be repeated here.
The structural representation for the focus recognition process unit embodiment two that Fig. 4 provides for the present invention, on the basis of Fig. 3,
Said apparatus can also include acquisition module 305, for CPU in NUMA system in the acquisition units time physical property information and
The physical property information of node register, and read the topology information of the NUMA system.
The structural representation for the focus recognition process unit embodiment three that Fig. 5 provides for the present invention, on the basis of Fig. 3,
Above-mentioned handover module 304 includes：Computing unit 501 and switch unit 502, wherein：
Computing unit 501, the weight for calculating each process run in the first CPU；Switch unit 502, is used
According to the weight, at least one process switching run in the first CPU is gone out to be currently located NUMA domains.
Further, above-mentioned computing unit 501, the CPU specifically for obtaining each process run in the first CPU
Occupancy, CPU holding times, interprocess communication number of times, cache hit rate and input and output readwrite bandwidth；Respectively entered according to described
CPU usage, CPU holding times, interprocess communication number of times, cache hit rate and the input and output readwrite bandwidth of journey, calculate institute
State the weight of each process.
Switch unit 502, specifically for the weight of each process is ranked up from high to low；The weight is more than
The process switching of predetermined threshold value to the first NUMA domains, wherein, the distance in the first NUMA domains to the focus section is less than institute
State the distance for being currently located NUMA domains to the focus section.
In addition, above-mentioned channel performance counting statistics information includes following any or combination：Passage takes ratio, the band of passage
The wide, utilization rate of passage.
Said apparatus is used to perform preceding method embodiment, and its implementing principle and technical effect is similar, will not be repeated here.
The embodiment of the present invention also provides another focus recognition process unit, including：Processor.
The processor, the physical property for accessing central processor CPU in NUMA system according to Non Uniform Memory Access is believed
The physical property information of breath and Node Controller, and the NUMA system topology information, obtain each in the NUMA system
The channel performance counting statistics information of passage；According to the channel performance counting statistics information of each passage in the NUMA system, know
Focus section in not described NUMA system；According to the focus section identified, identification causes the focus section extremely
Few first CPU；At least one process switching run in first CPU is gone out to be currently located NUMA domains.
In addition, the processor, the physical property information and node for being additionally operable to CPU in NUMA system in the acquisition units time is posted
The physical property information of storage, and read the topology information of the NUMA system.
Above-mentioned processor, the weight specifically for calculating each process run in the first CPU；According to the power
Weight, goes out to be currently located NUMA domains by least one process switching run in the first CPU.
Taken more specifically, above-mentioned processor obtains the CPU usage for each process run in the first CPU, CPU
Time, interprocess communication number of times, cache hit rate and input and output readwrite bandwidth；According to the CPU usage of each process,
CPU holding times, interprocess communication number of times, cache hit rate and input and output readwrite bandwidth, calculate the power of each process
Weight.
The weight of each process is ranked up by above-mentioned processor from high to low；The weight is more than predetermined threshold value
Process switching to the first NUMA domains, wherein, the distance in the first NUMA domains to the focus section is less than described be currently located
Distance of the NUMA domains to the focus section.
It should be noted that the channel performance counting statistics information includes following any or combination：Passage occupancy ratio,
The bandwidth of passage, the utilization rate of passage.
Said apparatus is used to perform preceding method embodiment, and its implementing principle and technical effect is similar, will not be repeated here.
One of ordinary skill in the art will appreciate that：Realizing all or part of step of above-mentioned each method embodiment can lead to
The related hardware of programmed instruction is crossed to complete.Foregoing program can be stored in a computer read/write memory medium.The journey
Sequence upon execution, performs the step of including above-mentioned each method embodiment；And foregoing storage medium includes：ROM, RAM, magnetic disc or
Person's CD etc. is various can be with the medium of store program codes.
Finally it should be noted that：Various embodiments above is merely illustrative of the technical solution of the present invention, rather than its limitations；To the greatest extent
The present invention is described in detail with reference to foregoing embodiments for pipe, it will be understood by those within the art that：Its according to
The technical scheme described in foregoing embodiments can so be modified, or which part or all technical characteristic are entered
Row equivalent substitution；And these modifications or replacement, the essence of appropriate technical solution is departed from various embodiments of the present invention technology
The scope of scheme.
Claims (12)
1. a kind of focus identifying processing method, it is characterised in that including：
The physical property information and the thing of Node Controller of central processor CPU in NUMA system are accessed according to Non Uniform Memory Access
Manage performance information, and the NUMA system topology information, the channel performance for obtaining each passage in the NUMA system counts
Statistical information；
According to the channel performance counting statistics information of each passage in the NUMA system, the hot zone in the NUMA system is recognized
Section, the focus section is focus Memory Controller Hub MC or focus passage；
According to the focus section identified, identification causes at least one the first CPU of the focus section；
The process switching that at least one in first CPU is run is to the first NUMA domains；
Wherein, the first NUMA domains are less than what at least one in the first CPU was run to the distance of the focus section
Process is currently located NUMA domains to the distance of the focus section.
2. according to the method described in claim 1, it is characterised in that described to be accessed according to Non Uniform Memory Access in NUMA system
CPU physical property information and the physical property information of Node Controller, and the NUMA system topology information, obtain institute
Before the channel performance counting statistics information for stating each passage in NUMA system, in addition to：
In the acquisition units time in NUMA system CPU physical property information and node register physical property information, and read
Take the topology information of the NUMA system.
3. according to the method described in claim 1, it is characterised in that described to be run at least one in the first CPU
Process switching to the first NUMA domains, including：
Calculate the weight for each process run in the first CPU；
According to the weight, the process switching that at least one in the first CPU is run to the first NUMA domains.
4. method according to claim 3, it is characterised in that each process run in calculating the first CPU
Weight, including：
Obtain the CPU usage of each process run in the first CPU, it is CPU holding times, interprocess communication number of times, slow
Deposit hit rate and input and output readwrite bandwidth；
It is defeated according to the CPU usage of each process, CPU holding times, interprocess communication number of times, cache hit rate and input
Go out readwrite bandwidth, calculate the weight of each process.
5. method according to claim 4, it is characterised in that described to be run at least one in the first CPU
Process switching to the first NUMA domains, including：
The weight of each process is ranked up from high to low；
The weight is more than the process switching of predetermined threshold value to the first NUMA domains.
6. the method according to any one of claim 1-5, it is characterised in that the channel performance counting statistics packet
Include following any or combination：Passage takes ratio, the bandwidth of passage, the utilization rate of passage.
7. a kind of focus recognition process unit, it is characterised in that including：
Acquisition module, for according to Non Uniform Memory Access access NUMA system in central processor CPU physical property information and
The physical property information of Node Controller, and the NUMA system topology information, obtain each passage in the NUMA system
Channel performance counting statistics information；
First identification module, for the channel performance counting statistics information according to each passage in the NUMA system, identification is described
Focus section in NUMA system, the focus section is focus Memory Controller Hub MC or focus passage；
Second identification module, for the focus section identified according to, identification causes at least one of the focus section
First CPU；
Handover module, for the process switching that is run at least one in the first CPU to the first NUMA domains；
Wherein, the first NUMA domains are less than what at least one in the first CPU was run to the distance of the focus section
Process is currently located NUMA domains to the distance of the focus section.
8. device according to claim 7, it is characterised in that also include：
Acquisition module, physical property information and the physics of node register for CPU in NUMA system in the acquisition units time
Performance information, and read the topology information of the NUMA system.
9. device according to claim 7, it is characterised in that the handover module, including：
Computing unit, the weight for calculating each process run in the first CPU；
Switch unit, for according to the weight, the process switching that at least one in the first CPU is run to described
One NUMA domains.
10. device according to claim 9, it is characterised in that the computing unit, specifically for obtaining described first
The CPU usage of each process run in CPU, CPU holding times, interprocess communication number of times, cache hit rate and input defeated
Go out readwrite bandwidth；According to the CPU usage of each process, CPU holding times, interprocess communication number of times, cache hit rate and
Input and output readwrite bandwidth, calculates the weight of each process.
11. device according to claim 10, it is characterised in that the switch unit, specifically for by each process
Weight be ranked up from high to low；The weight is more than the process switching of predetermined threshold value to the first NUMA domains.
12. the device according to claim any one of 7-11, it is characterised in that the channel performance counting statistics packet
Include following any or combination：Passage takes ratio, the bandwidth of passage, the utilization rate of passage.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310493598.8A CN103530191B (en) | 2013-10-18 | 2013-10-18 | Focus identifying processing method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310493598.8A CN103530191B (en) | 2013-10-18 | 2013-10-18 | Focus identifying processing method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103530191A CN103530191A (en) | 2014-01-22 |
| CN103530191B true CN103530191B (en) | 2017-09-12 |
Family
ID=49932228
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310493598.8A Active CN103530191B (en) | 2013-10-18 | 2013-10-18 | Focus identifying processing method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103530191B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| BR112017008407B1 (en) * | 2015-12-29 | 2023-04-04 | xFusion Digital Technologies Co., Ltd | CPU AND MULTIPLE CPU SYSTEM MANAGEMENT METHOD |
| CN105912441A (en) * | 2016-04-05 | 2016-08-31 | 浪潮电子信息产业股份有限公司 | Bandwidth monitoring method, device and system for NUMA chips |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060123423A1 (en) * | 2004-12-07 | 2006-06-08 | International Business Machines Corporation | Borrowing threads as a form of load balancing in a multiprocessor data processing system |
| US7739434B2 (en) * | 2008-01-11 | 2010-06-15 | International Business Machines Corporation | Performing a configuration virtual topology change and instruction therefore |
| US8245236B2 (en) * | 2008-02-27 | 2012-08-14 | International Business Machines Corporation | Lock based moving of threads in a shared processor partitioning environment |
| CN100562854C (en) * | 2008-03-11 | 2009-11-25 | 浙江大学 | The implementation method of load equalization of multicore processor operating system |
| US8443376B2 (en) * | 2010-06-01 | 2013-05-14 | Microsoft Corporation | Hypervisor scheduler |
| CN103324592B (en) * | 2013-06-24 | 2016-11-23 | 华为技术有限公司 | A kind of Data Migration control method, data migration method and device |
-
2013
- 2013-10-18 CN CN201310493598.8A patent/CN103530191B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN103530191A (en) | 2014-01-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110264326B (en) | Method, device and equipment for identifying abnormal account set and risk account set | |
| CN109803004B (en) | Block chain intelligent contract management method and device, electronic equipment and storage medium | |
| US9858228B2 (en) | Dynamic assignment of groups of resources in a peripheral component interconnect express network | |
| CN109710405B (en) | Block chain intelligent contract management method and device, electronic equipment and storage medium | |
| WO2015081757A1 (en) | Cold and hot data identification threshold calculation method, apparatus and system | |
| DE112016007516T5 (en) | DEVICES AND METHOD FOR A PROCESSOR ARCHITECTURE | |
| CN109587072A (en) | Distributed system overall situation speed limiting system and method | |
| EP3929745A1 (en) | Apparatus and method for a closed-loop dynamic resource allocation control framework | |
| CN106527985A (en) | Storage interaction device and storage system based on ceph | |
| CN104102549B (en) | A kind of method, apparatus and chip for realizing multithreading mutually exclusive operation | |
| Lee et al. | ReRe: A lightweight real-time ready-to-go anomaly detection approach for time series | |
| CN105827678B (en) | Communication means and node under a kind of framework based on High Availabitity | |
| CN103530191B (en) | Focus identifying processing method and device | |
| CN103309843B (en) | The collocation method of server and system | |
| Wang et al. | GeoCol: A geo-distributed cloud storage system with low cost and latency using reinforcement learning | |
| Wang et al. | A Data-Aware Remote Procedure Call Method for Big Data Systems. | |
| EP3929746A1 (en) | Apparatus and method for a resource allocation control framework using performance markers | |
| CN107341193B (en) | Method for inquiring mobile object in road network | |
| CN116841753A (en) | Stream processing and batch processing switching method and switching device | |
| CN106933882A (en) | A kind of big data incremental calculation method and device | |
| CN110618963A (en) | Heterogeneous computing hardware acceleration system and method integrating computing and storage | |
| CN108399145A (en) | A kind of CPU-GPU heterogeneous platforms share last level cache management method, framework and device | |
| CN114968073A (en) | Data prefetching method, equipment and system | |
| CN106027685A (en) | Peak access method based on cloud computation system | |
| CN104750640A (en) | Method and apparatus for arbitrating among multiple channels to access a resource |
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 | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20200421 Address after: 518129 Bantian HUAWEI headquarters office building, Longgang District, Guangdong, Shenzhen Patentee after: HUAWEI TECHNOLOGIES Co.,Ltd. Address before: 301, A building, room 3, building 301, foreshore Road, No. 310053, Binjiang District, Zhejiang, Hangzhou Patentee before: Huawei Technologies Co.,Ltd. |
|
| TR01 | Transfer of patent right |