CN106155810B - The input/output scheduling device of workload-aware in software definition mixing stocking system - Google Patents
The input/output scheduling device of workload-aware in software definition mixing stocking system Download PDFInfo
- Publication number
- CN106155810B CN106155810B CN201510193350.9A CN201510193350A CN106155810B CN 106155810 B CN106155810 B CN 106155810B CN 201510193350 A CN201510193350 A CN 201510193350A CN 106155810 B CN106155810 B CN 106155810B
- Authority
- CN
- China
- Prior art keywords
- workload
- queue
- value
- input
- delay
- 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 invention discloses a kind of input/output scheduling devices of workload-aware in software definition mixing stocking system comprising a queue management module, a work load characteristic database, a flow monitoring module and a flow model module.The queue management module further comprises a request receiving submodule, a request control submodule and a request scheduling submodule.Request and respective queue by queue management module to access hard disk and solid state hard disk.To which the performance of software definition mixing storage can be improved.Meanwhile the scheduler can be used for the request from multiple workloads simultaneously, to reach the performance requirement of each workload.
Description
Technical field
The present invention relates to a kind of input/output scheduling device, more particularly to working in a kind of software definition mixing stocking system
The input/output scheduling device of Load-aware.
Background technique
Computer operating system determines that block storage input/output operation will in what order using input/output scheduling device
Submit to storage roll.According to different purposes, the target of an input/output scheduling device may spend in hard disk with minimum and search
The target-seeking time pays the utmost attention to a part of input/output request of processing, to the program part Disk bandwidth being each currently running
Share, and/or guarantee that certain input/output requests start to execute before a specific cut-off time point.For example, Linux
Cut-off time point scheduler in kernel for guaranteeing a beginning service time for request, cut by forcing one by its function
Only time point prevents request hunger phenomenon (starvation) into all input/output operations and reaches.Thus, by making
With input/output queue out of the ordinary, ends the reading of time point scheduler preference and be excessively written.It runs database workload good
It is good.Another example is complete Fair Queue (Complete Fair Queuing) scheduler.Complete fair queuing scheduler puts
The synchronization request submitted by program is set to many by the queue of program, and then to each queue assignment timeslice, with access
The disk.To which complete fair queuing scheduler is suitable for continuously reading video or audio stream, and the work from general host
Load.
Above-mentioned scheduler is proposed based on the performance for improving hard disk or the stocking system being made of hard disk.It is below
The characteristic of some hard disc storage systems: firstly, multiple input/output request (reading and write-in the two) may be merged into a list
In one request, and handled under the movement that the single request reads write head at one.Therefore, mobile time of write head is read
Number can be reduced, to increase the circulation of hard disk.Second, input/output request is classified, and reads write head by reduction hard disk
Contact is mobile to improve search time.Based on those features, it is queued up to wait the input/output request of processing in the future can quilt
Merge and classifies.When using different schedulers, the workload with different characteristic can be handled with preferable performance.
A kind of mainstream of stocking system, is especially used in cloud system, be using mixed solid state hard disk and hard disk, rather than
Only it is made of hard disk, thus newest scheduler possibly can not reach themselves and require when they are attached to stocking system
The target to be reached.There is solid state hard disk the distinguishing feature different from hard disk to be described as follows.Firstly, solid state hard disk does not need to merge
And the input/output request of classification solid state hard disk, which imply need not merge and the time needed for classification.Input/output request
It should be able to be transmitted in solid state hard disk with prestissimo.Second, because many modern solid-state hard disks have multichannel, solid state hard disk is defeated
Enter/export the possible parallelization processing of request, and the multichannel can accommodate multiple input/output requests simultaneously.If using the tune
The case where stocking system of degree device is a kind of software definition mixing stocking system, to handle scheduler demand can be more complicated.?
In the presence of consideration solid state hard disk, it is necessary to change and improve existing scheduler
In addition, further investigation, the discharge characteristic of workload is another important problem.Any workload may
With certain characteristics different from other workloads, which may be input/output mode (sequence or random), reads/write
Enter ratio, solid state hard disk cache hit etc..For example, in-system decryption (On-Line Transaction
Processing) workload of database has stochastic inputs/output mode, read/write ratio greater than 1 and lesser
Store block size；And the workload of MongoDB have sequentially input/output mode, read/write ratio be less than 1 and larger
Storage block size.If two kinds of workloads execute under identical mixing stocking system, the scheduler developed recently cannot
Reach the performance requirement in service level agreement (Service Level Agreements) to the two, can at least there is noisy neighbour
Problem (Noisy Neighbor Problem) is occupied, those workloads are influenced.
To above-mentioned requirements, existing some technologies provide relevant solution, and wherein an example is disclosed in United States Patent (USP)
No. 8,756,369.In that patent, a stocking system includes a command catalog device, hard for an at least solid-state for determining
If a target storage equipment of disk order and a hard disk order, the solid state hard disk order are that calibration is consolidated to the one of the stocking system
State hard disc storage equipment then places the order into a solid state hard disk preparation queue, and if the hard disk order is to demarcate to arrive to be somebody's turn to do
One hard disc storage equipment of stocking system then places the hard disk order into a hard disk preparation queue.The stocking system also includes
One solid state hard disk preparation queue, the solid state hard disk order and a hard disk to queue calibration to the solid state hard disk storage facilities are pre-
The hard disk order of the hard disc storage equipment is arrived in standby queue to queue calibration.Meanwhile one command scheduler from the preparation queue
Hard disk and solid state hard disk order are taken out, and places those orders into a command processor.One based on the corresponding special order
The availability rank of one processing queue of target device, the command scheduler are placed one special from its respective preparation queue
(hard disk or solid state hard disk) order, into the command processor.So, which gives the processing queue those storages
Order.
Stocking system provided by U.S. Patent No. 8,756,369 is distinguished hard disk order (input/output request) and solid
State hard disk order, it facilitates the hardware operation to a workload.However, if being applied to more workloads, the storage system
System possibly can not be operated as it is default.On the other hand, it is run on for the workload on the stocking system for various,
There is no suitable mode to coordinate the request from each workload, so that each workload is able to satisfy service level agreement
Or service quality (Quality of Service) demand.Therefore, need one kind can in software definition mixing stocking system
The input/output scheduling device of workload is perceived, more than which can be used to solve the problems, such as.
Summary of the invention
In order to meet the demand, the present invention provides a kind of software definitions to mix workload in storage (SDHS) system
The input/output scheduling device of perception, software definition mixing storage include at least a hard disk and a solid state hard disk.The software definition
The input/output scheduling device for mixing workload-aware in stocking system includes a queue management module, a work load characteristic number
According to library and a flow monitoring module, the queue management module is to manage queue, read requests and write request, the queue management
Module includes a request receiving submodule, a request control submodule and a request scheduling submodule, wherein the request receives son
Module is temporarily to store those read requests and write request；The request control submodule to create workload queue,
It is arrived with those read requests are arranged with write request according to a scheduler configuration feature with those workload queues of dynamic configuration
In those workload queues；The request scheduling submodule is dispatched to create device queue and comes from the workload queue
Each read requests or write request into a specific device queue；The work load characteristic database is to store work
The characteristic of load is for access；The traffic monitoring module is to monitor and persistently record the one of the software definition mixing stocking system
The value of performance parameter, and the value for providing those performance parameters gives the request control submodule.
The scheduler configuration feature is to each workload queue, based on those of work load characteristic database offer
The value of characteristic and those performance parameters received calculates a queue depth and a waiting time, mixed to adjust the software definition
Close a performance guarantee value and a property regulation value that the performance parameter value of stocking system will fall within setting to the performance parameter in future
Between.
The input/output scheduling device of workload-aware, can also further wrap in above-mentioned software definition mixing stocking system
Containing a flow model module, storage flow of the discharge model module to model those requests from the workload,
And the prediction that those characteristics are provided on following specific time point stores flow.
Preferably, those characteristics are read/write ratio, merge ratio, solid state hard disk hit rate and storage block size.It should
Performance parameter is input-output operation number per second, circulation, delay time or the combination of aforementioned three.The performance guarantee value with
Property regulation value by the workload service level agreement (Service Level Agreement) or service quality
(Quality of Service) demand is defined.Each workload queue is classified as profound, medium level or shallow-layer
Secondary, each waiting time is classified as long delay, medium delay or short delay, wherein the queue of profound workload queue is deep
The queue depth for spending more medium level workload queue accommodates more read requests or write request；Medium level workload
The queue depth of queue accommodates more read requests or write request compared with the queue depth of shallow hierarchy workload queue；Block storage
Size is a median size；Long delay is longer than medium delay；Medium delay is longer than short delay.
In an example, if the value of the received input-output operation number per second or circulation is close to respective performance
Regulated value, the received delay time value are near or below performance guarantee value, the read/write ratio is greater than or equal to 1, and stores up
Counterfoil size is greater than or equal to the median size, then the queue depth of the workload queue is set as medium level, and to every
The waiting time of one workload queue is set as short delay.
In another example, if the value of the received input-output operation number per second or circulation is close to respective property
It is adjustable value, the received delay time value is near or below and can guarantee that value, the read/write ratio are greater than or equal to 1, and is somebody's turn to do
It stores block size and is less than the median size, then the queue depth of the workload queue is set as profound, and to each work
The waiting time of load queue is set to medium delay.
In another example, if the value of the received input-output operation number per second or circulation is close to respective property
Be adjustable value, the received delay time value is near or below performance guarantee value, and the read/write ratio is less than 1, then the work
The queue depth for making load queue is set as medium level, and is set as short to the waiting time of each workload queue and prolongs
When.
In another example, if the value of the received input-output operation number per second or circulation is near or below respectively
From performance guarantee value, the received delay time value is near or below performance guarantee value, the read/write ratio is greater than or waits
In 1, and the storage block size is greater than or equal to the median size, then the queue depth of the workload queue is set as shallow-layer
It is secondary, and short delay is set as to the waiting time of each workload queue.
In another example, if the value of the received input-output operation number per second or circulation is near or below respectively
From performance guarantee value, the received delay time value is near or below performance guarantee value, the read/write ratio is greater than or waits
In 1, and the storage block size is less than the median size, then the queue depth of the workload queue is set as medium level, and
Medium delay is set as to the waiting time of each workload queue.
In another example, if the value of the received input-output operation number per second or circulation is near or below respectively
From performance guarantee value, the received delay time value be near or below performance guarantee value, and the read/write ratio is less than 1,
Then the queue depth of the workload queue is set as shallow hierarchy, and is set as short to the waiting time of each workload queue
Delay.
In another example, if the value of the received input-output operation number per second or circulation is close to respective property
It is adjustable value, which keeps off or is greater than or equal to 1 lower than performance guarantee value, the read/write ratio,
And the storage block size is greater than or equal to the median size, then the queue depth of the workload queue is set as medium level,
And medium delay is set as to the waiting time of each workload queue.
In another example, if the value of the received input-output operation number per second or circulation is close to respective property
It is adjustable value, which keeps off or is greater than or equal to 1 lower than performance guarantee value, the read/write ratio,
And the storage block size is less than the median size, then the queue depth of the workload queue is set as profound, and to each
The waiting time of workload queue is set as long delay.
In another example, if the value of the received input-output operation number per second or circulation is close to respective property
It is adjustable value, which keeps off or is lower than performance guarantee value, the read/write ratio less than 1, and the storage
Counterfoil size is greater than or equal to the median size, then the queue depth of the workload queue is set as medium level, and to every
The waiting time of one workload queue is set as short delay.
In another example, if the value of the received input-output operation number per second or circulation is close to respective property
It is adjustable value, which keeps off or is lower than performance guarantee value, the read/write ratio less than 1, and the storage
Counterfoil size is less than the median size, then the queue depth of the workload queue is set as medium level, and to each work
The waiting time of load queue is set as medium delay.
In another example, if the value of the received input-output operation number per second or circulation is near or below respectively
From performance guarantee value, the received delay time value keep off or lower than performance guarantee value, and the storage block size be greater than or
Equal to the median size, then the queue depth of the workload queue is set as shallow hierarchy, and to each workload queue
Waiting time is set as short delay.
In another example, if the value of the received input-output operation number per second or circulation is near or below respectively
From performance guarantee value, the received delay time value keep off or lower than performance guarantee value, and be less than should for the storage block size
Median size, then the queue depth of the workload queue is set as shallow hierarchy, and when to the waiting of each workload queue
Between be set as medium delay.
In another example, if the solid state hard disk hit rate increases, the queue depth of the workload queue is maintained
It is identical or become more shallow, and identical is maintained to the waiting time of each workload queue or is become shorter, otherwise the work is negative
The queue depth for carrying queue remains identical or becomes deeper, and remains identical to the waiting time of each workload queue or become
It obtains longer.
In another example, if the merging ratio increases, the queue depth of the workload queue maintain it is identical or
Become more shallow, and identical is maintained to the waiting time of each workload queue or is become shorter, the otherwise workload queue
Queue depth remain identical or become deeper, and identical is maintained to the waiting time of each workload queue or become compared with
It is long.
Preferably, the median size is 8KB.
Detailed description of the invention
Fig. 1 is the storage system that invention software defines the input/output scheduling device of workload-aware in mixing stocking system
The structural schematic diagram of system；
Fig. 2 is the operation chart of request a control submodule and a request scheduling submodule；
Fig. 3 is the complete situation of input/output scheduling device workload.
Description of symbols: 1- remote user；5- cloud storage system；10- scheduler；50- host；55- file system；
60- device driver module；61- solid state hard disk；62- solid state hard disk；63- hard disk；64- hard disk；100- queue management module；
101- request receiving submodule；102- requests control submodule；103- request scheduling submodule；110- work load characteristic data
Library；120- traffic monitoring module；130- discharge model module；W1- workload queue group；W2- workload queue group；
Wn- workload queue group；QW1- queue depth；QW2- queue depth；QWn- queue depth；The first solid state hard disk of QS1- is set
Standby queue；QS2- the second solid-state hard disc equipment queue；QH1- the first hard disc apparatus queue；QH2- the second hard disc apparatus queue.
Specific embodiment
The present invention will be more specifically described referring to following embodiments.
As shown in Figure 1, it describes one embodiment of the invention.The cloud storage system 5 being shown in Fig. 1 is a software
Definition mixing this software definition of stocking system mixing storage, can be based on quick creation, the shifting of principle so that beyond the clouds under framework
Remove and manage stocking system.The storage that software definition mixing storage manages software from it usually in a manner of sas storage virtualization
It is separated in hardware.It is stored different from general software definition, software definition mixing stocking system includes more than one magnetic disc type
State.It is usually made of two kinds of disks, such as hard disk and solid state hard disk, for meeting the needs of different characteristic workload.From
And cloud storage system 5 can also apply to many workloads.These workloads can be an in-system decryption data
Library, video streaming services, virtual desktop architecture environment, e-mail server, backup server or archive server.It is all
It is stored as service or the framework of storage element using software definition mixing, is all the workload according to spirit of the invention.
Cloud storage system 5 receives the read or write request for carrying out many users in automatic network.It is of the invention in order to simplify
Illustrate, only a remote user 1 is for representing all users.Substantially, the input/output operation of cloud storage system 5 is to institute
Have for user all, but in different situations, processing order of priority and response delay for each workload
It is different.When reading the present embodiment, it is contemplated that issuing read/write there are many remote user 1 requests to cloud storage system
5, and wait-for-response simultaneously.
The input/output scheduler 10 of one workload-aware is installed in cloud storage system 5, it is weight of the invention
Want part.Because the workload in request source can be distinguished, scheduler 10 is to workload-aware.Scheduler 10 includes one
Queue management module 100, a work load characteristic database 110, a flow monitoring module 120 and a flow model module 130.
Each will be in being hereinafter described in detail.It should be noted that input/output scheduler 10 can be implemented in cloud in the form of software
On one host 50 of stocking system 5, it is also possible to the computer hardware to work for cloud storage system 5 in host 50.Alternatively, defeated
Enter/output scheduler 10 can part by hardware realization, and rest part is by software-driven, this is not limited by the present invention.
The major function of queue management module 100 is that management queue, read requests and write request, reading above-mentioned are asked
Seek the file system 55 with write request from host 50.Queue management module 100 includes three important submodules.They
It is a request receiving submodule 101, one request control submodule 102 and a request scheduling submodule 103 respectively.Request receives son
Module 101 can temporarily store read requests and write request from file system 55.It will be negative according to work after the request
It carries and classifies, and be then arranged into the write-in queue of a workload queue group and read in queue.The write-in queue is certain
One specific workload sequence write request, the reading queue are that identical workload sequentially keeps read requests.It will be again
Workload there are many emphasizing while cloud storage system 5 is used, equally also there are many workload queue group W1 to Wn
It is shown in Fig. 2.
Request control submodule 102 to be used to request to each workload, creation workload queue (write-in queue with
Read queue).After workload queue is set up, request control submodule 102 can configure function according further to a scheduler
Can, dynamic configuration workload queue.Depth that the configuration feature of the scheduler can operate to determine workload queue and every
The time span of the chances to be combined such as one request, this will be described in detail below.Therefore, request control submodule 102 arranges
Read requests are to write request into relevant workload queue.
Request scheduling submodule 103 is responsible for creation device queue.Each device queue includes to particular device, i.e. solid-state is hard
Disk 61,62 or hard disk 63,64, request (read and/or write-in).The request may be from a workload queue group or certain
Workload queue group.Request scheduling submodule 103 will dispatch each read requests from workload queue or write-in
It requests into a specific device queue.If the request of other queues has been handled before a request, which will
It can then be handled.In the design of certain hosts 50, scheduling can trigger corresponding storage via a device driver module 60 and set
Standby driver and reach.The request scheduling submodule 103 provides a kind of excellent function that other schedulers can not be reached, it will
Request and the request to solid state hard disk to hard disk separate.To, the request to solid state hard disk can be executed directly, it is not necessary to it is other
Request to hard disk waits in the queue, therefore the performance of cloud storage system 5 can be promoted.
Work load characteristic database 110 stores the characteristic of each workload, so that access is used.Those characteristics are scheduling
Device configuration feature is used to determine the factor of queue depth and waiting time.According to the present invention, those characteristics are read/write ratios
Rate, solid state hard disk hit rate merge ratio and storage block size.Read/write ratio is workload read requests quantity pair
The ratio of write request quantity.For a workload, it usually has a specific use pattern, and read requests quantity is more
In write request quantity or write request quantity more than read requests quantity.To certain special circumstances, a certain workload, such as
One backup server may have write request but no read requests.
Solid state hard disk hit rate refers to the frequency that the solid state hard disk in cloud storage system 5 is accessed, no matter those are asked
Seeking Truth is from which workload.The quantity or space for being used in the solid state hard disk of each workload may be negative according to work
It carries service condition and becomes.To which the solid state hard disk hit rate for each workload is becoming at any time.If the solid state hard disk is ordered
Middle rate increases, then the queue depth of the workload queue may remain identical or become more shallow, and to each workload team
The waiting time of column may remain identical or become shorter.Conversely, the queue depth of the workload queue may remain identical
Or become deeper, and identical may be maintained to the waiting time of each workload queue or become longer.Maintain or change team
Column depth or waiting time are based on depending on increase/reduction degree.Threshold value between maintaining and changing, and change the amount of degree
Value can just be set before the online running of stocking system 5 beyond the clouds.Method above-mentioned is a kind of criterion for laying down a regulation, "
Shallowly ", " short ", " depth ", and " length " equal quantificational descriptions will then illustrate.The use of comparison is to illustrate its increasing for existing rank
It adds deduct few.
Read/write ratio is a kind of pointer for expressing read/write mode.Merge the access that ratio only calculates hard disk,
It is a reading write head between identical movement, the ratio with the request of other requests merging treatment (change of processing priority)
Example.For example, when a definite-time starts (ta), there are 10 hard disk read requests in the queue.At the end of the time limit
(tb), request, which merges, causes the number of requests in queue to become 7, which is 0.3 in tb, its public affairs from below
Formula:
Wherein tb>ta
R (t) indicates that in time t, request obtains the function of quantity in one queue.It is higher to merge ratio, the circulation of hard disk
Measure bigger (this example, which is shown, is greater than 30%).However, if the enough length of the request etc. and obtain maximum circulation, delay time
It may become longer, and longer delay time can make user's experience be deteriorated.For a workload, its merging ratio
Rate may also change with the time.If merging ratio to increase, the queue depth of the workload queue may remain identical
Or become shallower, and identical may be maintained to the waiting time of each workload queue or become shorter.Otherwise the work is negative
The queue depth for carrying queue may remain identical or become deeper, and may maintain to the waiting time of each workload queue
It is identical or become longer.Similarly, " shallow ", " short ", " depth ", and " length " equal quantificational descriptions will then illustrate.The use of comparison
It is to illustrate it for existing rank and increase or decrease.
Storage block size is the basic storage element space of storage device, and hard disk or solid state hard disk have, it is from 4KB or more
It is small to 16MB or more than be likely to.In general, median size uses more typical with 8KB.With storage facilities science and technology
Evolution, the median size will will increase.For illustrative purposes, in the present specification, 8KB is set as the median size, however
This is not intended to limit the present invention.For different storage block size and situation, scheduler configuration feature is used to determine queue depth
It is different from the strategy of waiting time, it will be described below.
Traffic monitoring module 120 can monitor and persistently record the value of a performance parameter of cloud storage system 5, it can also be mentioned
Request control submodule 102 is given for the value of the performance parameter.Performance parameter can be input-output operation number per second, circulation
Amount, delay time or the combination of aforementioned three.Discharge model module 130 is used to model the storage stream from workload request
It measures and on following specific time point, provides the storage flow for predicting certain characteristics.The data used are modeled to come from
Traffic monitoring module 120.Any suitable method, algorithm or module can be applied, it is preferred to use by identical inventor in beauty
A kind of storage provided in state's patent application case the 14/290th, 533 sets discharge model, and identical skill can be obtained by this application
The collective reference of art.The storage flow of modelling from workload request, and the predicted value of storage flow, can provide and ask
Control submodule 102 is asked to refer to, in case future workload queue and the configuration of waiting time are used.
Hereinafter, the running of scheduler configuration feature will be illustrated with the examples of all situations.Scheduler configuration feature is based on
The characteristic and received performance parameter value that work load characteristic database provides calculate queue depth and to each workload team
The waiting time of column.Therefore, scheduler configuration feature can adjust the performance parameter value of cloud storage system 5, to fall in future
Between a performance guarantee value and a property regulation value of the setting to the performance parameter.Performance guarantee value and property regulation above-mentioned
Value is defined by the service level agreement or QoS requirement of workload.Each workload queue is classified as deep layer
Secondary, medium level or shallow hierarchy.Each waiting time is classified as long delay, medium delay or short delay.
According to spirit of the invention, for the workload queue of each classification, as running when be classified etc.
To the time, not absolute delimiter.One guide policy is that the queue depth of profound workload queue should be than medium layer
The queue depth of task load queue accommodates more read requests or write request；The team of medium level workload queue
Column depth accommodates more read requests or write request than the queue depth of shallow hierarchy workload queue；The size of block storage
For median size.Similarly, the more medium delay of long delay is long, the medium shorter delay length of delay.
When workload situation there are many as described above, influences scheduler configuration feature decision queue depth and waits
Between, all situations disclose as follows.In one embodiment, cloud storage system 5 supports video flowing, in-system decryption database
With mail server.These workloads are merely to illustrate, and do not limit application of the invention, different from the work of above-mentioned three
Load can use.Video flowing is a kind of sequence inputting/output kenel, and the read requests quantity having is compared with write request quantity
It is more, it needs to store block size more than or equal to 8K, generates the value of input-output operation number per second or circulation close to respective
Property regulation value, and the delay time having is near or below the performance guarantee value of delay time.As shown in Fig. 2, workload
A queue depth QW1 of workload queue is set as medium level (5) in queue group W1, to each workload queue
Waiting time is set as short delay (20ms).Request scheduling submodule 103 arranges the request from workload queue group W1
To the one first solid-state hard disc equipment queue QS1 for solid state hard disk 61 and for one first hard disc apparatus queue of hard disk 63
QH1。
In-system decryption database is stochastic inputs/output kenel, read/write request or reading with identical quantity
It takes number of requests more than write request quantity, needs to store block size more than or equal to 8K, generate input-output operation per second
Several or circulation value is near or below respective performance guarantee value, and the delay time having is near or below delay time
Performance guarantee value.A queue depth QW2 of workload queue is set as shallow hierarchy (2) in workload queue group W2, right
The waiting time of each workload queue is set as short delay (20ms).Request scheduling submodule 103 arranges negative from work
The request of queue group W2 is carried to the first solid-state hard disc equipment queue QS1 for solid state hard disk 61 and for solid state hard disk 62
One second solid-state hard disc equipment queue QS2.
Mail server is stochastic inputs/output kenel, and the read requests quantity having is more compared with write request quantity, (or
It is even identical), it needs to store block size less than 8K, generates the value of input-output operation number per second or circulation close to respectively
Property regulation value, and the delay time having be not close to or lower than delay time performance guarantee value.Workload queue
A queue depth QWn of workload queue is set as profound (7) in group Wn, when to the waiting of each workload queue
Between be set as long delay (100ms).Request scheduling submodule 103 arranges the request from workload queue group Wn to being used for
First hard disc apparatus queue QH1 of hard disk 63 and the second hard disc apparatus queue QH2 for hard disk 64.
The complete combination table of workload situation is listed in Fig. 3., it is clear that video flowing, in-system decryption database,
And mail server is distinctly No. 1, No. 5 and No. 10 situation.Remaining situation is described as follows.
For No. 2 situation, if the value of received input-output operation number per second or circulation is close to respective
Property regulation value, received delay time value are near or below performance guarantee value, read/write ratio is greater than or equal to 1, and stores up
Counterfoil size is less than median size, then the queue depth of the workload queue is set as profound, and to each workload
The waiting time of queue is set as medium delay.
For No. 3 situation, if the value of received input-output operation number per second or circulation is close to respective
Property regulation value, received delay time value are near or below performance guarantee value, read/write ratio less than 1, and block storage is big
Small to be greater than or equal to median size, then the queue depth of the workload queue is set as medium level, and negative to each work
The waiting time for carrying queue is set as short delay.
For No. 4 situation, if the value of received input-output operation number per second or circulation is close to respective
Property regulation value, received delay time value are near or below performance guarantee value, read/write ratio less than 1, and block storage is big
It is less than median size, then the queue depth of the workload queue is set as medium level, and to each workload queue
Waiting time be set as short delay.
For No. 6 situation, if the value of received input-output operation number per second or circulation is near or below
Respective performance guarantee value, received delay time value are near or below performance guarantee value, read/write ratio is greater than or equal to
1, and store block size and be less than median size, then the queue depth of the workload queue is set as medium level, and to each
The waiting time of workload queue is set as medium delay.
For No. 7 situation, if the value of received input-output operation number per second or circulation is near or below
Respective performance guarantee value, received delay time value are near or below performance guarantee value, read/write ratio less than 1, and store up
Counterfoil size is greater than or equal to median size, then the queue depth of the workload queue is set as shallow hierarchy, and to each work
The waiting time for making load queue is set as short delay.
For No. 8 situation, if the value of received input-output operation number per second or circulation is near or below
Respective performance guarantee value, received delay time value are near or below performance guarantee value, read/write ratio less than 1, and store up
Counterfoil size is less than median size, then the queue depth of the workload queue is set as shallow hierarchy, and to each workload
The waiting time of queue is set as short delay.
For No. 9 situation, if the value of received input-output operation number per second or circulation is close to respective
Property regulation value, received delay time value are kept off or lower than performance guarantee value, read/write ratio more than or equal to 1, and
It stores block size and is greater than or equal to median size, then the queue depth of the workload queue is set as medium level, and to every
The waiting time of one workload queue is set as medium delay.
For o.11 situation, if the value of received input-output operation number per second or circulation is close to respective
Property regulation value, received delay time value are kept off or lower than performance guarantee value, read/write ratio less than 1, and block storage
Size is greater than or equal to median size, then the queue depth of the workload queue is set as medium level, and to each work
The waiting time of load queue is set as short delay.
For No. 12 situation, if the value of received input-output operation number per second or circulation is close to respective
Property regulation value, received delay time value are kept off or lower than performance guarantee value, read/write ratio less than 1, and block storage
Size is less than median size, then the queue depth of the workload queue is set as medium level, and to each workload team
The waiting time of column is set as medium delay.
For No. 13 situation, if the value of received input-output operation number per second or circulation is near or below
Respective performance guarantee value, received delay time value are kept off or are greater than or wait lower than performance guarantee value, read/write ratio
In 1, and storing block size and be greater than or equal to median size, then the queue depth of the workload queue is set as shallow hierarchy, and
Short delay is set as to the waiting time of each workload queue.
For No. 14 situation, if the value of received input-output operation number per second or circulation is near or below
Respective performance guarantee value, received delay time value are kept off or are greater than or wait lower than performance guarantee value, read/write ratio
It in 1, and stores block size and is less than median size, then the queue depth of the workload queue is set as shallow hierarchy, and to each
The waiting time of workload queue is set as medium delay.
For No. 15 situation, if the value of received input-output operation number per second or circulation is near or below
Respective performance guarantee value, received delay time value are kept off or lower than performance guarantee value, read/write ratio less than 1, and
It stores block size and is greater than or equal to median size, then the queue depth of the workload queue is set as shallow hierarchy, and to each
The waiting time of workload queue is set as short delay.
For No. 16 situation, if the value of received input-output operation number per second or circulation is near or below
Respective performance guarantee value, received delay time value are kept off or lower than performance guarantee value, read/write ratio less than 1, and
It stores block size and is less than median size, then the queue depth of the workload queue is set as shallow hierarchy, and negative to each work
The waiting time for carrying queue is set as medium delay.
Although the present invention is disclosed above with embodiment, however, it is not to limit the invention, any affiliated technology neck
Have usually intellectual in domain, without departing from the spirit and scope of the present invention, when can make it is a little change and retouch, therefore this
Subject to the protection scope of invention ought be defined depending on this case scope of the claims.
Claims (21)
1. the input/output scheduling device of workload-aware in a kind of software definition mixing stocking system, software definition mixing storage
Deposit system includes an at least hard disk and a solid state hard disk, which is characterized in that workload in the software definition mixing stocking system
The input/output scheduling device of perception includes a queue management module, a work load characteristic database and a flow monitoring module,
Wherein:
For the queue management module to manage queue, read requests and write request, which includes that a request connects
Receive submodule, a request control submodule and a request scheduling submodule, wherein
The request receiving submodule is temporarily to store those read requests and write request；
The request control submodule is to create workload queue, according to a scheduler configuration feature with those works of dynamic configuration
Make load queue and arranges those read requests and write request into those workload queues；
The request scheduling submodule to create device queue, and dispatch each read requests from the workload queue or
Write request is into a specific device queue；
The work load characteristic database is to store the characteristic of workload for access；
The traffic monitoring module to monitor and persistently record the software definition mixing stocking system a performance parameter value, and
The value for providing those performance parameters gives the request control submodule；
Wherein the scheduler configuration feature is to each workload queue, based on those of work load characteristic database offer
The value of characteristic and those performance parameters received calculates a queue depth and a waiting time, mixed to adjust the software definition
Close a performance guarantee value and a property regulation value that the performance parameter value of stocking system will fall within setting to the performance parameter in future
Between.
2. the input/output scheduling device of workload-aware in software definition mixing stocking system according to claim 1,
It is characterized in that, further including a flow model module, the discharge model module is to model from the workload
The storage flow of those requests, and the prediction storage flow of those characteristics is provided on following specific time point.
3. the input/output scheduling device of workload-aware in software definition mixing stocking system according to claim 1,
It is characterized in that, those characteristics are read/write ratio, merge ratio, solid state hard disk hit rate and storage block size.
4. the input/output scheduling device of workload-aware in software definition mixing stocking system according to claim 3,
It is characterized in that, the performance parameter is input-output operation number per second, circulation, delay time or the combination of aforementioned three.
5. the input/output scheduling device of workload-aware in software definition mixing stocking system according to claim 3,
It is characterized in that, the performance guarantee value is with property regulation value by service level agreement or the QoS requirement institute of the workload
Definition.
6. the input/output scheduling device of workload-aware in software definition mixing stocking system according to claim 4,
It is characterized in that, each workload queue is classified as profound, medium level or shallow hierarchy, each waiting time is classified
For long delay, medium delay or short delay, wherein the more medium level workload of queue depth of profound workload queue
The queue depth of queue accommodates more read requests or write request；The queue depth of medium level workload queue is compared with shallow-layer
The queue depth of task load queue accommodates more read requests or write request；Storage block size is a median size；It is long
Delay is longer than medium delay；Medium delay is longer than short delay.
7. the input/output scheduling device of workload-aware in software definition mixing stocking system according to claim 6,
It is characterized in that, if the value of received input-output operation number per second or circulation close to respective property regulation value, connect
The delay time value of receipts is near or below performance guarantee value, the read/write ratio is greater than or equal to 1, and it is big to store block size
In or equal to the median size, then the queue depth of the workload queue is set as medium level, and to each workload
The waiting time of queue is set as short delay.
8. the input/output scheduling device of workload-aware in software definition mixing stocking system according to claim 6,
It is characterized in that, if the value of received input-output operation number per second or circulation close to respective property regulation value, connect
The delay time value of receipts, which is near or below, can guarantee that value, the read/write ratio are greater than or equal to 1, and the storage block size is small
In the median size, then the queue depth of the workload queue is set as profound, and to each workload queue etc.
It is set to medium delay to the time.
9. the input/output scheduling device of workload-aware in software definition mixing stocking system according to claim 6,
It is characterized in that, if the value of received input-output operation number per second or circulation close to respective property regulation value, connect
The delay time value of receipts is near or below performance guarantee value, and the read/write ratio is less than 1, then the workload queue
Queue depth is set as medium level, and is set as short delay to the waiting time of each workload queue.
10. the input/output scheduling device of workload-aware in software definition mixing stocking system according to claim 6,
It is characterized in that, if the value of received input-output operation number per second or circulation is near or below respective performance guarantee
Value, received delay time value are near or below performance guarantee value, the read/write ratio is greater than or equal to 1, and the block storage
Size is greater than or equal to the median size, then the queue depth of the workload queue is set as shallow hierarchy, and to each work
The waiting time of load queue is set as short delay.
11. the input/output scheduling device of workload-aware in software definition mixing stocking system according to claim 6,
It is characterized in that, if the value of received input-output operation number per second or circulation is near or below respective performance guarantee
Value, received delay time value are near or below performance guarantee value, the read/write ratio is greater than or equal to 1, and the block storage
Size is less than the median size, then the queue depth of the workload queue is set as medium level, and to each workload
The waiting time of queue is set as medium delay.
12. the input/output scheduling device of workload-aware in software definition mixing stocking system according to claim 6,
It is characterized in that, if the value of received input-output operation number per second or circulation is near or below respective performance guarantee
Value, received delay time value are near or below performance guarantee value, and the read/write ratio is less than 1, then the workload team
The queue depth of column is set as shallow hierarchy, and is set as short delay to the waiting time of each workload queue.
13. the input/output scheduling device of workload-aware in software definition mixing stocking system according to claim 6,
It is characterized in that, if the value of received input-output operation number per second or circulation close to respective property regulation value, connect
The delay time value of receipts is kept off or is greater than or equal to 1 lower than performance guarantee value, the read/write ratio, and the block storage is big
Small to be greater than or equal to the median size, then the queue depth of the workload queue is set as medium level, and to each work
The waiting time of load queue is set as medium delay.
14. the input/output scheduling device of workload-aware in software definition mixing stocking system according to claim 6,
It is characterized in that, if the value of received input-output operation number per second or circulation close to respective property regulation value, connect
The delay time value of receipts is kept off or is greater than or equal to 1 lower than performance guarantee value, the read/write ratio, and the block storage is big
It is less than the median size, then the queue depth of the workload queue is set as profound, and to each workload queue
Waiting time be set as long delay.
15. the input/output scheduling device of workload-aware in software definition mixing stocking system according to claim 6,
It is characterized in that, if the value of received input-output operation number per second or circulation close to respective property regulation value, connect
The delay time value of receipts is kept off or lower than performance guarantee value, the read/write ratio less than 1, and the storage block size is greater than
Or being equal to the median size, then the queue depth of the workload queue is set as medium level, and to each workload team
The waiting time of column is set as short delay.
16. the input/output scheduling device of workload-aware in software definition mixing stocking system according to claim 6,
It is characterized in that, if the value of received input-output operation number per second or circulation close to respective property regulation value, connect
The delay time value of receipts is kept off or lower than performance guarantee value, the read/write ratio less than 1, and the storage block size is less than
The median size, then the queue depth of the workload queue is set as medium level, and to each workload queue etc.
It is set as medium delay to the time.
17. the input/output scheduling device of workload-aware in software definition mixing stocking system according to claim 6,
It is characterized in that, if the value of received input-output operation number per second or circulation is near or below respective performance guarantee
Value, received delay time value are kept off or lower than performance guarantee values, and the storage block size is greater than or equal to the median size,
Then the queue depth of the workload queue is set as shallow hierarchy, and is set as short to the waiting time of each workload queue
Delay.
18. the input/output scheduling device of workload-aware in software definition mixing stocking system according to claim 6,
It is characterized in that, if the value of received input-output operation number per second or circulation is near or below respective performance guarantee
Value, received delay time value are kept off or lower than performance guarantee values, and the storage block size is less than the median size, then the work
The queue depth for making load queue is set as shallow hierarchy, and is set as medium to the waiting time of each workload queue and prolongs
When.
19. the input/output scheduling device of workload-aware in software definition mixing stocking system according to claim 6,
It is characterized in that, the queue depth of the workload queue remains identical or becomes if the solid state hard disk hit rate increases
More shallowly, and identical is maintained to the waiting time of each workload queue or is become shorter, otherwise the team of the workload queue
Column depth remains identical or becomes deeper, and remains identical to the waiting time of each workload queue or become longer.
20. the input/output scheduling device of workload-aware in software definition mixing stocking system according to claim 6,
It is characterized in that, the queue depth of the workload queue remains identical or becomes more shallow if the merging ratio increases, and
Identical is maintained to the waiting time of each workload queue or is become shorter, queue depth's dimension of the otherwise workload queue
It holds identical or becomes deeper, and identical is maintained to the waiting time of each workload queue or becomes longer.
21. the input/output scheduling device of workload-aware in software definition mixing stocking system according to claim 6,
It is characterized in that, the median size is 8KB.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510193350.9A CN106155810B (en) | 2015-04-22 | 2015-04-22 | The input/output scheduling device of workload-aware in software definition mixing stocking system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510193350.9A CN106155810B (en) | 2015-04-22 | 2015-04-22 | The input/output scheduling device of workload-aware in software definition mixing stocking system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106155810A CN106155810A (en) | 2016-11-23 |
| CN106155810B true CN106155810B (en) | 2018-12-25 |
Family
ID=58058445
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510193350.9A Active CN106155810B (en) | 2015-04-22 | 2015-04-22 | The input/output scheduling device of workload-aware in software definition mixing stocking system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106155810B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109614236B (en) * | 2018-12-07 | 2023-04-18 | 深圳前海微众银行股份有限公司 | Cluster resource dynamic adjustment method, device and equipment and readable storage medium |
| CN111208943B (en) * | 2019-12-27 | 2023-12-12 | 天津中科曙光存储科技有限公司 | IO pressure scheduling system of storage system |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101719082A (en) * | 2009-12-24 | 2010-06-02 | 中国科学院计算技术研究所 | Method and system for dispatching application requests in virtual calculation platform |
| CN101957780A (en) * | 2010-08-17 | 2011-01-26 | 中国电子科技集团公司第二十八研究所 | Resource state information-based grid task scheduling processor and grid task scheduling processing method |
| CN102394931A (en) * | 2011-11-04 | 2012-03-28 | 北京邮电大学 | Cloud-based user visit request scheduling method |
| CN102799487A (en) * | 2012-06-29 | 2012-11-28 | 记忆科技(深圳)有限公司 | IO (input/output) scheduling method and apparatus based on array/LUN (Logical Unit Number) |
| CN103299271A (en) * | 2011-01-11 | 2013-09-11 | 惠普发展公司，有限责任合伙企业 | Concurrent request scheduling |
| CN104463492A (en) * | 2014-12-23 | 2015-03-25 | 国家电网公司 | Operation management method of electric power system cloud simulation platform |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9047410B2 (en) * | 2012-07-18 | 2015-06-02 | Infosys Limited | Cloud-based application testing |
-
2015
- 2015-04-22 CN CN201510193350.9A patent/CN106155810B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101719082A (en) * | 2009-12-24 | 2010-06-02 | 中国科学院计算技术研究所 | Method and system for dispatching application requests in virtual calculation platform |
| CN101957780A (en) * | 2010-08-17 | 2011-01-26 | 中国电子科技集团公司第二十八研究所 | Resource state information-based grid task scheduling processor and grid task scheduling processing method |
| CN103299271A (en) * | 2011-01-11 | 2013-09-11 | 惠普发展公司，有限责任合伙企业 | Concurrent request scheduling |
| CN102394931A (en) * | 2011-11-04 | 2012-03-28 | 北京邮电大学 | Cloud-based user visit request scheduling method |
| CN102799487A (en) * | 2012-06-29 | 2012-11-28 | 记忆科技(深圳)有限公司 | IO (input/output) scheduling method and apparatus based on array/LUN (Logical Unit Number) |
| CN104463492A (en) * | 2014-12-23 | 2015-03-25 | 国家电网公司 | Operation management method of electric power system cloud simulation platform |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106155810A (en) | 2016-11-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20160299697A1 (en) | Workload-aware i/o scheduler in software-defined hybrid storage system | |
| US10958719B2 (en) | Pairwise comparison for load balancing | |
| KR101086514B1 (en) | Continuous media priority aware storage scheduler | |
| JP6290462B2 (en) | Coordinated admission control for network accessible block storage | |
| JP3672236B2 (en) | Method, system, and program product for managing a central processing unit in a computer environment | |
| CN102958166B (en) | A kind of resource allocation methods and resource management platform | |
| JP3813056B2 (en) | Processing channel subsystem pending I / O work queue based on priority | |
| US9244742B2 (en) | Distributed demand-based storage quality of service management using resource pooling | |
| CA2382017C (en) | Workload management in a computing environment | |
| US10338970B2 (en) | Multi-platform scheduler for permanent and transient applications | |
| US20150295970A1 (en) | Method and device for augmenting and releasing capacity of computing resources in real-time stream computing system | |
| US9858117B2 (en) | Method and system for scheduling input/output resources of a virtual machine | |
| US10810143B2 (en) | Distributed storage system and method for managing storage access bandwidth for multiple clients | |
| US10250673B1 (en) | Storage workload management using redirected messages | |
| KR20010050503A (en) | Method, system and program products for managing groups of partitions of a computing environment | |
| Yao et al. | LsPS: A job size-based scheduler for efficient task assignments in Hadoop | |
| US10686721B2 (en) | Storage device access mediation | |
| CN113342477A (en) | Container group deployment method, device, equipment and storage medium | |
| US9563366B2 (en) | Using queues corresponding to attribute values associated with units of work and sub-units of the unit of work to select the units of work and their sub-units to process | |
| CN106155810B (en) | The input/output scheduling device of workload-aware in software definition mixing stocking system | |
| Tai et al. | SLA-aware data migration in a shared hybrid storage cluster | |
| Nzanywayingoma et al. | Task scheduling and virtual resource optimising in Hadoop YARN-based cloud computing environment | |
| JP6127754B2 (en) | Program, exclusive control request distribution method and system | |
| Zhang et al. | Utility functions in autonomic workload management for DBMSs | |
| TWI509411B (en) | Workload-aware i/o scheduler in software-defined hybrid storage system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| REG | Reference to a national code |
Ref country code: HK Ref legal event code: DE Ref document number: 1231197 Country of ref document: HK |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |