KR20200114274A - I/0 optimization apparatus and method for improving loading performance of an application - Google Patents

Abstract

The present invention relates to an I/O optimization device for improving the loading performance of an application and a method thereof. The I/O optimization device for improving the loading performance of an application comprises: a first work set generation part that generates a first work set through I/O scheduling including I/O alignment and I/O merging; an I/O padding part that performs I/O padding for filling a padded page between a plurality of page chunks constituting the first work set for the generated first work set; and a work set generation part that generates a second work set in which the I/O size is enlarged through the I/O padding. Accordingly, the present invention can shorten the loading speed of an application in a computing system by increasing the I/O size to increase effective storage I/O throughput.

Description

I/O optimization device and method for improving application loading performance {I/0 OPTIMIZATION APPARATUS AND METHOD FOR IMPROVING LOADING PERFORMANCE OF AN APPLICATION}

The present invention relates to an input/output optimization technology for improving application loading performance, and more particularly, an input/output optimization device and method for reducing the loading speed of an application in a computing system by increasing the effective storage input/output throughput by increasing the size of the input/output. It is about.

Recent smart devices and mobile devices support simultaneous execution of multiple user applications using an operating system based on demand paging. Demand paging has proven to be effective in efficiently sharing limited system resources and maintaining a constantly changing working set of applications in memory. However, request paging can resume application execution after page fetching is completed when a page fault occurs. Therefore, a single page size (4 KB) can be reduced when starting an application and activating a background application caused by a large number of page faults. Generates a blocking read command. On the other hand, flash storage devices mounted on smart devices and mobile devices are internally composed of multiple flash chips and are implemented to overcome the low transmission speed of a single chip through parallelization. To use this, input/output commands from the host device are simultaneously Multiple numbers are transmitted and processed in parallel, or a host device receives a large I/O command of hundreds of KB or more and internally divides it to perform parallel processing.

Therefore, the I/O pattern generated when the request paging is brought in the application has a problem that the potential performance of the flash storage device is not properly utilized, but the existing operating system does not sufficiently apply the optimization technique that considers these characteristics of the flash storage device.

Korean Patent Publication No. 10-2007-0068801 (2007.07.02)

An embodiment of the present invention is to provide an input/output optimization apparatus and method for shortening the loading speed of an application in a computing system by increasing the size of input/output to increase effective storage input/output throughput.

An embodiment of the present invention is an application that shortens the loading time of a working set by filling a padded page between page chunks constituting a working set for a previously learned application working set and increasing the size of input/output. To provide an input/output optimization device and method for improving the loading performance of

An embodiment of the present invention defines a padding vector that uses empty spaces between page chunks as padding candidates, and provides optimal padding to obtain a minimum fetch time for a working set through a dynamic programming model. It is to provide an input/output optimization apparatus and method for improving the loading performance of an application that determines a vector.

Among embodiments, an input/output optimization device for improving the loading performance of an application includes a first work set generation unit that generates a first work set through input/output scheduling including input/output sorting and input/output merging, and the generated first work set. On the other hand, an input/output padding unit performing input/output padding to fill a padded page between a plurality of page chunks constituting the first work set, and a second working set in which the size of input/output is increased through the input/output padding. And a second work set generation unit that generates a final work set.

The input/output scheduling unit may arrange the plurality of page chunks in a logical block address (LBA) order through the input/output alignment.

The first work set generation unit may generate the first work set by merging a plurality of adjacent page chunks among the plurality of aligned page chunks into a single page chunk through the input/output merging.

The input/output padding unit may define a padding vector that determines empty spaces between the plurality of page chunks as padding candidates.

The input/output padding unit may detect a padding vector having a minimum loading time for the first working set through a dynamic programming model.

The dynamic programming model may detect the padding vector in consideration of the sizes of the padding candidates and the sizes of adjacent page chunks surrounding each of the padding candidates.

The dynamic programming model may detect the padding vector in consideration of whether input/output padding is performed for nearby padding candidates and storage input/output throughput for a different starting LBA and a transmission size.

The input/output padding unit may dynamically determine whether to perform the input/output padding for the padding candidates through the detected padding vector.

Among embodiments, an input/output optimization method for improving the loading performance of an application includes generating a first work set through input/output scheduling including input/output sorting and input/output merging, and for the generated first work set, the first Performing input/output padding to fill a padded page between a plurality of page chunks constituting a work set, and generating a second work set in which the size of input/output is enlarged through the input/output padding as a final work set It may include steps.

Among embodiments, the method for optimizing input/output for improving the loading performance of an application includes profiling a first work set generated at a time when an application is executed or a background application is activated, from the generated first work set, Creating a second work set in which the size of input/output is increased by filling in a padded page between a plurality of page chunks constituting the first work set, when the execution of the application or activation of the background application is detected Pausing the application or the background application, and applying the second work set to the application or the background application and executing it again.

The disclosed technology can have the following effects. However, since it does not mean that a specific embodiment should include all of the following effects or only the following effects, it should not be understood that the scope of the rights of the disclosed technology is limited thereby.

The apparatus and method for optimizing input/output for improving application loading performance according to an embodiment of the present invention can reduce an application loading speed in a computing system by increasing an effective storage input/output throughput by increasing the size of the input/output.

The apparatus and method for optimizing input/output for improving the loading performance of an application according to an embodiment of the present invention include padding pages between page chunks constituting a working set and input/output with respect to a previously learned application working set. By increasing the size of the task set, you can shorten the loading time of the work set.

The apparatus and method for optimizing input/output for improving the loading performance of an application according to an embodiment of the present invention define a padding vector using empty spaces between page chunks as padding candidates, and minimize a working set through a dynamic programming model. It is possible to determine an optimal padding vector for obtaining the fetch time of.

1 is a diagram illustrating an input/output optimization system for improving loading performance of an application according to an embodiment of the present invention.
FIG. 2 is a block diagram illustrating an input/output optimizing device for improving loading performance of an application in FIG. 1.
3 is a diagram illustrating an input/output throughput of a storage according to an input/output size according to an exemplary embodiment.
4 is an exemplary diagram showing a method of determining a padding vector through a dynamic programming model of an input/output padding unit according to an embodiment of the present invention.
5 is a diagram illustrating an input/output padding result of an input/output padding unit according to an embodiment of the present invention.
6 is a flowchart illustrating an input/output optimization method for improving an application loading performance according to an embodiment of the present invention.
7 is a flowchart illustrating an input/output optimization method for improving an application loading performance according to another embodiment of the present invention.
8 is a graph showing the effect of improving the loading speed of an application according to an embodiment of the present invention.

Since the description of the present invention is merely an embodiment for structural or functional description, the scope of the present invention should not be construed as being limited by the embodiments described in the text. That is, since the embodiments can be variously changed and have various forms, the scope of the present invention should be understood to include equivalents capable of realizing the technical idea. In addition, since the object or effect presented in the present invention does not mean that a specific embodiment should include all of them or only those effects, the scope of the present invention should not be understood as being limited thereto.

Meanwhile, the meaning of terms described in the present application should be understood as follows.

Terms such as "first" and "second" are used to distinguish one component from other components, and the scope of rights is not limited by these terms. For example, a first component may be referred to as a second component, and similarly, a second component may be referred to as a first component.

When a component is referred to as being "connected" to another component, it should be understood that although it may be directly connected to the other component, another component may exist in the middle. On the other hand, when it is mentioned that a certain component is "directly connected" to another component, it should be understood that no other component exists in the middle. On the other hand, other expressions describing the relationship between components, that is, "between" and "directly between" or "neighboring to" and "directly neighboring to" should be interpreted as well.

Singular expressions are to be understood as including plural expressions unless the context clearly indicates otherwise, and terms such as “comprise” or “have” refer to implemented features, numbers, steps, actions, components, parts, or It is to be understood that it is intended to designate that a combination exists and does not preclude the presence or addition of one or more other features or numbers, steps, actions, components, parts, or combinations thereof.

In each step, the identification code (for example, a, b, c, etc.) is used for convenience of explanation, and the identification code does not describe the order of each step, and each step has a specific sequence clearly in context. Unless otherwise stated, it may occur differently from the stated order. That is, each of the steps may occur in the same order as specified, may be performed substantially simultaneously, or may be performed in the reverse order.

The present invention can be embodied as computer-readable codes on a computer-readable recording medium, and the computer-readable recording medium includes all types of recording devices storing data that can be read by a computer system. . Examples of computer-readable recording media include ROM, RAM, CD-ROM, magnetic tape, floppy disk, and optical data storage devices. Further, the computer-readable recording medium is distributed over a computer system connected by a network, so that the computer-readable code can be stored and executed in a distributed manner.

All terms used herein have the same meaning as commonly understood by one of ordinary skill in the field to which the present invention belongs, unless otherwise defined. Terms defined in commonly used dictionaries should be construed as having meanings in the context of related technologies, and cannot be construed as having an ideal or excessive formal meaning unless explicitly defined in the present application.

The loading time of an application is one of the key performance indicators of a computing system because it is a factor that has a decisive influence on the performance that users can perceive. As the size of mobile applications increases rapidly, the importance of application loading performance is increasingly emphasized in mobile devices, and the latest mobile OS (Operating System) relies on request paging to bring code and data pages of user applications into memory. In particular, since I/O requests are generated only when the requested page is missed, the weakness of mobile storage devices is remarkably revealed when loading a rapidly changing working set such as starting an application or bringing a background application to the foreground.

In order to alleviate the inefficiency of requested paging, an aggressive merging scheme is initiated as an input/output optimization apparatus and method (hereinafter, input/output optimization apparatus 110) for improving application loading performance according to an embodiment of the present invention. do. In one embodiment, the aggressive merging technique may correspond to an explicit application loading method composed of an input/output optimization technique including input/output (I/O) alignment, input/output merging, and input/output padding. In conclusion, the input/output optimization device 110 may increase the effective storage throughput by increasing the size of the input/output related to application loading.

An explicit application loading method of the aggressive merge technique Demand paging can implicitly load an application by fetching only missing pages in an on-demand manner. Demand paging is efficient at keeping track of constantly changing working sets, but because the size of the optimization window is limited to a single I/O request, it can severely limit the opportunities for I/O optimization. In order to overcome this limitation, the input/output optimization device 110 of the present invention includes: 1) profiling a working set of a target application, 2) determining when to start loading an application, 3) pausing the target application, Disclosed is a method of explicitly loading or fetching a working set of an application through 4) loading or fetching a working set through input/output optimization technology, and 5) restarting the target application.

In one embodiment, the input/output optimizing apparatus 110 may be applied when a work set is predetermined by profiling. Accordingly, the input/output optimization device 110 may supplement the requested paging. For example, the input/output optimization device 110 may be applied when starting an application or activating a deactivated background application. In one embodiment, in the case of starting an application, the input/output optimization device 110 may acquire a work set immediately after execution of the application is completed using a block layer I/O tracing mechanism (blktrace) as an input/output tracking tool. More specifically, the input/output optimization device 110 may empty a page cache and secure input/output requests generated during an application execution process. In another embodiment, when activating a deactivated background application, the input/output optimization apparatus 110 may determine a task set immediately before the application is deactivated in the background as the target task set. For example, the input/output optimization device 110 may assume that most of the pages of the working set have been removed from memory while the application is deactivated in the background. In one embodiment, the input/output optimization device 110 may track the set of removed pages in the same manner as the Linux swap system.

In an embodiment, the input/output optimization device 110 may accurately detect when a user executes an application or activates a background application through an operating system. However, it is not always beneficial to load an application upon execution or activation of such an application. For example, when most pages of a working set remain in memory, starting the loading of the application may delay the execution or activation of the application. Accordingly, the operating system can check the page cache state and selectively perform the method of the present invention only when, for example, a part of the working set in the cache is below the threshold. In an embodiment, the input/output optimizing apparatus 110 may start an input/output optimization apparatus and method for improving an application loading speed assuming that a cold start scenario, that is, a page of a working set does not exist in memory.

I/O sorting and I/O merging are performed by the I/O scheduler, so that I/O optimization for I/O in the queue can be performed. I/O sorting and I/O merging are effective for non-blocking I/O such as write requests and asynchronous reads, but I/O schedulers have little impact on the application loading process, where blocking read I/Os are dominated. More specifically, the maximum number of blocking read I/Os appearing in the I/O queue is limited to 1, and the next I/O is executed only after the current blocking read I/O is completed.

In one embodiment, the I/O optimization device 110 may initiate an explicit application loading method to access a predetermined working set before sending an actual I/O request, so that all page chunks may be perfectly aligned. For example, I/O alignment itself does not directly affect the loading time of a working set of flash memory, but it indirectly provides more I/O consolidation opportunities, greatly increasing the size of I/O while loading a working set. have.

I/O sorting and I/O merging are limited in that they do not change the amount of data in the working set. In an embodiment, the input/output optimization apparatus 110 may alleviate this limitation by continuously applying an input/output padding technique to input/output alignment and input/output merging. For example, the input/output optimization device 110 may increase the amount of transmitted data by inserting a padding page between chunks of adjacent pages. That is, the present invention can increase the size of the input/output through input/output padding, thereby increasing the effective input/output throughput during application loading. However, in order to successfully perform input/output padding, the gain through the expansion of the input/output size must exceed the overhead caused by the padding pages. Simply finding a certain size to fill all padding candidates below the threshold cannot minimize the loading time of the application. Instead, the input/output optimization device 110 determines the distance between adjacent page chunks, the size of the padding candidates, the size of the adjacent page chunks, whether padding for other nearby padding candidates, and storage input/output throughput for other starting LBAs and transmission sizes. In consideration of this, optimal input/output padding can be performed.

In one embodiment, the input/output optimization apparatus 110 proposes an effective input/output padding technique based on a dynamic programming model. In the present invention, when input/output sorting and input/output merging are completed, the working set of the target application consisting of n page chunks is given by W. Here, W = (b0, b1, ..., bn-1), and bi may correspond to page chunks of adjacent pages arranged in LBA order. In one embodiment, the blank space between adjacent page chunks may be greater than or equal to the size of a single page. Given the inputs W0,n-1, the related padding vector P can be defined. Here, it may correspond to P = (p0, p1, ..., pn-2). Here, pi corresponds to a padding candidate for an empty space between bi and bi+1. In one embodiment, if it is determined to insert a padded page between bi and bi+1, each pi may be set to 1. For example, P = (1, 1, 1, ..., 1) may mean that all empty spaces in W are filled with padded pages to generate a single I/O request covering the entire working set.

The input/output optimization device 110 may solve the problem of determining the padding by finding an optimal padding vector P so that the loading time of the target application program is minimized for a given working set W.

In an embodiment, the input/output optimizing device 110 may determine an optimal padding vector that minimizes the loading time of an application through a dynamic programming model. For example, determine the subset Wi,j = (bi, ..., bj) for the working set W, and set the padding vector for the subset to Pi,j-1 = (pi, pi+1, .. ., pj-1). Then, read(i,j) is determined as the storage access time for loading Wi,j through Pi,j-1. In one embodiment, read(i,j) may be calculated by a single input/output request that includes a section from bi to bj. Finally, we define t(i,j) as the time to load Wi,j through the optimal padding vector. In an embodiment, a relationship between t(i,j) and its subproblem t(s,t) may be distinguished through Equation 1 below.

[Equation 1]

In one embodiment, if i=j, there is only one page chunk in the working set, so no I/O padding occurs. If j is greater than i, then the padding vector Pi,j-1 is initialized to (X,X,...,X). Here, X may mean a state in which input/output padding has not yet been determined. Thereafter, the min function can find the minimum time to load Wi,j by comparing the next input/output padding decisions.

One) When t(i,j) is read(i,j), empty space is filled through input/output padding for all padding candidates. That is, Pi,j-1 = (1,1,...,1).

을 선택하고, Ps를 0으로 설정하고 나머지를 X로 설정한다. 예를 들어, s가 i+1로 설정되면, pi.j-1은 (X,O,X,...,X)가 되고 t(i,j)는 t(i,i+1)+t(i+2,j)가 된다.2) one candidate

Select, set Ps to 0 and the rest to X. For example, if s is set to i+1, pi.j-1 becomes (X,O,X,...,X) and t(i,j) becomes t(i,i+1)+ becomes t(i+2,j).

t(i,j) can be calculated only after completing the calculation of the underlying subproblem. For example, the subproblem may correspond to t(i,i+1) and t(i+2,j) for Ps with s=i+1. The input/output optimizing device 110 may use a top-down approach with memorization or a bottom-up approach that approaches a table to evaluate all required sub-problems. Both methods can significantly reduce computational overhead by evaluating each subproblem only once.

Finally, by applying Equation 1 to t(0,n-1), the input/output optimizing apparatus 110 may calculate an optimal padding vector P0,n-2 for a given input W0,n-1. 4 corresponds to an exemplary diagram for W0,3 = (b0,b1,b2,b3). 4(c)-(e) shows that t(0,3) has 6 different subproblems, of which t(0,2) is determined recursively in FIG. 4(fh). I can. For example, if t(0,0) and t(0,1) are already calculated in Fig. 4(cd) and stored in memory, the second calculation in Fig. 4(gh) is performed by simply retrieving the stored values. Can be.

Equation 1 requires a storage performance model that returns the read access time for a given starting LBA and size. You can develop a measurement-based model that uses only the input/output size to predict the read access time. In particular, for a given input/output size, the present invention may repeat a task of measuring a read latency value with a randomly generated LBA. Among the obtained values, the input/output optimization device 110 may select an intermediate value rather than an average in order to mitigate the influence due to the singularity.

Since it takes too much time to perform measurements for all possible I/O sizes, in the present invention, the read access time increases linearly when I/O requests are larger than 128 KB, and linear extrapolation to I/Os larger than 128 KB. You can use the method and construct a lookup table for I/O sizes of up to 128 KB.

In conclusion, the input/output optimization device 110 according to an embodiment of the present invention shows that increasing the input/output size by measuring the read throughput of the mobile storage for various input/output sizes can be an effective method of improving application loading performance. Start. In addition, the input/output optimizing device 110 may disclose an active merging technique composed of an explicit application loading method along with input/output alignment, input/output merge *?*, and input/output padding techniques. Lastly, the input/output optimization device 110 may calculate and apply a dynamic programming model to find an optimal padding determination (padding vector) for input/output padding. As a result of the experiment, the average input/output size on the Google Nexus 5 smartphone increased by 5.6X, reducing the application loading time by 30%. Hereinafter, an input/output optimizing apparatus 110 according to an embodiment of the present invention will be described in detail with reference to FIGS. 1 to 8.

1 is a diagram illustrating an input/output optimization system (hereinafter, an input/output optimization system) for improving an application loading performance according to an embodiment of the present invention.

Referring to FIG. 1, the input/output optimization system 100 may include an input/output optimization device 110 (hereinafter referred to as input/output optimization device), a user terminal 120, and a storage 130 for improving application loading performance.

The input/output optimizing device 110 may optimize an input/output command to reduce a loading speed of an application in the starting stage of the application or the activation stage of the deactivated background application. More specifically, the input/output optimization device 110 may be connected to the user terminal 120 to reduce a loading speed in the process of executing or activating an application through the user terminal 120 of the user and, as a result, improve performance. In one embodiment, the input/output optimization device 110 may achieve a maximum input/output throughput of a flash storage that cannot be achieved in a conventional paging-based operating system. For example, the input/output optimization device 110 may improve loading performance through an aggressive merging scheme that increases the storage throughput by increasing the size of the input/output through input/output padding during application loading.

The user terminal 120 may correspond to a computing device that is connected to the input/output optimization device 110 to provide various applications to a user. In one embodiment, the user terminal 120 may be implemented as a smartphone or a tablet PC, and is not necessarily limited thereto, and may be implemented as various devices capable of executing applications. The user terminal 120 may be connected to the input/output optimization device 110 through a network, and at least one user terminal 120 may be connected to the input/output optimization device 110 at the same time, preferably, the user terminal 120 Can be implemented inside of

The storage 130 may correspond to various types of memories or storage devices. The storage 130 may be implemented as a nonvolatile or volatile memory, and may correspond to a storage device used to store all data required to execute an application of the user terminal 120. For example, the storage 130 may correspond to a flash memory.

FIG. 2 is a block diagram illustrating an input/output optimizing device (hereinafter referred to as an input/output optimizing device) for improving the loading performance of the application shown in FIG. 1.

Referring to FIG. 2, the input/output optimization device 110 includes a first work set generation unit 210, an input/output padding unit 220, and a second work set generation unit 230.

A working set may mean that unit tasks are configured as a set so that a task can be performed at a task stage such as starting or activating an application. The first work set generation unit 210 may generate a first work set through input/output scheduling. Here, input/output scheduling may mean that when an operating system (OS) receives a request to read or write a disk from a plurality of processors, it sets a priority and manages it. For example, I/O scheduling may be performed by various types of I/O schedulers. In an embodiment, the first work set generation unit 210 may generate the first work set through input/output scheduling including input/output sorting and input/output merging. Here, I/O reordering and I/O merging are performed by the I/O scheduler, and optimization for input/output (I/O) pending in the queue is performed. I can. In an embodiment, the first work set generation unit 210 may arrange a plurality of page chunks in a logical block address (LBA) order through input/output alignment. Here, the first working set may be composed of a plurality of page chunks corresponding to a bundle of a plurality of pages, and the LBA designates a location of a data block stored in a computer storage device. This is a common scheme used, and can correspond to a simple linear addressing scheme. In an embodiment, the first work set generation unit 210 may generate a first work set by merging adjacent page chunks from among a plurality of page chunks arranged through input/output merging into a single page chunk. In an embodiment, the first work set generation unit 210 may increase the size of input/output by increasing the probability of input/output merging by arranging page chunks through input/output alignment in the process of generating the first work set.

The input/output padding unit 220 performs input/output padding (I/O padding) filling a padding page between a plurality of page chunks constituting the first work set for the generated first work set. I can. Here, the padded page may correspond to a specific page inserted to fill an empty space existing between a plurality of page chunks each including a plurality of pages. In addition, input/output padding may mean additionally inserting a padding page between a plurality of page chunks constituting the first working set. More specifically, the input/output padding unit 220 may increase the size of the input/output by performing input/output padding, thereby reducing a loading speed of an application and improving performance. In an embodiment, the input/output padding unit 220 may define a padding vector for determining blank spaces between a plurality of page chunks as padding candidates in a process of performing input/output padding. More specifically, the input/output padding unit 220 may define an optimal padding vector such that a gain obtained by increasing the size of input/output through input/output padding exceeds an overhead due to additional insertion of a padding page. That is, the input/output padding unit 220 may increase the size of the input/output by performing efficient input/output padding through the optimal padding vector. More specifically, the input/output padding unit 220 includes padding candidates to fill the padding page among a plurality of padding candidates in the process of performing input/output padding for the first work set generated through input/output scheduling including input/output alignment and input/output merging. It is possible to detect an optimal padding vector to determine the.

The working set of a mobile application can typically consist of hundreds or thousands of page chunks, and the computational overhead of I/O sorting and I/O merging can be negligible. On the other hand, the input/output optimizing device 110 uses a dynamic programming model in a process of performing input/output padding through the input/output padding unit 220 and has overhead required for calculation and storage in terms of time and space. In an embodiment, the input/output padding unit 220 may divide the first working set into a plurality of subsets whenever padding candidates larger than 128 KB are generated to reduce overhead. For example, the input/output padding unit 220 can reduce the number of table items from 555,985 (2,171 KB) to 3,756 (15 KB) in the process of optimizing the messenger application, and accordingly, can reduce the calculation time from 213 seconds to 0.07 seconds. .

In an embodiment, the input/output padding unit 220 may detect a padding vector having a minimum loading time for the first working set through a dynamic programming model. More specifically, the input/output padding unit 220 may detect the padding vector in consideration of the size of the padding candidates and the size of adjacent page chunks surrounding each of the padding candidates through the dynamic programming model. The input/output padding unit 220 may detect a padding vector in consideration of whether input/output padding for other adjacent padding candidates is performed and a storage input/output throughput for a different starting LBA and a transmission size through a dynamic programming model. In conclusion, the input/output padding unit 220 may dynamically determine whether to perform input/output padding for padding candidates through the detected padding vector.

이 주어지면, 관련된 패딩 벡터 P = (p0, p1, ..., pn-2)를 정의할 수 있다. 여기에서, pi는 bi와 bi+1 사이의 빈 공간에 대한 패딩 후보에 해당한다. 일 실시예에서, 입출력 패딩부(220)는 만약 bi와 bi+1 사이에 패딩 페이지를 삽입하는 것으로 결정하면, 각pi를 1로 설정할 수 있다. 예를 들어, P = (1, 1, 1, ..., 1)은 제1 작업집합(W)의 모든 빈 공간을 패딩 페이지로 채워 전체 작업집합을 포괄하는 단일의 입출력 요청을 생성하는 것을 의미할 수 있다. 일 실시예에서, 입출력 패딩부(220)는 제1 작업집합(W)에 대한 부분집합(subset) Wi,j = (bi, ..., bj)를 결정하고, 부분집합에 대한 패딩 벡터를 Pi,j-1 = (pi, pi+1, ..., pj-1)로 결정할 수 있다. 입출력 패딩부(220)는 read(i,j)를 Pj,j-1을 통해 Wi,j를 로드하기 위한 스토리지 액세스 시간으로 결정할 수 있다. 여기에서, read(i,j)는 bi에서 bj까지 구간을 포함하는 단일 입출력 요청에 의해 산출될 수 있다. 입출력 패딩부(220)는 t(i,j)를 최적의 패딩 벡터를 통해 Wi,j를 로드하는 시간으로 결정할 수 있다. 보다 구체적으로, 입출력 패딩부(220)는 아래 수학식1을 통해

에서 t(i,j)와 이의 하위문제인 t(s,t) 사이의 관계를 구별할 수 있다.In an embodiment, when input/output alignment and input/output merging are completed, the input/output padding unit 220 may generate a work set W of the target application composed of n page chunks as the first work set. Here, W = (b0, b1, ..., bn-1), and bi may correspond to page chunks of adjacent pages arranged in the LBA order. The empty space between adjacent page chunks may be greater than or equal to a single page size. Input/output padding unit 220 is input

Given is, we can define the associated padding vector P = (p0, p1, ..., pn-2). Here, pi corresponds to a padding candidate for an empty space between bi and bi+1. In one embodiment, if the input/output padding unit 220 determines to insert a padding page between bi and bi+1, each pi may be set to 1. For example, P = (1, 1, 1, ..., 1) fills all empty spaces in the first work set (W) with padded pages to create a single I/O request covering the entire work set. It can mean. In one embodiment, the input/output padding unit 220 determines a subset Wi,j = (bi, ..., bj) for the first working set W, and calculates a padding vector for the subset. It can be determined as Pi,j-1 = (pi, pi+1, ..., pj-1). The input/output padding unit 220 may determine read(i,j) as a storage access time for loading Wi,j through Pj,j-1. Here, read(i,j) can be calculated by a single input/output request that includes a section from bi to bj. The input/output padding unit 220 may determine t(i,j) as a time to load Wi,j through an optimal padding vector. More specifically, the input/output padding unit 220 is through Equation 1 below.

The relationship between t(i,j) and its subproblem t(s,t) can be distinguished.

[Equation 1]

를 선택하여 Ps를 0으로 결정하고 나머지를 X로 결정할 수 있다. 예를 들어, 입출력 패딩부(220)는 s를 i+1로 결정하면 Pi,j+1를 (X,0,X,...,X)로 t(i,j)를 t(i,i+1)+t(i+2,j)로 결정할 수 있다. 일 실시예에서, 입출력 패딩부(220)는 근본적 하위문제의 계산을 완료한 후에 t(i,j)를 산출할 수 있다. 예를 들어, 근본적 하위문제는 s=i+1에서 Ps에 대한 t(i,i+1) 및 t(i+2,j)에 해당할 수 있다. 일 실시예에서, 입출력 패딩부(220)는 암기와 함께 하향식 접근법을 사용하거나 표와 함께 상향식 접근법을 사용하여 모든 필요한 하위 문제들을 평가할 수 있다. 두 방법 모두 각 하위 문제를 한 번씩만 평가하여 계산 오버 헤드를 크게 줄일 수 있다. 결론적으로, 입출력 패딩부(220)는 t(0,n-1)에 대해 수학식1을 적용함으로써 주어진 입력

에 대해서 최적의 패딩 벡터

를 산출할 수 있다.For example, when i=j, the input/output padding unit 220 does not perform input/output padding because only one page chunk exists in the first working set. When j is greater than i, the input/output padding unit 220 may initialize the padding vector Pi,j-1 to (X,X,...,X). Here, X means a state in which input/output padding has not yet been determined. Thereafter, the input/output padding unit 220 may find the minimum time to load Wi,j through the min function of Equation 1. For example, when t(i,j) becomes read(i,j), the input/output padding unit 220 is for all padding candidates (ie, Pi,j-1 = (1,1,... ,1)) I/O padding can be performed. In addition, the input/output padding unit 220 determines whether or not to pad the input/output, a single padding candidate

By selecting, Ps can be determined as 0 and the remainder as X. For example, if the input/output padding unit 220 determines s as i+1, then Pi,j+1 is (X,0,X,...,X) and t(i,j) is t(i, It can be determined as i+1)+t(i+2,j). In an embodiment, the input/output padding unit 220 may calculate t(i,j) after completing the calculation of the fundamental subproblem. For example, the fundamental subproblem may correspond to t(i,i+1) and t(i+2,j) for Ps at s=i+1. In one embodiment, the input/output padding unit 220 may evaluate all necessary subproblems using a top-down approach with memorization or a bottom-up approach with a table. Both methods can significantly reduce computational overhead by evaluating each subproblem only once. In conclusion, the input/output padding unit 220 applies Equation 1 to t(0,n-1)

The optimal padding vector for

Can be calculated.

The second work set generation unit 230 may generate a second work set in which the size of the input/output is enlarged through input/output padding. More specifically, the second work set generation unit 230 performs input/output padding for the first work set by the input/output padding unit 220, and accordingly, the second work set in which the size of the input/output is enlarged is a final work set. Can be created with As a result, the input/output optimization device 110 may increase the throughput of storage and increase the loading speed of the application by executing the application with the final work set generated through the second work set generation unit 230.

The controller 240 controls the overall operation of the input/output optimization device 110 in the process of the input/output optimization device 110 performing input/output optimization for improving the loading performance of an application, and the first work set generator 210, the input/output Data and control flow between the padding unit 220 and the second work set generation unit 230 may be managed.

3 is a graph showing a throughput of a storage input/output according to an input/output size according to an exemplary embodiment.

In FIG. 3, a graph showing a change in an input/output throughput of a storage according to an enlargement of an input/output size to be performed by an input/output optimization apparatus according to an embodiment of the present invention may be disclosed. 3 is a graph showing throughput according to input/output size of Google Nexus 5 (32GB eMMC storage). For example, the size of the input/output observed during the loading process of the application is observed to be about 43.4KB. As can be seen in FIG. 3, an input/output size of at least 256KB is required to obtain the maximum input/output throughput of the eMMC storage. Accordingly, the input/output optimization device 110 can implement the input/output throughput of the storage close to the maximum input/output throughput by expanding the input/output throughput to at least 256 KB.

4 is a diagram illustrating a method of performing input/output padding according to another embodiment of the present invention.

In FIG. 4, the input/output optimization device 110 may solve t(0,3) for W0,3 = (b0,b1,b2,b3) according to an embodiment. 4(ce) shows that t(0,3) has 6 different subproblems, of which t(0,2) can be recursively obtained from FIG. 4(fh). . For example, if the input/output optimization device 110 has already calculated t(0,0) and t(0,1) in FIG. 4(cd) and stored in the memory, the second calculation in FIG. 4(gh) is only It can be executed by retrieving the stored value from memory.

5 is a diagram illustrating a result of input/output padding according to an embodiment of the present invention.

5(a) corresponds to a part of a logical block map showing the result of input/output padding for a messenger among applications according to an embodiment of the present invention. 5(b) is a diagram showing details of the determined input/output padding.

In FIG. 5A, the input/output optimization device 110 may effectively enlarge the size of adjacent page chunks through input/output padding to increase the size of input/output while loading a messenger application. For example, in FIG. 5 (a), the input/output optimization device 110 indicates that the amount of padding pages in all test applications may correspond to 85.0 MB, which is 7.4% of the size of the original first working set, and as a result, input/output The size of the input/output of the second working set generated according to the padding may increase by 5.6X on average. In FIG. 5B, the input/output optimization device 110 may determine padding for padding candidates of various sizes through input/output padding. Referring to FIG. 5B, for example, the input/output optimization apparatus 110 may preferentially perform input/output padding as the size of the padding candidate is smaller.

6 is a flowchart illustrating an input/output optimization method for improving a loading speed according to an embodiment of the present invention.

In FIG. 6, the input/output optimization device 110 may generate a first work set through input/output scheduling including input/output alignment and input/output merging through the first work set generation unit 210 (step S610). The input/output optimization device 110 may perform input/output padding for filling a padded page between a plurality of page chunks constituting the first work set with respect to the first work set through the input/output padding unit 220 (step S620). The input/output optimizing device 110 may generate a second work set in which the size of the input/output is enlarged by the input/output padding through the second work set generation unit 230 as a final work set (step S630).

7 is a flowchart illustrating an input/output optimization method for improving a loading speed according to another embodiment of the present invention.

In FIG. 7, the input/output optimization device 110 may secure an input/output performance model for input/output of various sizes of a storage device (storage 130) (step S710). In one embodiment, the input/output optimization device 110 secures a performance model by applying input/output commands of various sizes to the flash storage (eMMC, etc.) built into the user terminal 120 executing the optimization target application and measuring time. can do.

The input/output optimization device 110 monitors and analyzes the input/output command of the application to extract a work set when the application is fetched (step S720).

The input/output optimizing device 110 may first perform input/output sorting and input/output merging on page chunks of the extracted work set, and then generate and store an optimal work set through a dynamic programming model (step S730). In one embodiment, the input/output optimization device 110 generates a work set in which input/output alignment and input/output merge is performed as a first work set, and then, optimizes the first work set by performing input/output padding through a dynamic programming model. The resulting second work set can be created and saved as a final work set.

The input/output optimizing device 110 may pause the application so that the application is not loaded by the request paging immediately upon detecting an event such as startup of an application or activation of a background application (step S740).

The input/output optimizing device 110 may resume execution or activation of the application after importing the second work set (step S750).

8 is a graph showing an effect according to an embodiment of the present invention.

FIG. 8(a) is a graph showing loading times of various types of applications according to each input/output optimization step, and FIG. 8(b) is a graph showing the loading performance of applications in the presence of background input/output interference.

In Fig. 8(a), although I/O reordering itself hardly affects the loading time, it can be seen that the loading time is shortened by 19.3% through I/O merging. The input/output optimization device 110 may achieve a reduction in loading time of 29.9% in total through additional input/output padding. In FIG. 8A, the reduction in the loading speed through the input/output optimization device 110 may differ depending on the type of application. For example, in the case of temple run 2, it can be seen that 27.1% of the speed reduction was achieved due to input/output merging, while the additional speed reduction due to input/output padding was only 2.9%P (percentage point). Conversely, in the case of messenger, only 8.0% of speed reduction is obtained through input/output merging, while additional speed reduction of 25.1%P can be obtained through additional application of input/output padding.

In Fig. 8(b), the effect of various (three) types of background input/output interference on the loading time of an application is shown. Video playback, FTP download, and application update show low, medium and high input/output intensity, respectively. In an embodiment, in order to reproduce similar input/output interference, blktrace is applied to obtain a background input/output sequence of each process, and the R/W ratio, input/output per second, and input/output size may be extracted. In conclusion, it can be seen from FIG. 8(b) that as the intensity of background input/output interference increases, the difference between the baseline application performance and the application performance due to optimization through the input/output padding of the input/output optimization device 110 increases.

Although the above has been described with reference to preferred embodiments of the present invention, those skilled in the art will variously modify and change the present invention within the scope not departing from the spirit and scope of the present invention described in the following claims. You will understand that you can do it.

100: I/O optimization system to improve application loading performance
110: I/O optimization device for improving application loading performance
120: user terminal
130: storage
210: first work set generation unit 220: input/output padding unit
230: second work set generation unit
240: control unit

Claims (9)

A first work set generation unit for generating a first work set through input/output scheduling including input/output alignment and input/output merging;
An input/output padding unit for performing input/output padding for filling the padded page between a plurality of page chunks constituting the first work set, for the generated first work set; And
An input/output optimizing device for improving loading performance of an application, comprising: a second work set generation unit generating a second work set in which the size of the input/output is enlarged through the input/output padding.
The method of claim 1, wherein the first work set generation unit
An input/output optimization device for improving loading performance of an application, comprising arranging the plurality of page chunks in a logical block address (LBA) order through the input/output alignment.
The method of claim 2, wherein the first work set generation unit
I/O for improving the loading performance of an application, comprising merging a plurality of page chunks adjacent to each other among the plurality of aligned page chunks into a single page chunk through the input/output merging to generate the first working set Optimization device.
The method of claim 1, wherein the input/output padding unit
And defining a padding vector for determining blank spaces between the plurality of page chunks as padding candidates.
The method of claim 4, wherein the input/output padding unit
An input/output optimization device for improving loading performance of an application, characterized in that detecting a padding vector having a minimum loading time for the first working set through a dynamic programming model.
The method of claim 5, wherein the input/output padding unit
And detecting the padding vector by considering the size of the padding candidates or the spacing between adjacent page chunks surrounding each of the padding candidates through the dynamic programming model.
The method of claim 6, wherein the input/output padding unit
I/O for improving loading performance of an application, comprising detecting the padding vector by considering the size of adjacent page chunks surrounding each of the padding candidates and whether input/output padding is performed for other adjacent padding candidates through the dynamic programming model Optimization device.
The method of claim 7, wherein the input/output padding unit
An input/output optimizing apparatus for improving application loading performance, characterized in that it is dynamically determined whether to perform the input/output padding for the padding candidates based on the detected padding vector.
Generating a first work set through input/output scheduling including input/output alignment and input/output merging;
Performing input/output padding for filling a padded page between a plurality of page chunks constituting the first work set on the generated first work set; And
And generating a second work set in which the size of the input/output is enlarged through the input/output padding as a final work set.

Images