CN107132990A - Reading I O scheduling method and device based on super fusion storage - Google Patents
Reading I O scheduling method and device based on super fusion storage Download PDFInfo
- Publication number
- CN107132990A CN107132990A CN201610108881.8A CN201610108881A CN107132990A CN 107132990 A CN107132990 A CN 107132990A CN 201610108881 A CN201610108881 A CN 201610108881A CN 107132990 A CN107132990 A CN 107132990A
- Authority
- CN
- China
- Prior art keywords
- strange land
- local replica
- copy
- virtual machine
- time delays
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F3/00—Input arrangements for transferring data to be processed into a form capable of being handled by the computer; Output arrangements for transferring data from processing unit to output unit, e.g. interface arrangements
- G06F3/06—Digital input from, or digital output to, record carriers, e.g. RAID, emulated record carriers or networked record carriers
- G06F3/0601—Interfaces specially adapted for storage systems
- G06F3/0602—Interfaces specially adapted for storage systems specifically adapted to achieve a particular effect
- G06F3/061—Improving I/O performance
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F3/00—Input arrangements for transferring data to be processed into a form capable of being handled by the computer; Output arrangements for transferring data from processing unit to output unit, e.g. interface arrangements
- G06F3/06—Digital input from, or digital output to, record carriers, e.g. RAID, emulated record carriers or networked record carriers
- G06F3/0601—Interfaces specially adapted for storage systems
- G06F3/0668—Interfaces specially adapted for storage systems adopting a particular infrastructure
- G06F3/067—Distributed or networked storage systems, e.g. storage area networks [SAN], network attached storage [NAS]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a kind of reading I O scheduling method based on super fusion storage, the reading I O scheduling method based on super fusion storage comprises the following steps:When virtual machine is run on the node of local replica, local replica limit index parameter is obtained;The higher limit of IO pressure values on the local replica is determined according to the limit index parameter;Whether the current IO pressure values of real-time judge local replica are more than the higher limit;If so, then the read request of the virtual machine is sent to strange land copy;If it is not, then the read request of the virtual machine is sent to local replica.The invention also discloses a kind of reading I O scheduling device based on super fusion storage.The present invention improves the performance for reading IO.
Description
Technical field
The present invention relates to communication technique field, more particularly to a kind of reading I O scheduling side based on super fusion storage
Method and device.
Background technology
At present, on super integrated environment, each node is both memory node and calculate node, void
When plan machine is run on super integrated environment, it reads IO performances is influenceed by the physical location where copy.It is existing
There is reading I O scheduling algorithm mainly using IO localizations, i.e. automatic identification local replica is read, to be then attached to
Read I O operation is carried out on this local replica.This scheme takes full advantage of the advantage of localization, but works as
IO reaches that the hardware just occurred during certain pressure on local replica reaches performance bottleneck, and I O access ability will
Limited by local hardware performance, it is impossible to play the ability of distributed type assemblies, so as to cause the property for reading IO
Can be poor.
The content of the invention
It is a primary object of the present invention to provide a kind of reading I O scheduling method based on super fusion storage and dress
Put, it is intended to which IO performance is read in lifting.
To achieve the above object, a kind of reading I O scheduling method bag based on super fusion storage that the present invention is provided
Include following steps:
When virtual machine is run on the node of local replica, local replica limit index parameter is obtained;
The higher limit of IO pressure values on the local replica is determined according to the limit index parameter;
Whether the current IO pressure values of real-time judge local replica are more than the higher limit;
If so, then the read request of the virtual machine is sent to strange land copy;
If it is not, then the read request of the virtual machine is sent to local replica.
Preferably, the read request by the virtual machine sends to strange land copy and included:
When the strange land copy is two or more, then the read request of the virtual machine is rotated into transmission successively
To each strange land copy.
In addition, to achieve the above object, the present invention also provides a kind of reading I O scheduling based on super fusion storage
Method comprises the following steps:
When virtual machine is run on the node of local replica, the limit index parameter of strange land copy is obtained;
Network transfer delay and strange land IO time delays;And according to the IO pressure condition history numbers on local replica
According to and the limit index parameter of local replica calculate the IO time delays that handle on local replica, according to strange land copy
On IO pressure conditions historical data and strange land copy limit index parameter calculate current network congestion time delay
And strange land copy IO congestion time delays;The network transfer delay, which is that the current read request of virtual machine is corresponding, asks
Seek the network transfer delay of the strange land copy of block;The strange land IO time delays are the current read request pair of virtual machine
The strange land IO time delays for the request block answered;The IO time delays handled on the local replica are that virtual machine is current
The IO time delays that the corresponding request block of read request is handled on local replica;
Strange land copy IO is extended to when calculating the network transfer delay, strange land IO time delays, current network congestion
Congestion time delay sum, and it is set as the corresponding request block of the current read request of the virtual machine in strange land copy
The IO time delays of upper processing;
Judge whether the IO time delays handled on the strange land copy are less than the IO handled on the local replica
Time delay;
Sent if so, then performing the read request by the virtual machine to strange land copy step;
Sent if it is not, then performing the read request by the virtual machine to local replica step.
Preferably, the limit of the IO pressure conditions historical data and local replica according on local replica
Index parameter, which calculates the IO time delays handled on local replica, to be included:
IO pressure conditions on detection local replica, form local replica IO pressure condition historical datas in real time;
The local replica IO pressure conditions historical data includes request block size and corresponding time delay;
The limit index ginseng of local replica is extended to according to average request block size, the mean time on local replica
Number calculates the IO time delays that the corresponding request block of the current read request is handled on local replica.
Preferably, the limit of the IO pressure conditions historical data and strange land copy according on the copy of strange land
Index parameter extends to strange land copy IO congestion time delays when calculating current network congestion to be included:
IO pressure conditions on detection strange land copy, form strange land copy IO pressure condition historical datas in real time;
The strange land copy IO pressure conditions historical data includes request block size and corresponding time delay;
Current net is calculated according to the limit index parameter that average request block size, mean time extend to strange land copy
Network congestion time delay and strange land copy IO congestion time delays.
In addition, to achieve the above object, the present invention also provides a kind of reading I O scheduling based on super fusion storage
Device, the reading I O scheduling device based on super fusion storage includes:
First acquisition module, it is local secondary for when virtual machine is run on the node of local replica, obtaining
This limit index parameter;
Determining module, for determining IO pressure values on the local replica according to the limit index parameter
Higher limit;
Whether the first judge module, be more than the upper limit for the current IO pressure values of real-time judge local replica
Value;
Output module, for when the current IO pressure values of local replica are more than the higher limit, by the void
The read request of plan machine is sent to strange land copy;When the current IO pressure values of local replica are less than or equal to the upper limit
During value, the read request of the virtual machine is sent to local replica.
Preferably, the output module will be specifically for when the strange land copy is two or more, then will
The read request of the virtual machine rotates transmission to each strange land copy successively.
In addition, to achieve the above object, the present invention also provides a kind of reading I O scheduling based on super fusion storage
Device, the reading I O scheduling device based on super fusion storage includes:
Second acquisition module, for when the strange land copy is two or more, obtaining the pole of strange land copy
Limit index parameter;
First computing module, during for being transmitted according to the limit index parameter calculating network of the strange land copy
Prolong and strange land IO time delays;And according to the IO pressure conditions historical data on local replica and local replica
Limit index parameter calculates the IO time delays handled on local replica, according to the IO pressure feelings on the copy of strange land
Condition historical data and the limit index parameter of strange land copy extend to strange land copy IO when calculating current network congestion
Congestion time delay;The network transfer delay is the strange land pair of the corresponding request block of the current read request of virtual machine
This network transfer delay;The strange land IO time delays are the corresponding request block of the current read request of virtual machine
Strange land IO time delays;The IO time delays handled on the local replica are that the current read request of virtual machine is corresponding
The IO time delays that request block is handled on local replica;
Second computing module, is gathered around for calculating the network transfer delay, strange land IO time delays, current network
Strange land copy IO congestion time delay sums are extended to during plug, and are set as the current read request correspondence of the virtual machine
The IO time delays that are handled on the copy of strange land of request block;
Whether the second judge module, the IO time delays for judging to handle on the strange land copy are less than described
The IO time delays handled on ground copy;If so, it is described by the virtual machine then to trigger the output module execution
Read request send to strange land copy operation;Otherwise trigger the output module perform it is described will it is described virtually
The read request of machine is sent to local replica operation.
Preferably, second computing module specifically for,
IO pressure conditions on detection local replica, form local replica IO pressure condition historical datas in real time;
The local replica IO pressure conditions historical data includes request block size and corresponding time delay;And according to
The limit index parameter that average request block size, mean time on local replica extend to local replica calculates institute
State the IO time delays that the corresponding request block of current read request is handled on local replica.
Preferably, second computing module specifically for,
IO pressure conditions on detection strange land copy, form strange land copy IO pressure condition historical datas in real time;
The strange land copy IO pressure conditions historical data includes request block size and corresponding time delay;And according to
When the limit index parameter that average request block size, mean time extend to strange land copy calculates current network congestion
Prolong and strange land copy IO congestion time delays.
When the embodiment of the present invention is run by virtual machine on the node of local replica, local replica pole is obtained
Limit index parameter;The higher limit of IO pressure values on the local replica is determined according to the limit index parameter;
Whether the current IO pressure values of real-time judge local replica are more than the higher limit;If so, then will be described virtual
The read request of machine is sent to strange land copy;If it is not, then the read request of the virtual machine is sent to local pair
This.So as to ensure using IO localizations are read in the case of the IO pressure of local replica is less, when local
Read request is sent to strange land copy in the case that copy IO pressure is larger and carries out read operation, therefore is improved
Read IO performance.
Brief description of the drawings
The schematic flow sheet for the embodiment of reading I O scheduling method one that Fig. 1 is stored for the present invention based on super fusion;
The schematic flow sheet for another embodiment of reading I O scheduling method that Fig. 2 is stored for the present invention based on super fusion;
Fig. 3 illustrates for the functional module of the embodiment of reading I O scheduling device one of the present invention based on super fusion storage
Figure;
Fig. 4 shows for the functional module of another embodiment of reading I O scheduling device of the present invention based on super fusion storage
It is intended to.
The realization, functional characteristics and advantage of the object of the invention will be done further referring to the drawings in conjunction with the embodiments
Explanation.
Embodiment
It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, limit is not used to
The fixed present invention.
The present invention provides a kind of reading I O scheduling method based on super fusion storage, and reference picture 1 is implemented one
In example, it should be included based on the reading I O scheduling method of super fusion storage:
Step S10, when virtual machine is run on the node of local replica, obtains local replica limit index
Parameter;
The reading I O scheduling method based on super fusion storage that the present invention is provided is mainly used in distributed type assemblies
In, it is scheduled for the read request to virtual machine.Specifically, in distributed type assemblies, generally include
Two or more copy.
Super fusion:Hyper-Converged Infrastructure, or referred to as " HCI " refers in same set of list
Not only possess the resources such as calculating, network, storage and server virtualization and technology in element device, and
Also include the elements such as backup software, snapping technique, data de-duplication, online data compression, and cover more
Unit can be got up by network polymerization, realize modular seamless (scale-out) extending transversely,
Form unified resource pool.
On super integrated environment, each node is that memory node is also calculate node, and virtual machine exists
When being run on super integrated environment, it reads IO performances is influenceed by the physical location where copy.
Local replica:It is that computing resource is also storage resource to refer to the same node under super integrated environment,
When the node that virtual machine is run has a copy of corresponding virtual machine image file, the copy is referred to as
Local replica;Conversely, the copy being present on other nodes is referred to as strange land copy or distal end copy (this hair
Referred to as strange land copy in bright embodiment).
Above-mentioned limit index parameter includes IOPS (number for referring to processing I/O Request per second) and maximum throughput
Amount.When non-storage limit index parameter on local replica, the limit index parameter can be tested first,
For example processing 4KB in 1S can be tested by the abilities of the fritter I O read requests for testing local replica
Request block number so as to obtaining IOPS indexs;Can also be by testing asking for 128KB or 1MB
The throughput data of block is asked to measure the maximum throughput of local replica.
Step S20, the upper limit of IO pressure values on the local replica is determined according to the limit index parameter
Value;
Above-mentioned limit index parameter according to getting can determine the limit pressure of IO on local replica,
Can be directly using the limit pressure as above-mentioned higher limit, can also be above-mentioned according to preset ratio-dependent
Higher limit, such as higher limit can be the 90% of limit pressure.Specific set-up mode can basis
It is actually needed and is configured, does not limit further herein.This can also be obtained according to limit index parameter
Go out the theoretical IO time delays that a formula one is used to calculate any block size on local replica, i.e., formula one is defeated
Enter test block size (obtaining limit index parameter to obtain), test time delay and (obtain limit index parameter
When obtain) and block size (being read request corresponding request block);It is output as theoretical IO time delays;Its
Middle input parameter test block size and test time delay, the two parameters are worth when being according to actual test,
Then it is considered that they are a constants, so formula one is described as after can simplifying, input as any piece
Size, is output as theoretical IO time delays.There is this formula and also make IO predictions on local replica to be follow-up
Time delay provides foundation.
Whether step S30, the current IO pressure values of real-time judge local replica are more than the higher limit;If so,
Step S40 is then performed, step S50 is otherwise performed;
Step S40, the read request of the virtual machine is sent to strange land copy;
Step S50, the read request of the virtual machine is sent to local replica.
In the present embodiment, the current IO of local replica pressure condition can be detected in real time, according to local replica
Current IO pressure conditions determine current IO pressure values.Then, whether the current IO pressure values of real-time judge
More than above-mentioned higher limit.When more than above-mentioned higher limit, it is determined that the IO pressure of local replica it is larger or
The IO limiting pressures of local replica are reached, the read request of virtual machine is now sent to strange land copy, so that
Mitigate the IO pressure of local replica.
When the embodiment of the present invention is run by virtual machine on the node of local replica, local replica pole is obtained
Limit index parameter;The higher limit of IO pressure values on the local replica is determined according to the limit index parameter;
Whether the current IO pressure values of real-time judge local replica are more than the higher limit;If so, then will be described virtual
The read request of machine is sent to strange land copy;If it is not, then the read request of the virtual machine is sent to local pair
This.So as to ensure using IO localizations are read in the case of the IO pressure of local replica is less, when local
Read request is sent to strange land copy in the case that copy IO pressure is larger and carries out read operation, therefore is improved
Read IO performance.
Further, because the quantity of strange land copy can be configured according to actual needs, when having two
It is individual or two with strange land copy when, in order to improve read IO performance, strange land copy rotation can be carried out.
Specifically, based on above-described embodiment, in the present embodiment, above-mentioned steps S40 includes:
When the strange land copy is two or more, then the read request of the virtual machine is rotated into transmission successively
To each strange land copy.
In the present embodiment, rotated successively for many strange land copies, for example, read strange land copy A for the first time, then
Second of reading strange land copy B, third time is read strange land copy C, rotated successively repeatedly.This scheme can be with
The reading IO pressure of scattered list copy, and then shorten the time of processing read request, and then reach that IO is read in lifting
Performance effect.
Further, reference picture 2, the present invention also provides the reading I O scheduling method the based on super fusion storage
Two embodiments, the reading I O scheduling method based on super fusion storage that the present embodiment is provided includes:
Step S60, when virtual machine is run on the node of local replica, the limit for obtaining strange land copy refers to
Mark parameter;
Step S70, calculating the reading of the virtual machine currently according to the limit index parameter of the strange land copy please
Seek the network transfer delay and strange land IO time delays of the strange land copy of corresponding request block;And according to local replica
On IO pressure conditions historical data and the limit index parameter of local replica to calculate the virtual machine current
The IO time delays that the corresponding request block of read request is handled on local replica, are pressed according to the IO on the copy of strange land
Power situation historical data and the limit index parameter of strange land copy extend to strange land pair when calculating current network congestion
This IO congestion time delays;
Step S80, is extended to when calculating the network transfer delay, strange land IO time delays, current network congestion
Strange land copy IO congestion time delay sums obtain the corresponding request block of the current read request of the virtual machine in strange land
The IO time delays handled on copy;
Step S90, judges whether the IO time delays handled on the strange land copy are less than on the local replica
The IO time delays of processing;If so, then performing step S100, step S110 is otherwise performed.
Step S100, the read request of the virtual machine is sent to strange land copy;
Step S110, the read request of the virtual machine is sent to local replica.
In the present embodiment, it is larger that above-mentioned higher limit is generally set, relatively the limiting value of local replica,
When the IO pressure of local replica is larger and not less than above-mentioned higher limit, will easily occur locating on the copy of strange land
The IO time delays of reason are also smaller than the IO time delays handled on the local replica, if still sending read request
To local replica, then the time delay for reading IO will be longer, be unfavorable for improving the performance for reading IO.The present embodiment exists
Execution sends read request to local replica, judge the IO time delays that are handled on the strange land copy with it is described
The relation of the IO time delays handled on local replica, according to the IO time delays handled on the strange land copy with it is described
Whether the relation of the IO time delays handled on local replica, it is determined whether shunted to read request, i.e., will read
Request is sent to strange land copy, is handled for strange land copy.
If it is understood that there is no the pole of the strange land copy of strange land copy in the equipment of storage local replica
Index parameter is limited, strange land can also be tested according to the same mode of the limit index parameter of test local replica
The limit index parameter of copy, so as to obtain the limit index parameter of strange land copy.
Specifically, the limit of above-mentioned IO pressure conditions historical data and local replica according on local replica
Index parameter calculates the IO that the corresponding request block of the current read request of the virtual machine is handled on local replica
Time delay includes:
IO pressure conditions on detection local replica, form local replica IO pressure condition historical datas in real time;
The local replica IO pressure conditions historical data includes request block size and corresponding time delay;
The limit index ginseng of local replica is extended to according to average request block size, the mean time on local replica
Number calculates the IO time delays that the corresponding request block of the current read request is handled on local replica.
It is understood that in the present embodiment, above-mentioned current read request is corresponding to ask block in local replica
The IO time delays of upper processing refer to that local replica handles the time spent in current read request is estimated.
The limit index parameter of above-mentioned IO pressure conditions historical data and strange land copy according on the copy of strange land
Extending to strange land copy IO congestion time delays when calculating current network congestion includes:
IO pressure conditions on detection strange land copy, form strange land copy IO pressure condition historical datas in real time;
The strange land copy IO pressure conditions historical data includes request block size and corresponding time delay;
Current net is calculated according to the limit index parameter that average request block size, mean time extend to strange land copy
Network congestion time delay and strange land copy IO congestion time delays.
The embodiment of the present invention, is sentenced by predicting the IO time delays of local replica and the IO time delays of strange land copy
It is disconnected whether read request to be shunted, thus local replica IO pressure not close to before above-mentioned higher limit
Shunting can be started, so as to further improve the performance for reading IO.
Specifically, the reading of the virtual machine currently is being calculated according to the limit index parameter of the strange land copy
It is delayed when the network transfer delay and strange land IO of the strange land copy of the corresponding request block of request, can be by pre-
The formula two put calculates the theoretical network propagation delay time and strange land IO time delays of any block size, and this is preset
The input of formula two is test block size, tests time delay, the network bandwidth and block size;It is output as theoretical net
The network propagation delay time network transfer delay of above-mentioned strange land copy (be) and theory IO time delays (as strange land
IO time delays);Wherein input parameter test block size, test time delay, the network bandwidth these three parameters are bases
Actual test result is obtained, then it is considered that they are a constants.I.e. in formula two, input is appointed
What block size, is output as theoretical network propagation delay time and theoretical strange land IO time delays;So as to calculate net to be follow-up
Network congestion time delay and strange land IO congestions time delay provide calculation basis.
Specifically, refer to according to the IO pressure conditions historical data and the limit of local replica on local replica
When mark parameter calculates the IO that the corresponding request block of the current read request of the virtual machine is handled on local replica
Delay, can calculate the IO time delays for obtaining handling on local replica, the formula three according to preset formula three
Input be historical data (average block, average delay), the theoretical time delay of historical data (utilize formula one
Calculate average block theoretical time delay) and current read request block size;It is output as the local of current read request
IO time delays on copy.
Specifically, refer to according to the IO pressure conditions historical data and the limit of strange land copy on the copy of strange land
Mark parameter extends to be delayed during the copy IO congestions of strange land when calculating current network congestion, can be according to preset public affairs
Formula four is obtained, the input of the formula four is the historical data (average block, average delay) on the copy of strange land,
The theoretical time delay of historical data is (when calculating theoretical network time delay and the strange land IO of average block using formula two
Prolong) and current read request block size;When being output as current network congestion time delay and strange land copy IO congestions
Prolong.
Further, because the quantity of strange land copy can be configured according to actual needs, when having two
It is individual or two with strange land copy when, in order to improve read IO performance, strange land copy rotation can be carried out.
Specifically, based on above-described embodiment, in the present embodiment, above-mentioned steps S40 includes:
When the strange land copy is two or more, then the read request of the virtual machine is rotated into transmission successively
To each strange land copy.
In the present embodiment, rotated successively for many strange land copies, for example, read strange land copy A for the first time, then
Second of reading strange land copy B, third time is read strange land copy C, rotated successively repeatedly.This scheme can be with
The reading IO pressure of scattered list copy, and then shorten the time of processing read request, and then reach that IO is read in lifting
Performance effect.
The present invention also provides a kind of reading I O scheduling device based on super fusion storage, and reference picture 3 is real one
Apply in example, the reading I O scheduling device based on super fusion storage that the present invention is provided includes:
First acquisition module 100, for when virtual machine is run on the node of local replica, obtaining local
Copy limit index parameter;
The reading I O scheduling device based on super fusion storage that the present invention is provided is mainly used in distributed type assemblies
In, it is scheduled for the read request to virtual machine.Specifically, in distributed type assemblies, generally include
Two or more copy.
Super fusion:Hyper-Converged Infrastructure, or referred to as " HCI " refers in same set of list
Not only possess the resources such as calculating, network, storage and server virtualization and technology in element device, and
Also include the elements such as backup software, snapping technique, data de-duplication, online data compression, and cover more
Unit can be got up by network polymerization, realize modular seamless (scale-out) extending transversely,
Form unified resource pool.
On super integrated environment, each node is that memory node is also calculate node, and virtual machine exists
When being run on super integrated environment, it reads IO performances is influenceed by the physical location where copy.
Local replica:It is that computing resource is also storage resource to refer to the same node under super integrated environment,
When the node that virtual machine is run has a copy of corresponding virtual machine image file, the copy is referred to as
Local replica;Conversely, the copy being present on other nodes is referred to as strange land copy or distal end copy (this hair
Referred to as strange land copy in bright embodiment).
Above-mentioned limit index parameter includes IOPS (number for referring to processing I/O Request per second) and maximum throughput
Amount.When non-storage limit index parameter on local replica, the limit index parameter can be tested first,
For example processing 4KB in 1S can be tested by the abilities of the fritter I O read requests for testing local replica
Request block number so as to obtaining IOPS indexs;Can also be by testing asking for 128KB or 1MB
The throughput data of block is asked to measure the maximum throughput of local replica.
Determining module 200, for determining IO pressure on the local replica according to the limit index parameter
The higher limit of value;
Above-mentioned limit index parameter according to getting can determine the limit pressure of IO on local replica,
Can be directly using the limit pressure as above-mentioned higher limit, can also be above-mentioned according to preset ratio-dependent
Higher limit, such as higher limit can be the 90% of limit pressure.Specific set-up mode can basis
It is actually needed and is configured, does not limit further herein.
First judge module 300, for the current IO pressure values of real-time judge local replica whether more than described
Higher limit;
Output module 400, for when the current IO pressure values of local replica are more than the higher limit, by institute
The read request for stating virtual machine is sent to strange land copy;Described in being less than or equal to when the current IO pressure values of local replica
During higher limit, the read request of the virtual machine is sent to local replica.
In the present embodiment, the current IO of local replica pressure condition can be detected in real time, according to local replica
Current IO pressure conditions determine current IO pressure values.Then, whether the current IO pressure values of real-time judge
More than above-mentioned higher limit.When more than above-mentioned higher limit, it is determined that the IO pressure of local replica it is larger or
The IO limiting pressures of local replica are reached, the read request of virtual machine is now sent to strange land copy, so that
Mitigate the IO pressure of local replica.
When the embodiment of the present invention is run by virtual machine on the node of local replica, local replica pole is obtained
Limit index parameter;The higher limit of IO pressure values on the local replica is determined according to the limit index parameter;
Whether the current IO pressure values of real-time judge local replica are more than the higher limit;If so, then will be described virtual
The read request of machine is sent to strange land copy;If it is not, then the read request of the virtual machine is sent to local pair
This.So as to ensure using IO localizations are read in the case of the IO pressure of local replica is less, when local
Read request is sent to strange land copy in the case that copy IO pressure is larger and carries out read operation, therefore is improved
Read IO performance.
Further, reference picture 4, the present invention also provides the reading I O scheduling device the based on super fusion storage
Two embodiments, the reading I O scheduling device based on super fusion storage that the present embodiment is provided includes:
Second acquisition module 500, the limit index parameter for obtaining strange land copy;
First computing module 600, for calculating described virtual according to the limit index parameter of the strange land copy
The network transfer delay and strange land IO time delays of the strange land copy of the corresponding request block of the current read request of machine;And
According to being calculated the IO pressure conditions historical data on local replica and the limit index parameter of local replica
The IO time delays that the corresponding request block of the current read request of virtual machine is handled on local replica, according to strange land pair
When the limit index parameter of IO pressure conditions historical data and strange land copy in sheet calculates current network congestion
Extend to strange land copy IO congestion time delays;
Second computing module 700, for calculating the network transfer delay, strange land IO time delays, current net
Strange land copy IO congestion time delay sums are extended to during network congestion, and to obtain the current read request of the virtual machine corresponding
The IO time delays that request block is handled on the copy of strange land;
Whether the second judge module 800, the IO time delays for judging to handle on the strange land copy are less than institute
State the IO time delays handled on local replica;If so, then triggering output module 900 by the reading of the virtual machine
Request is sent to strange land copy;Otherwise the output module 900 is triggered to send out the read request of the virtual machine
Deliver to local replica.
In the present embodiment, it is larger that above-mentioned higher limit is generally set, relatively the limiting value of local replica,
When the IO pressure of local replica is larger and not less than above-mentioned higher limit, will easily occur locating on the copy of strange land
The IO time delays of reason are also smaller than the IO time delays handled on the local replica, if still sending read request
To local replica, then the time delay for reading IO will be longer, be unfavorable for improving the performance for reading IO.The present embodiment exists
Execution sends read request to local replica, judge the IO time delays that are handled on the strange land copy with it is described
The relation of the IO time delays handled on local replica, according to the IO time delays handled on the strange land copy with it is described
Whether the relation of the IO time delays handled on local replica, it is determined whether shunted to read request, i.e., will read
Request is sent to strange land copy, is handled for strange land copy.
If it is understood that there is no the pole of the strange land copy of strange land copy in the equipment of storage local replica
Index parameter is limited, strange land can also be tested according to the same mode of the limit index parameter of test local replica
The limit index parameter of copy, so as to obtain the limit index parameter of strange land copy.
Specifically, above-mentioned second computing module specifically for,
IO pressure conditions on detection local replica, form local replica IO pressure condition historical datas in real time;
The local replica IO pressure conditions historical data includes request block size and corresponding time delay;And according to
The limit index parameter that average request block size, mean time on local replica extend to local replica calculates institute
State the IO time delays that the corresponding request block of current read request is handled on local replica.
It is understood that in the present embodiment, above-mentioned current read request is corresponding to ask block in local replica
The IO time delays of upper processing refer to that local replica handles the time spent in current read request is estimated.
Above-mentioned second computing module is specifically additionally operable to,
IO pressure conditions on detection strange land copy, form strange land copy IO pressure condition historical datas in real time;
The strange land copy IO pressure conditions historical data includes request block size and corresponding time delay;And according to
When the limit index parameter that average request block size, mean time extend to local replica calculates current network congestion
Prolong and strange land copy IO congestion time delays.
The embodiment of the present invention, is sentenced by predicting the IO time delays of local replica and the IO time delays of strange land copy
It is disconnected whether read request to be shunted, thus local replica IO pressure not close to before above-mentioned higher limit
Shunting can be started, so as to further improve the performance for reading IO.
Further, because the quantity of strange land copy can be configured according to actual needs, when having two
It is individual or two with strange land copy when, in order to improve read IO performance, strange land copy rotation can be carried out.
Specifically, based on above-described embodiment, in the present embodiment, above-mentioned output module 400 is specifically for working as institute
When stating strange land copy for two or more, then the read request of the virtual machine is rotated into transmission to each strange land successively
On copy.
When the strange land copy is two or more, then the read request of the virtual machine is rotated into transmission successively
To each strange land copy.
In the present embodiment, rotated successively for many strange land copies, for example, read strange land copy A for the first time, then
Second of reading strange land copy B, third time is read strange land copy C, rotated successively repeatedly.This scheme can be with
The reading IO pressure of scattered list copy, and then shorten the time of processing read request, and then reach that IO is read in lifting
Performance effect.
The preferred embodiments of the present invention are these are only, are not intended to limit the scope of the invention, it is every
The equivalent structure or equivalent flow conversion made using description of the invention and accompanying drawing content, or directly or
Connect and be used in other related technical fields, be included within the scope of the present invention.
Claims (10)
1. a kind of reading I O scheduling method based on super fusion storage, it is characterised in that described to be melted based on super
The reading I O scheduling method for closing storage comprises the following steps:
When virtual machine is run on the node of local replica, local replica limit index parameter is obtained;
The higher limit of IO pressure values on the local replica is determined according to the limit index parameter;
Whether the current IO pressure values of real-time judge local replica are more than the higher limit;
If so, then the read request of the virtual machine is sent to strange land copy;
If it is not, then the read request of the virtual machine is sent to local replica.
2. the reading I O scheduling method as claimed in claim 1 based on super fusion storage, it is characterised in that
The read request by the virtual machine, which is sent to strange land copy, to be included:
When the strange land copy is two or more, then the read request of the virtual machine is rotated into transmission successively
To each strange land copy.
3. a kind of reading I O scheduling method based on super fusion storage, it is characterised in that described to be melted based on super
Closing the reading I O scheduling method of storage includes:
When virtual machine is run on the node of local replica, the limit index parameter of strange land copy is obtained;
According to the limit index parameter calculating network propagation delay time and strange land IO time delays of the strange land copy;
And calculated according to the IO pressure conditions historical data on local replica and the limit index parameter of local replica
The IO time delays handled on local replica, according to the IO pressure conditions historical data on the copy of strange land and strange land
The limit index parameter of copy extends to strange land copy IO congestion time delays when calculating current network congestion;It is described
When network transfer delay is the network transmission of the strange land copy of the corresponding request block of the current read request of virtual machine
Prolong;The strange land IO time delays are the strange land IO time delays of the corresponding request block of the current read request of virtual machine;
The IO time delays handled on the local replica are the corresponding request block of the current read request of virtual machine local
The IO time delays handled on copy;
Strange land copy is extended to when calculating the network transfer delay, strange land IO time delays, current network congestion
IO congestion time delay sums, and it is set as the corresponding request block of the current read request of the virtual machine in strange land pair
The IO time delays handled in sheet;
Judge whether the IO time delays handled on the strange land copy are less than the IO handled on the local replica
Time delay;
If so, then the read request of the virtual machine is sent to strange land copy;
If it is not, then the read request of the virtual machine is sent to local replica.
4. the reading I O scheduling method as claimed in claim 3 based on super fusion storage, it is characterised in that
The limit index parameter meter of the IO pressure conditions historical data and local replica according on local replica
Calculating the IO time delays handled on local replica includes:
IO pressure conditions on detection local replica, form local replica IO pressure condition historical datas in real time;
The local replica IO pressure conditions historical data includes request block size and corresponding time delay;
The limit index ginseng of local replica is extended to according to average request block size, the mean time on local replica
Number calculates the IO time delays that the corresponding request block of the current read request is handled on local replica.
5. the reading I O scheduling method as claimed in claim 3 based on super fusion storage, it is characterised in that
The limit index parameter meter of the IO pressure conditions historical data and strange land copy according on the copy of strange land
Extending to strange land copy IO congestion time delays when calculating current network congestion includes:
IO pressure conditions on detection strange land copy, form strange land copy IO pressure condition historical datas in real time;
The strange land copy IO pressure conditions historical data includes request block size and corresponding time delay;
Current net is calculated according to the limit index parameter that average request block size, mean time extend to strange land copy
Network congestion time delay and strange land copy IO congestion time delays.
6. a kind of reading I O scheduling device based on super fusion storage, it is characterised in that described to be melted based on super
Closing the reading I O scheduling device of storage includes:
First acquisition module, it is local secondary for when virtual machine is run on the node of local replica, obtaining
This limit index parameter;
Determining module, for determining IO pressure values on the local replica according to the limit index parameter
Higher limit;
First judge module, for the current IO pressure values of real-time judge local replica whether more than described
Limit value;
Output module, for when the current IO pressure values of local replica are more than the higher limit, by described in
The read request of virtual machine is sent to strange land copy;Described in being less than or equal to when the current IO pressure values of local replica
During higher limit, the read request of the virtual machine is sent to local replica.
7. the reading I O scheduling device as claimed in claim 6 based on super fusion storage, it is characterised in that
The output module is specifically for when the strange land copy is two or more, then by the virtual machine
Read request rotates transmission to each strange land copy successively.
8. a kind of reading I O scheduling device based on super fusion storage, it is characterised in that described to be melted based on super
Closing the reading I O scheduling device of storage also includes:
Second acquisition module, for when virtual machine is run on the node of local replica, obtaining strange land pair
This limit index parameter;
First computing module, during for being transmitted according to the limit index parameter calculating network of the strange land copy
Prolong and strange land IO time delays;And according to the IO pressure conditions historical data on local replica and local replica
Limit index parameter calculates the IO time delays handled on local replica, according to the IO pressure feelings on the copy of strange land
Condition historical data and the limit index parameter of strange land copy extend to strange land copy when calculating current network congestion
IO congestion time delays;The network transfer delay is the strange land of the corresponding request block of the current read request of virtual machine
The network transfer delay of copy;The strange land IO time delays are the current corresponding request of read request of virtual machine
The strange land IO time delays of block;The IO time delays handled on the local replica are the current read request pair of virtual machine
The IO time delays that the request block answered is handled on local replica;
Second computing module, for calculating the network transfer delay, strange land IO time delays, current network
Strange land copy IO congestion time delay sums are extended to during congestion, and are set as the current read request of the virtual machine
The IO time delays that corresponding request block is handled on the copy of strange land;
Second judge module, it is described whether the IO time delays for judging to handle on the strange land copy are less than
The IO time delays handled on local replica;If so, it is described by the void then to trigger the output module execution
The read request of plan machine is sent to the operation of strange land copy;Otherwise the output module execution is triggered described by described in
The read request of virtual machine is sent to local replica operation.
9. the reading I O scheduling device as claimed in claim 8 based on super fusion storage, it is characterised in that
Second computing module specifically for,
IO pressure conditions on detection local replica, form local replica IO pressure condition historical datas in real time;
The local replica IO pressure conditions historical data includes request block size and corresponding time delay;And root
The limit index parameter for extending to local replica according to average request block size, the mean time on local replica is calculated
The corresponding IO time delays for asking block to be handled on local replica of the current read request.
10. the reading I O scheduling device as claimed in claim 8 based on super fusion storage, its feature exists
In, second computing module specifically for,
IO pressure conditions on detection strange land copy, form strange land copy IO pressure condition historical datas in real time;
The strange land copy IO pressure conditions historical data includes request block size and corresponding time delay;And root
The limit index parameter for extending to strange land copy according to average request block size, mean time calculates current network congestion
Time delay and strange land copy IO congestion time delays.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610108881.8A CN107132990B (en) | 2016-02-26 | 2016-02-26 | Read IO scheduling method and device based on super-fusion storage |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610108881.8A CN107132990B (en) | 2016-02-26 | 2016-02-26 | Read IO scheduling method and device based on super-fusion storage |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107132990A true CN107132990A (en) | 2017-09-05 |
CN107132990B CN107132990B (en) | 2021-05-04 |
Family
ID=59721535
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610108881.8A Active CN107132990B (en) | 2016-02-26 | 2016-02-26 | Read IO scheduling method and device based on super-fusion storage |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107132990B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109471847A (en) * | 2018-09-18 | 2019-03-15 | 华中科技大学 | A kind of I/O jamming control method and control system |
CN109951531A (en) * | 2019-02-27 | 2019-06-28 | 广东唯一网络科技有限公司 | Super fusion cloud computing system |
Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101202746A (en) * | 2006-12-15 | 2008-06-18 | 华为技术有限公司 | Method for generating node identifier and method and device of load balancing |
CN101645925A (en) * | 2009-05-15 | 2010-02-10 | 中国科学院声学研究所 | Selection method of neighbor nodes related to physical topology in P2P system |
US20130055260A1 (en) * | 2011-08-24 | 2013-02-28 | Radware, Ltd. | Techniques for workload balancing among a plurality of physical machines |
CN103095804A (en) * | 2011-12-13 | 2013-05-08 | 微软公司 | Load Balancing In Cluster Storage Systems |
CN103559077A (en) * | 2013-11-08 | 2014-02-05 | 北京华胜天成科技股份有限公司 | Optimized automatic transfer method and system for virtual machines |
CN103577115A (en) * | 2012-07-31 | 2014-02-12 | 华为技术有限公司 | Data configuration processing method and data configuration processing device |
CN103605574A (en) * | 2013-11-18 | 2014-02-26 | 浪潮(北京)电子信息产业有限公司 | Virtual machine resource scheduling method and system for server clusters |
CN104580019A (en) * | 2014-12-26 | 2015-04-29 | 小米科技有限责任公司 | Network service supplying method and device |
CN104639645A (en) * | 2015-02-12 | 2015-05-20 | 广州神马移动信息科技有限公司 | Method and device for balancing network load, and cluster service system |
-
2016
- 2016-02-26 CN CN201610108881.8A patent/CN107132990B/en active Active
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101202746A (en) * | 2006-12-15 | 2008-06-18 | 华为技术有限公司 | Method for generating node identifier and method and device of load balancing |
CN101645925A (en) * | 2009-05-15 | 2010-02-10 | 中国科学院声学研究所 | Selection method of neighbor nodes related to physical topology in P2P system |
US20130055260A1 (en) * | 2011-08-24 | 2013-02-28 | Radware, Ltd. | Techniques for workload balancing among a plurality of physical machines |
CN103095804A (en) * | 2011-12-13 | 2013-05-08 | 微软公司 | Load Balancing In Cluster Storage Systems |
CN103577115A (en) * | 2012-07-31 | 2014-02-12 | 华为技术有限公司 | Data configuration processing method and data configuration processing device |
CN103559077A (en) * | 2013-11-08 | 2014-02-05 | 北京华胜天成科技股份有限公司 | Optimized automatic transfer method and system for virtual machines |
CN103605574A (en) * | 2013-11-18 | 2014-02-26 | 浪潮(北京)电子信息产业有限公司 | Virtual machine resource scheduling method and system for server clusters |
CN104580019A (en) * | 2014-12-26 | 2015-04-29 | 小米科技有限责任公司 | Network service supplying method and device |
CN104639645A (en) * | 2015-02-12 | 2015-05-20 | 广州神马移动信息科技有限公司 | Method and device for balancing network load, and cluster service system |
Non-Patent Citations (1)
Title |
---|
刘胜等: "《片上网络延时差异对存储系统公平性的影响及对策》", 《计算机学报》 * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109471847A (en) * | 2018-09-18 | 2019-03-15 | 华中科技大学 | A kind of I/O jamming control method and control system |
CN109471847B (en) * | 2018-09-18 | 2020-06-09 | 华中科技大学 | I/O congestion control method and control system |
CN109951531A (en) * | 2019-02-27 | 2019-06-28 | 广东唯一网络科技有限公司 | Super fusion cloud computing system |
CN109951531B (en) * | 2019-02-27 | 2021-05-07 | 广东唯一网络科技有限公司 | Super-fusion cloud computing system |
Also Published As
Publication number | Publication date |
---|---|
CN107132990B (en) | 2021-05-04 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Neto et al. | ULOOF: A user level online offloading framework for mobile edge computing | |
CN106470133B (en) | System pressure testing method and device | |
CN103150238B (en) | A kind of service request processing method, simulated performance method of testing and device | |
CN104184813B (en) | The load-balancing method and relevant device and group system of virtual machine | |
US7315807B1 (en) | System and methods for storage area network simulation | |
CN105446827B (en) | Date storage method and equipment when a kind of database failure | |
US10277475B2 (en) | System and method for dynamically testing networked target systems | |
CN103984619A (en) | Performance monitoring method of network storage system | |
CN109586952A (en) | Method of server expansion, device | |
CN108075934A (en) | A kind of network quality monitoring method, apparatus and system | |
CN105577414A (en) | Elastically telescopic processing method, device and system for VM (Virtual Machine) resources | |
CN109218141A (en) | A kind of malfunctioning node detection method and relevant apparatus | |
CN108200218B (en) | Method and device for realizing load balance and electronic equipment | |
CN110298185A (en) | Model training method and device, electronic equipment and storage medium | |
CN113411221A (en) | Power communication network fault simulation verification method, device, equipment and storage medium | |
CN109491788A (en) | A kind of virtual platform implementation of load balancing and device | |
CN108052444A (en) | A kind of method and apparatus of performance test for mobile application | |
CN107992386A (en) | A kind of router memory method for testing pressure and system | |
CN109324962A (en) | A kind of test method and equipment of the application program based on cloud internet of things technology | |
CN107132990A (en) | Reading I O scheduling method and device based on super fusion storage | |
CN113676534A (en) | AI algorithm flow and service all-in-one machine based on edge calculation | |
Hanafy et al. | A new infrastructure elasticity control algorithm for containerized cloud | |
CN108632823A (en) | A kind of method for switching network, terminal and computer storage media | |
CN106060048A (en) | Network resource access method and network resource access device | |
CN107453786B (en) | Method and device for establishing electric power communication network model |
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 | ||
CB02 | Change of applicant information | ||
CB02 | Change of applicant information |
Address after: 518000 the first floor of A1 building, Nanshan Zhiyuan 1001, Nanshan District Xue Yuan Avenue, Shenzhen, Guangdong. Applicant after: SANGFOR TECHNOLOGIES Inc. Address before: 518052 the first floor of A1 building, Nanshan Zhiyuan 1001, Nanshan District Xue Yuan Avenue, Shenzhen, Guangdong. Applicant before: Sangfor Technologies Co.,Ltd. |
|
GR01 | Patent grant | ||
GR01 | Patent grant |