CN102508618A - Magnetic disk energy consumption optimization method based on request dynamic delay dispatching - Google Patents

Abstract

The invention discloses a magnetic disk energy consumption optimization method based on request dynamic delay dispatching and aims at solving the technical problem to dynamically regulate the magnetic disk dispatching delay time, possibly delay the dispatching moment on the premise of meeting the request average response time requirement and reduce the magnetic disk energy. The method has the technical scheme that a magnetic disk request dispatching system consisting of a queuing system and a submitting system is built, the queuing system receives data reading and writing requests, sequences the requests and submits the sequenced requests to the submitting system, and the submitting system carries out statistics on the request average response time in the dispatching delay time and dynamically regulates the dispatching delay time according to the set request average response time threshold. When the method is adopted, the magnetic disk channel finding distance can be shortened, and the magnetic disk channel finding time is reduced, so the magnetic disk busy time is reduced, the magnetic disk energy consumption is reduced, and in addition, the condition of too long request average response time caused by delay processing request of the magnetic disk is eliminated through dynamically regulating the delay time.

Description

A kind of magnetic disk energy consumption optimization method based on request dynamic delay dispatching

Technical field

The present invention relates to the energy consumption optimization method of disk scheduling, the method that disk energy consumption optimization is espespecially carried out by the dynamic adjustment dispatch delay time.

Background technology

Disk is the critical piece for constituting storage system, mainly by parts such as magnetic head, head arm, drive motor, discs.Data in disk are stored on the disc inside disk.Data carry out tissue on disc in the way of cylinder/magnetic track/sector.Each disc is divided into multiple magnetic tracks, and each magnetic track is the concentric circles on disc.The magnetic track of same position constitutes a cylinder on multiple different discs.Each magnetic track is divided into multiple different sectors, and sector is the base unit that disk carries out reading and writing data.The process being written and read to request data, that is, magnetic head is to the access process of sector.The structure chart of disk is as shown in Figure 1.

The magnetic track of some disc, sector in the request of data address correspondence disk of application program.Disk asks user writable the processing of (hereinafter, read-write requests are referred to as request) to include tracking, rotation and three processes of reading and writing data.

1) magnetic track where first, magnetic head is moved to target data along disc radius, referred to as seeks process；

2) magnetic head then, is kept in the track location, the sector that disc is rotated to where data, referred to as rotary course；

3) the read-write process of log-on data, until the data manipulation asked is completed.

During above-mentioned 3, two processes of tracking and rotation constitute data position fixing process altogether.The time of three processes uses t respectively_seek、t_rotationAnd t_transRepresent, the processing time t of request_requestFor three's sum, that is, there is t_request=t_seek+t_rotation+t_trans.When no request needs processing, magnetic head is no longer moved, but disc keeps rotation, after being reached so as to new request, can be responded as early as possible.During tracking, the track number that magnetic head is moved through is referred to as seeking distance, and tracking distance is bigger, then seeks the time longer.

When disk processing is asked, it is active.Under active state, disk is in one during above-mentioned 3.Now magnetic head either performs tracking operation or performs data read-write operation, and disc rotates at a high speed.When no request processing, disk is in idle condition.Power consumption under active state is high, uses p_activeRepresent, idle condition it is low in energy consumption, use p_idleRepresent, then have p_active＞ p_idle.It is assumed that in the t times, the time that disk is active is t_active, the time of idle condition is t_idle, then energy consumption E of the disk within the time be：

E=p_active×t_active+p_idle×t_idle       (1)

The time that the active state time of disk is consumed by disk processing request.It is assumed that within the t times, disk coprocessing n is asked, then t_activeFor n request processing time sum.

Disk influences the energy consumption of disk to the execution sequence of request.It is not continuous that the request of disk, which is submitted to, the location of on disk.Different requests are distributed on the different magnetic track of disk.If the order that disk is reached according to request, that is prerequisite variable (First Come First Service, FCFS) it is handled, magnetic head is then caused to frequently execute tracking operation on disc, so that tracking operation span when request is handled every time is larger, and magnetic head moving direction frequently changes so that tracking distance and tracking time are longer.In the processing procedure that disk is carried out to request, it is adjusted by the processing sequence to request, it is possible to reduce the tracking distance of magnetic head, the tracking time is reduced, so as to reduce the time of disk in ambulatory situations, it is possible to achieve the purpose of reduction disk energy expenditure.This adjustment process is referred to as disk requests scheduling.One specific scheduling is referred to as a disk requests dispatching method.Fig. 2 and Fig. 3 illustrates validity of the request scheduling for the reduction tracking time using a simply example.It is assumed that request sequence includes 5 requests, x₁, x₂, x₃, x₄, x₅, the magnetic track of access is respectively：5,20,12,8,22.It is assumed that magnetic head initial position is on 0 magnetic track.Under default condition, using frist come,frist serve, this sequence is performed, the tracking distance of magnetic head is 5+15+8+4+14=46, as shown in Figure 2.By asking scheduling to be ranked up 5 requests according to the magnetic track of access, the execution sequence of request is adjusted, the execution sequence after adjustment is：5,8,12,20,22.On this condition, the tracking distance of magnetic head is 22, as shown in Figure 3.After being adjusted, the tracking distance of magnetic head is greatly reduced, and according to tracking distance and the relation between the tracking time, the tracking time also greatly reduces.According to formula (1), the energy consumption of disk is reduced.

Disk energy consumption is the chief component of storage system power consumption, in the environment of current data storage requirement increasingly increases, and how to reduce disk energy consumption turns into one of key issue that " green is calculated " urgent need overcomes.How Mobile state adjustment is entered to scheduling time delay, avoid because delay dispatching makes the response time called request excessive, ensure that disk energy consumption saving is the technical problem that those skilled in the art extremely pay close attention to, and there is no the open report of the disk energy optimizing method technology based on request dynamic delay dispatching at present again simultaneously.

The content of the invention

The technical problem to be solved in the present invention is how to dynamically adjust disk scheduling time delay, under the premise of request average response time requirement is met, as far as possible delay dispatching opportunity, reduces disk energy.The disk energy optimizing method basic thought based on request dynamic delay dispatching of the invention is：After request is reached, postpone to submit to disk, the request reached in the dispatch delay time is ranked up, the request after sequence is submitted disk processing by the position where track location and magnetic head that sort by accesses for these requests.The request that disk is reached when handling request, submits disk processing after next minor sort.By being sorted to request, tracking time during disk processing request is reduced, energy expenditure reduction.The dispatch delay time points to the time interval that disk submits request.

Delay dispatching has postponed the time that request is submitted to disk, and the sequence to request reduces tracking time when disk processing is asked.The dispatch delay time is longer, and the request for participating in sequence is more, and the tracking time of disk is shorter in the case of identical disk access data volume.After delay dispatching, dispatch delay overlong time can increase the response time of request.In addition, under fixed dispatch delay time conditions, the response time of request is related to the arrival feature of request.The dispatch delay time is fixed, and the request reached in the unit interval is more, then the response time asked is longer.Thus, the dispatch delay time can not immobilize, it is necessary to be adjusted according to specific request arrival situation.

The technical scheme is that：Using request dispatch delay time adjustable strategies, for the request reached in the dispatch delay time, it is ranked up according to current head position, and after the dispatch delay time is reached, submits to disk and handled；Count in the dispatch delay time, the average response time of request, according to the request average response time threshold value of setting, the dynamic adjustment dispatch delay time, realize and save disk energy expenditure, it is ensured that the purpose of request response.

Concrete technical scheme comprises the following steps：

The first step, builds disk requests scheduling system, and the system is by request queue subsystem (abbreviation queuing system), and subsystem (abbreviation submission system) composition is submitted in request.Queuing system is a software, is responsible for receiving the reading and writing data request that application program is submitted, request is put into the first request sequence σ, every t_delayTime will be sent to submission system after the request sequence in σ.

Submission system is a software, is responsible for every t_delayTime receives the request sequence after the queuing that queuing system is submitted, and request sequence is added into the second request sequence σ_sub.Submission system is by the second request sequence σ_subRequest submit to disk execution, and count the average response time of request, the time interval t of request submitted according to average response time and average response time threshold value amendment queuing system_delay.First request sequence σ and the second request sequence σ_subThe dynamic link table in committed memory space, its reading and writing data number of request for being submitted with application program of memory space taken and change.

Second step, sets initial guess, including：

2.1 set request average response time threshold value t_res, by t_resIt is set as the historical statistics value of requirement or request average response time of the user program to request response time；

2.2 set dispatch delay time t_delayInitial value be t_res；

2.3 set the position h of magnetic head as 0 magnetic track；

3rd step, queuing system receives reading and writing data request, and request is ranked up, the request after sequence is submitted into submission system.The request sequence that queuing system is submitted is added to the second request sequence σ by submission system_subIn, and by σ_subIn request submit to disk in order.

3.1 queuing systems receive the reading and writing data request x that application program is submitted_i, record request arrival timeAnd x will be asked_iIt is put into the first request sequence σ, the numbering that i reaches for request, i is integer；

3.2 every time t_delay, to the request sequence being stored in σ<x₁, x₂..., x_i..., x_n>It is ranked up, n is t_delayThe sum of request is reached in period, n is integer, and the request sequence after sequence is represented with σ '.Sort method is：

If 3.2.1

Then n request is ranked up according to track address is ascending, l represents the minimum track address for asking to access in σ, r represents maximum track address；

If 3.2.2

Then n request is ranked up according to track address is descending；

3.3 according to the sequences of the middle requests of σ ', and it is the residing magnetic track of last request after sequence to update the track address h residing for magnetic head；

Request sequence σ ' after sequence is sent to submission system by 3.4 queuing systems, and queuing system empties the first request sequence σ；

3.5 submission systems receive the σ ' that queuing system is submitted, and σ ' is added into the second request sequence σ_sub；

3.6 submission systems take out σ successively_subIn be located at the request of sequence header, disk processing is submitted to, and in σ_subMiddle deletion request；

4th step, submission system statistics disk processing σ_subThe average response time of middle request, and according to average response time and average response time threshold value, to the dispatch delay time t of queuing system_delayIt is adjusted.

4.1 disks have performed σ_subIn request x_jAfterwards, numberings of the j for request in σ ', 1≤j≤n, submission system records its deadline

Calculate x_jResponse time be

It is x_jArrival time；

4.2 submission systems count dispatch delay time t_delayInterior, disk completes the average response time of all requestsM is the number of request that disk is completed in the dispatch delay time.Will

With request response time threshold value t_resContrast, the adjustment queuing system new dispatch delay time is t_delay', method is as follows：

If 4.2.1 $\stackrel{\&OverBar;}{t_{\mathrm{res}}}>t_{\mathrm{res}},$ The two difference of order $t_{\mathrm{diff}}=\stackrel{\&OverBar;}{t_{\mathrm{res}}}-t_{\mathrm{res}},$ Then ${t_{\mathrm{delay}}}^{\&prime;}=t_{\mathrm{delay}}-\frac{1}{2}\&times;t_{\mathrm{diff}};$

If 4.2.2 $\stackrel{\&OverBar;}{t_{\mathrm{res}}}<t_{\mathrm{res}},$ $t_{\mathrm{diff}}=t_{\mathrm{res}}-\stackrel{\&OverBar;}{t_{\mathrm{res}}},$ Then ${t_{\mathrm{delay}}}^{\&prime;}=t_{\mathrm{delay}}+\frac{1}{2}\&times;t_{\mathrm{diff}};$

4.3 submission systems are by the t after adjustment_delay' it is sent to queuing system；

4.4 queuing systems adjust dispatch delay time t_delayFor t_delay=t_delay', turn the 3rd step.

In step 4.2, disk completes the average response time of request in the submission system statistics dispatch delay time.Within the dispatch delay time, the number of request that disk is completed is m.When the number of request of arrival is more, in fact it could happen that all processing is completed for the request that m ＜ n situation, i.e. disk reached the upper dispatch delay time not within the dispatch delay time.In this case, during the submission system statistical average response time, the second request sequence σ_subIn still include untreated request.But, according to following 2 points, σ_subIn request will not unrestrictedly increase：1) data access of application program will not be constantly in high request frequency state；2) request that submission system is received is derived from operating system, and the request control function of operating system limits total number of request of whole system.Step 4.4 shows that the present invention is an endless loop, as long as not shut down because actual conditions are exactly computer, and disk requests scheduling system constantly receives the reading and writing data request of application program submission, and constantly request is handled.Unless computer shutdown or power-off, the present invention just force to shut down, the giant computer of HLRS does not shut down usually in actual state, therefore energy optimization need to continue to carry out.

Following effect can be reached using the present invention：

1) 3.2 step of the invention submits disk processing after being ranked up to the request reached in scheduling time delay, has reached that shortening disk seeks distance, has reduced the tracking time, so as to reduce disk busy time, reduces the purpose of disk energy expenditure.

2) present invention 4.2 is according to the average response time, request response time threshold value asked in the dispatch delay time, and the size of dynamic adjustment time delay is eliminated because disk delay disposal asks the excessive situation of triggered request average response time.

Brief description of the drawings

Fig. 1 is the structural representation of disk.

Fig. 2 is to move schematic diagram using prerequisite variable scheduling mode magnetic head.

Fig. 3 is to be ranked up rear magnetic head to request using the present invention to move schematic diagram.

Fig. 4 is general flow chart of the present invention.

Fig. 5 is second step of the present invention, and queuing system receives request, and the request after sequence is submitted into submission system, and the flow chart for submitting disk to handle.

Embodiment

Fig. 1 is the structural representation of disk.The figure illustrates the institutional framework of magnetic head and disc.Disc is divided into several different magnetic tracks, and magnetic track is divided into sector again, and sector is the base unit of reading and writing data.Magnetic head is moved between different magnetic tracks, and the data in sector are written and read.

Fig. 2 is that when using prerequisite variable scheduling mode, disk performs 5 request x in example₁, x₂, x₃, x₄, x₅When, magnetic head movement schematic diagram.The order that magnetic head is reached according to request, the magnetic track being moved to where request.When performing second request, magnetic head is moved to magnetic track 20.When performing the 3rd request, magnetic head needs to be moved to magnetic track 12 in the opposite direction.5 requests are performed, magnetic head moving direction needs to change twice.

Fig. 3 is that after being ranked up using the present invention to request, disk performs 5 request x in example₁, x₂, x₃, x₄, x₅When, magnetic head movement schematic diagram.After being ranked up to 5 requests according to residing magnetic track, magnetic head is moved respectively to magnetic track 5,8,12,20,22.5 requests are performed, magnetic head moving direction does not change.

Fig. 4 is the general flow chart of the present invention, is specifically included：

The first step, builds disk requests scheduling system, including queuing system and Request System.

Second step, sets initial guess.

3rd step, queuing system receives reading and writing data request, and request is ranked up, the request after sequence is submitted into submission system, the request sequence that queuing system is submitted is added to the second request sequence σ by submission system_subIn, and by σ_subIn request submit to disk in order.

4th step, submission system statistics disk processing σ_subThe average response time of middle request, and according to average response time and average response time threshold value, to the dispatch delay time t of queuing system_delayIt is adjusted, turns the 3rd step.

Fig. 5 describes the flow of the 3rd step of the invention, specifically includes：

3.1 queuing systems receive the reading and writing data request that application program is submitted, and will ask x_iIt is put into request sequence；

3.2 every time t_delay, the request being stored in request sequence is ranked up, sort method is：

If 3.2.1

Then n request is ranked up according to track address is ascending, l represents the minimum track address for asking to access in σ, r represents maximum track address；

If 3.2.2

Then n request is ranked up according to track address is descending；

3.3 update the track address h residing for magnetic head for the residing magnetic track of last request after sorting；

Request sequence after sequence is sent to submission system by 3.4 queuing systems, and queuing system empties request sequence；

3.5 submission systems receive the request sequence that queuing system is submitted；

3.6 submission systems take out the request of sequence header successively, submit to disk processing, and delete the request in the sequence.

Claims (3)

1. a kind of magnetic disk energy consumption optimization method based on request dynamic delay dispatching, it is characterised in that comprise the following steps：

The first step, structure disk requests scheduling system, the system are made up of queuing system, submission system, and queuing system is a software, is responsible for receiving the reading and writing data request of application program submission, request is put into the first request sequence σ, every t_delayTime will be sent to submission system after the request sequence in σ；Submission system is a software, is responsible for every t_delayTime receives the request sequence after the queuing that queuing system is submitted, and request sequence is added into the second request sequence σ_sub, and by σ_subRequest submit to disk execution, and count the average response time of request, the time interval t of request submitted according to average response time and average response time threshold value amendment queuing system_delay；

Second step, sets initial guess, including：

2.1 set request average response time threshold value t_res, by t_resIt is set as the historical statistics value of requirement or request average response time of the user program to request response time；

2.2 set dispatch delay time t_delayInitial value be t_res；

2.3 set the position h of magnetic head as 0 magnetic track；

3rd step, queuing system receives reading and writing data request, and request is ranked up, the request after sequence is submitted into submission system, the request sequence that queuing system is submitted is added to the second request sequence σ by submission system_subIn, and by σ_subIn request submit to disk in order, specific method is：

3.1 queuing systems receive the reading and writing data request x that application program is submitted_i, record request arrival time

And x will be asked_iIt is put into the first request sequence σ, the numbering that i reaches for request, i is integer；

3.2 every time t_delay, to the request sequence being stored in σ<x₁, x₂..., x_i..., x_n>It is ranked up, n is t_delayThe sum of request is reached in period, n is integer, and the request sequence after sequence is represented with σ '；

3.3 according to the sequences of the middle requests of σ ', and it is the residing magnetic track of last request after sequence to update the track address h residing for magnetic head；

Request sequence σ ' after sequence is sent to submission system by 3.4 queuing systems, and queuing system empties the first request sequence σ；

3.5 submission systems receive the σ ' that queuing system is submitted, and σ ' is added into the second request sequence σ_sub；

3.6 submission systems take out σ successively_subIn be located at the request of sequence header, disk processing is submitted to, and in σ_subMiddle deletion request；

4th step, submission system statistics disk processing σ_subThe average response time of middle request, and according to average response time and average response time threshold value, to the dispatch delay time t of queuing system_delayIt is adjusted, method is：

4.1 disks have performed σ_subIn request x_jAfterwards, numberings of the j for request in σ ', 1≤j≤n, submission system records its deadline

Calculate x_jResponse time be

It is x_jArrival time；

4.2 submission systems count dispatch delay time t_delayInterior, disk completes the average response time of all requests

M is the number of request that disk is completed in the dispatch delay time；Will

With request response time threshold value t_resContrast, the adjustment queuing system new dispatch delay time is t_delay′；

4.3 submission systems are by the t after adjustment_delay' it is sent to queuing system；

4.4 queuing systems adjust dispatch delay time t_delayFor t_delay=t_delay', turn the 3rd step.

2. a kind of magnetic disk energy consumption optimization method based on request dynamic delay dispatching as claimed in claim 1, it is characterised in that to the request sequence being stored in σ<x₁, x₂..., x_i..., x_n>The method being ranked up is：

If 3.2.1

Then n request is ranked up according to track address is ascending, l represents the minimum track address for asking to access in σ, r represents maximum track address；

If 3.2.2

Then n request is ranked up according to track address is descending.

3. a kind of magnetic disk energy consumption optimization method based on request dynamic delay dispatching as claimed in claim 1, it is characterised in that the adjustment queuing system new dispatch delay time is t_delay' method be：

If 4.2.1 $\stackrel{\&OverBar;}{t_{\mathrm{res}}}>t_{\mathrm{res}},$ The two difference of order $t_{\mathrm{diff}}=\stackrel{\&OverBar;}{t_{\mathrm{res}}}-t_{\mathrm{res}},$ Then ${t_{\mathrm{delay}}}^{\&prime;}=t_{\mathrm{delay}}-\frac{1}{2}\&times;t_{\mathrm{diff}};$

If 4.2.2 $\stackrel{\&OverBar;}{t_{\mathrm{res}}}<t_{\mathrm{res}},$ $t_{\mathrm{diff}}=t_{\mathrm{res}}-\stackrel{\&OverBar;}{t_{\mathrm{res}}},$ Then ${t_{\mathrm{delay}}}^{\&prime;}=t_{\mathrm{delay}}+\frac{1}{2}\&times;t_{\mathrm{diff}}.$

Images