KR102541455B1 - Transcoding task allocation Method and System for Cost-Optimization Considering Live Video Streaming Quality in Edge-Computing Environment - Google Patents

Abstract

A method and system for allocating transcoding tasks for cost optimization considering live video streaming quality in an edge computing environment are presented. The transcoding job allocation method for cost optimization considering live video streaming quality in the edge computing environment proposed in the present invention is performed in the central transcoding server in the order of cost-effectiveness compared to popularity-weighted video quality in the edge computing environment through the collection server. Priority is given to the transcoding job to be performed, and the corresponding transcoding job is transferred to the central transcoding server or the edge server according to the network range and computational load for each transcoding job given the priority through the collection server. Allocating, reallocating or removing some of the work through the aggregation server to the central transcoding server operated by a live streaming provider in order to minimize the effect of network coverage and computational load on the popularity-weighted video quality. and providing live video streaming to a plurality of viewer terminals through a transmission server after performing the transcoding task allocated through the collection server in the central transcoding server or the edge server.

Description

{Transcoding task allocation method and system for Cost-Optimization Considering Live Video Streaming Quality in Edge-Computing Environment}

The present invention relates to a transcoding job allocation method and system for cost optimization considering live video streaming quality in an edge computing environment.

Live streaming of real-time content, such as video sports events, is gaining popularity. For example, Twitch, one of the top gaming live streaming platforms, accounts for 1.8% of US internet traffic. In a live streaming scenario, a broadcaster's channel continuously captures and encodes a video stream, uploads it to a local server, and then delivers it to the viewer who requests it. This continuous computing and transfer process requires a lot of computing resources.

To accommodate changing network conditions, live streaming services typically use dynamic adaptive streaming over HTTP (DASH). Here, each source video stream is transcoded into a version with a different bitrate so that the video delivered to the viewer matches the network bandwidth. But because transcoding requires more computation than a single regional server can do, broadcasters only offer a limited number of versions of video streams. For example, Twitch.tv only offers transcoding services for premium channels. Therefore, many viewers receive video at sub-optimal bitrates, degrading viewer QoE.

For this, service providers can use cloud services, but if the cloud server is far from the broadcasting source, it will not meet the latency requirements of live streaming. In the edge computing environment that has emerged as a result, an edge server (ES) is placed near the data to be processed. Gartner predicts that by 2023, more than 50% of large enterprises will have six or more Internet of Things (IoT) applications that require edge computing, and the number of edge infrastructures will increase.

Live streaming service providers can use MEC to transcode video streams on edge servers closer to the channel location, offloading the transcoding workload previously delivered to central aggregation servers. Because it is in the streaming data path, latency is hardly increased, but each edge server's CPU processing capacity and network range are limited, requiring effective allocation of transcoding tasks. It also needs to be able to minimize the total cost of using edge infrastructure.

Existing techniques aimed at reducing the computational load in video streaming using edge computing are mostly applied to VoD systems, and their contribution to live streaming systems that require continuous real-time transcoding is insufficient. In a live streaming system, several techniques for using a viewer's device for some transcoding work have been proposed, but it is difficult to provide a reliable service due to the instability of a viewer's connection.

Korean Patent Registration No. 10-1569502 (2015.11.10)

The technical problem to be achieved by the present invention is to provide a method and system for maximizing total popularity-weighted video quality under cost constraints by controlling the utilization rate of the edge server of a live streaming server in an edge computing environment. For transcoding in the live streaming system using edge server computing, each channel is streamlined to maximize the overall video quality transmitted to the user terminal within the service provider's cost limit constraints, each edge server's processing capacity and network coverage constraints. You select the bitrate version you need and assign each edge server the transcoding job that creates it.

In one aspect, the transcoding job allocation method for cost optimization considering live video streaming quality in an edge computing environment proposed in the present invention is based on popularity-weighted video quality (PWQ) in an edge computing environment through a collection server. ) prioritizing the transcoding jobs performed in the central transcoding server in order of high cost efficiency, the network coverage and computational load for each transcoding job given the priority through the collection server. Allocating the corresponding transcoding task to the central transcoding server or the edge server according to the above, in order to minimize the effect of the network coverage and computational load on the popularity-weighted video quality, a live streaming provider operates some of the work through the collection server After the central transcoding server or the edge server performs the reassignment or removal step of the central transcoding server and the transcoding task assigned through the collection server, the live broadcast to a plurality of viewer terminals through the transmission server. and providing video streaming.

In the step of prioritizing the transcoding work performed in the central transcoding server in the order of cost efficiency compared to popularity-weighted video quality in the edge computing environment through the aggregation server, a list of channels exists in the aggregation server, and each Capture, encode and upload the video stream of the channel, define the version of the video stream with the bitrate requested from multiple viewer terminals as a combination of binary values, calculate the popularity-weighted video quality, and limit the processing capacity and network range Allocate a transcoding task according to the popularity-weighted video quality to maximize the video quality transmitted to a plurality of user terminals in the Remove.

The step of allocating the corresponding transcoding job to the central transcoding server or the edge server according to the network range and computational load for each transcoding job given the priority through the aggregation server is to ensure that each channel is connected to the network of the transcoding server. Using a binary variable indicating whether it is within the range, a binary value is given according to whether or not the transcoding operation for the lowest bitrate version of the channel is performed.

The step of providing live video streaming to a plurality of viewer terminals through a transmission server after performing the transcoding task allocated through the collection server in the central transcoding server or the edge server, If the version having the bit rate has not been transcoded, the version having the highest bit rate among versions lower than the requested bit rate among the versions for which the transcoding task has been executed is transmitted.

In another aspect, the transcoding job allocation system for cost optimization considering live video streaming quality in an edge computing environment proposed by the present invention compares popularity-weighted video quality (PWQ) in an edge computing environment. An aggregation server that prioritizes transcoding jobs performed on the central transcoding server in order of cost efficiency, and according to the network coverage and computational load for each transcoding job given the priority through the aggregation server. A central transcoding server and an edge server assigned corresponding transcoding tasks - the central transcoding server reallocates some tasks through the aggregation server to minimize the effect of network coverage and computational load on the popularity-weighted video quality receiving or some of the work is removed - and a transmission server that provides live video streaming to a plurality of viewer terminals after performing the transcoding work assigned through the collection server in the central transcoding server or the edge server.

According to embodiments of the present invention, the CPU processing capacity of each edge server, wireless network range, Transcoding tasks can be assigned to each edge server, taking into account the broadcaster's cost budget and the viewer's video quality. In addition, according to an embodiment of the present invention, the proposed technique can achieve an average of 1.4 times more PWQ than the conventional technique in the same cost budget, and can contribute to effective cost management of the live streaming system in the MEC environment. .

1 is a diagram showing the configuration of a transcoding job allocation system for cost optimization considering live video streaming quality in an edge computing environment according to an embodiment of the present invention.
2 is a flowchart illustrating a transcoding job assignment method for cost optimization considering live video streaming quality in an edge computing environment according to an embodiment of the present invention.
3 is a diagram illustrating a transcoding task determination and edge allocation algorithm according to an embodiment of the present invention.
4 is a diagram illustrating an algorithm for reallocating and removing transcoding jobs for cost reduction according to an embodiment of the present invention.
5 is a diagram comparing popularity-weighted quality according to cost limitation between a transcoding job allocation technique according to an embodiment of the present invention and a conventional technique.
6 is a diagram comparing popularity-weighted quality according to the number of channels between a transcoding job allocation technique according to an embodiment of the present invention and a conventional technique.
7 is a diagram comparing popularity-weighted quality according to the number of edge servers between a transcoding job allocation technique according to an embodiment of the present invention and a conventional technique.
8 is a diagram comparing popularity-weighted quality in an on-off cost model according to cost limitation between a transcoding job assignment technique according to an embodiment of the present invention and a technique in the prior art.
9 is a diagram comparing popularity-weighted quality of a transcoding task assignment technique according to an embodiment of the present invention in a linear cost model and an on-off cost model.

In a live video streaming system using a multiple-access edge-computing (MEC) architecture, the present invention provides CPU processing capacity of each edge server, wireless network range, broadcaster's cost budget and viewers. A method and system for allocating transcoding jobs to each edge server in consideration of video quality. The present invention relates to a method and system for solving a cost problem through utilization control of each edge server in a system performing live streaming in an edge computing environment.

The present invention first prioritizes transcoding jobs that are more cost-effective than Popularity-Weighted Video Quality (PWQ), and assigns each job to an edge server in consideration of network coverage and computational load. Thereafter, cost budgets are met by reassigning or removing some of the work to a central transcoding server operated by the live streaming provider, while minimizing the impact of these changes on the overall PWQ.

In this way, we propose a heuristic algorithm that optimizes the service provider's cost limit by allocating the transcoding job to the edge server in consideration of the streaming video quality and the channel network of the edge server.

According to simulation results according to an embodiment of the present invention, the proposed technique achieves an average of 1.4 times more PWQ than the conventional technique in the same cost budget. This difference is more effective with smaller cost margins, more channels, and fewer edge servers.

In addition, according to the experimental results according to the embodiment of the present invention, it is shown that the proposed technique can contribute to effective cost management of the live streaming system in the MEC environment. Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a diagram showing the configuration of a transcoding job allocation system for cost optimization considering live video streaming quality in an edge computing environment according to an embodiment of the present invention.

In the proposed edge computing environment, the transcoding job allocation system for cost optimization considering live video streaming quality includes an ingest server (110), a central transcoding server (CTS) (120), an edge server ( Edge server; ES) (131, 132, ..., 133) and delivery server (Delivery server) (140).

The collection server 110 according to an embodiment of the present invention prioritizes the transcoding tasks performed by the central transcoding server in order of cost-effectiveness compared to Popularity-Weighted Video Quality (PWQ) in an edge computing environment. grant

A list of channels exists in the collection server 110 according to an embodiment of the present invention, and video streams of each channel are captured, encoded, and uploaded.

The collection server 110 according to an embodiment of the present invention defines a video stream of a version having a bit rate requested from a plurality of viewer terminals (Users) as a combination of binary values and calculates a popularity-weighted video quality. Then, the transcoding task is allocated according to the popularity-weighted video quality to maximize the video quality transmitted to a plurality of user terminals within the processing capacity constraint and the network range constraint, and then the popularity-weighted video quality until the cost budget constraint is met. Remove the assigned tasks to minimize losses.

The central transcoding server 120 and the edge servers 131, 132, ..., 133 according to an embodiment of the present invention have network coverage and computational load for each transcoding job prioritized through the aggregation server. According to this, the corresponding transcoding task is assigned.

The central transcoding server 120 may have some tasks reassigned or some tasks removed through the aggregation server 110 to minimize the effect of network coverage and computational load on the popularity-weighted video quality.

The central transcoding server 120 and the edge servers 131, 132, ..., 133 according to an embodiment of the present invention use a binary variable indicating whether each channel is within the network range of the transcoding server, A binary value is given depending on whether or not transcoding is performed for the lowest bitrate version of the channel.

The transmission server 140 according to an embodiment of the present invention performs the transcoding task allocated through the collection server 110 in the central transcoding server 120 or the edge server 131, 132, ..., 133 Then, live video streaming is provided to a plurality of viewer terminals (Users).

The transmission server 140 according to an embodiment of the present invention, when the version having the requested bit rate from a plurality of viewer terminals (Users) does not perform the transcoding job, the requested bit rate among the versions for which the transcoding job has been executed. Among versions lower than the rate, the version with the highest bit rate is transmitted.

Referring to FIG. 1, the proposed transcoding task allocation system for cost optimization in consideration of live video streaming quality in an edge computing environment will be described in more detail.

According to an embodiment of the present invention, a broadcaster's video source is transmitted in each channel, and these channels are managed in the aggregation server 110. In the collection server 110, a list of channels exists, and the video stream of each channel is continuously captured, encoded, and uploaded. The original video stream of each channel is provided by the central transcoding server 120 operated by the live streaming provider. Can be transcoded. In addition, the upload path from the channel to the collection server 110 passes through or is close to a plurality of edge servers 131, 132, ..., 133, and these edge servers 131, 132, ..., 133 Assume that it can be used for transcoding.

The aggregation server 110 determines which server will execute each transcoding job. Then, the delivery server 140 may provide a video version at an appropriate bit rate to a plurality of viewer terminals.

(j=0, ..., N^ES)를 CPU 프로세싱 능력이 가장 가장 높은 TS에 대비한 비율로 표현한다.N ^ES is the number of edge servers, and the total number of transcoding servers (TS) is N ^ES +1. This is because TS 0 is a central transcoding server (CTS). Since different TSs have different CPU processing capacities, in the present invention, the CPU processing capacities of TS j

(j=0, ..., N ^ES ) is expressed as a ratio of TS with the highest CPU processing capability.

는 각 채널 i가 TS j의 네트워크 범위 내에 있는지 여부를 나타내는 이진 변수이다. i 채널이 TS j의 네트워크 범위 내에 있으면

; 그렇지 않으면

이다. 모든 트랜스코딩 작업은 CTS(TS 0)에서 실행할 수 있으므로

이다. Each edge server is assigned to a base station to handle the transcoding task of channels within network coverage. Therefore, each edge server has network range restrictions according to its distance from the channel.

is a binary variable indicating whether each channel i is within network range of TS j. If channel i is within network range of TS j

; Otherwise

am. All transcoding jobs can run on CTS (TS 0), so

am.

2 is a flowchart illustrating a transcoding job assignment method for cost optimization considering live video streaming quality in an edge computing environment according to an embodiment of the present invention.

The proposed transcoding task allocation method for cost optimization considering the live video streaming quality in the edge computing environment is cost-effective compared to Popularity-Weighted Video Quality (PWQ) in the edge computing environment through the collection server. Step 210 of prioritizing transcoding jobs performed in the central transcoding server in order (210). For each transcoding job given the priority through the collection server, the corresponding transcoding job is based on the network coverage and computational load. Allocating coding work to a central transcoding server or edge server (220), in order to minimize the effect of network coverage and computational load on the popularity-weighted video quality, some work is operated by a live streaming provider through the collection server Reassigning or removing (230) to the central transcoding server, and after performing the transcoding task assigned through the collection server in the central transcoding server or the edge server, a plurality of viewers through the transmission server and providing a live video streaming to the terminal (240).

In step 210, transcoding jobs performed in the central transcoding server are prioritized in the order of cost-effectiveness to popularity-weighted video quality in the edge computing environment through the aggregation server.

A list of channels exists in the collection server according to an embodiment of the present invention, and video streams of each channel are captured, encoded, and uploaded.

The collection server according to an embodiment of the present invention defines a video stream of a version having a bitrate requested from a plurality of viewer terminals as a combination of binary values and calculates a popularity-weighted video quality. Then, the transcoding task is allocated according to the popularity-weighted video quality to maximize the video quality transmitted to a plurality of user terminals within the processing capacity constraint and the network range constraint, and then the popularity-weighted video quality until the cost budget constraint is met. Remove the assigned tasks to minimize losses.

In step 220, for each transcoding job given the priority through the aggregation server, the corresponding transcoding job is allocated to the central transcoding server or the edge server according to the network coverage and computational load.

The central transcoding server and the edge server according to an embodiment of the present invention are assigned transcoding jobs given priority through the aggregation server according to the network range and calculation load.

The central transcoding server and the edge server according to an embodiment of the present invention use a binary variable indicating whether each channel is within the network range of the transcoding server, and whether or not the transcoding operation for the lowest bit rate version of the corresponding channel is executed. It is given a binary value according to

In step 230, in order to minimize the impact of network coverage and computational load on popularity-weighted video quality, some tasks through the aggregation server are reassigned to the central transcoding server operated by the live streaming provider or removed. .

In step 240, the central transcoding server or the edge server performs the transcoding task allocated through the aggregation server, and then provides live video streaming to a plurality of viewer terminals through the transmission server.

The transmission server according to an embodiment of the present invention, when a version having a bit rate requested from a plurality of viewer terminals does not perform a transcoding job, among versions in which a transcoding job has been executed, the highest version among the versions lower than the requested bit rate Send a version with a higher bitrate.

Hereinafter, a transcoding task assignment process for cost optimization in consideration of live video streaming quality in the proposed edge computing environment will be described in more detail.

3 is a diagram illustrating a transcoding task determination and edge allocation algorithm according to an embodiment of the present invention.

4 is a diagram illustrating an algorithm for reallocating and removing transcoding jobs for cost reduction according to an embodiment of the present invention.

는 채널 i의 k번째 가장 낮은 비트레이트 버전 (i=1, ..., N^ch 및 k=1, ..., N^ver)이다. 가장 높은 비트레이트 버전은 항상 사용할 수 있으므로 트랜스코딩으로 생성될 수 있는 버전 N^tran의 수는 N^ver-1이다.N ^ch is the total number of channels. N ^ver is the number of different bitrate versions used.

is the kth lowest bitrate version of channel i (i=1, ..., N ^ch and k=1, ..., N ^ver ). Since the highest bitrate version is always available, the number of versions N ^tran that can be generated by transcoding is N ^ver -1.

를 버전

, (i=1,..., N^ch 및 k=1, ..., N^tran)에 대한 트랜스코딩 작업이라고 가정한다. 각 작업

의 엣지 서버 CPU 사용량 요구 사항

를 가장 강력한 ES의 CPU 프로세싱 능력에 대비한 비율로 표현한다.

version

, (i = 1, ..., N ^ch and k = 1, ..., N ^tran ). each task

Edge server CPU usage requirements of

is expressed as a ratio of the CPU processing power of the most powerful ES.

를 버전

의 액세스 확률, 즉 일반 시청자가 시청할 확률이라고 한다. 각 채널의 확률은 현재 조회수에서 파생될 수 있다.

를

버전의 비디오 품질 인덱스라고 하자.

의 값은 비디오 품질 측정 도구를 사용하여 얻을 수 있다. 본 발명에서는 Netflix에서 개발한 VMAF(Video Multi-Method Assessment fusion) 지수를 사용하여 이를 표현하였다. 본 발명에서는 더 인기 있는 채널에 더 높은 품질을 제공하기 위해

버전의 인기 가중치 동영상 품질(PWQ),

을 도입하고, 아래와 같이 정의 한다:

version

is referred to as the access probability of, that is, the probability that a general viewer will watch. The probability of each channel can be derived from the current number of hits.

cast

Let be the video quality index of the version.

The value of can be obtained using a video quality measurement tool. In the present invention, this is expressed using the VMAF (Video Multi-Method Assessment fusion) index developed by Netflix. In the present invention, in order to provide higher quality to more popular channels

Popularity-weighted video quality (PWQ) of the version;

is introduced, and defined as:

를 사용하여 버전

의 트랜스코딩을 통한 생성 여부를 나타낸다. 즉

가 실행되면

; 그렇지 않으면

이다.binary variable

version using

Indicates whether to generate through transcoding. in other words

is executed

; Otherwise

am.

In live streaming, each viewer terminal can be expected to request the highest bitrate version that can be supported in the terminal and current network conditions. If the bitrate of the delivered version is higher than requested, rebuffering can significantly reduce QoE, which may cause playback to stop altogether. To prevent this in any case, all video should be transcoded to the lowest bitrate offered by the channel in question.

라고 하며 아래와 같이 정의한다: If the version with the requested bit rate cannot be used because it has not been transcoded, the version with the highest bit rate among the transcoded versions lower than the requested bit rate is transmitted. If the requested bitrate is k, this version of the video stream

and is defined as:

는

값의 조합에 의해 결정되므로 채널 i의 총 PWQ는 다음과 같이 계산할 수 있다:

Is

Since it is determined by the combination of values, the total PWQ of channel i can be calculated as:

Next, the cost of using the edge server according to an embodiment of the present invention will be described.

는 정규화된 비용 값

과 CPU 사용량 u에 비례한다. (

이다) 이는 아래와 같이 표현할 수 있다:Live streaming service providers have to pay to use edge servers. Embodiments of the present invention primarily consider a linear cost model in which cost is proportional to CPU usage. Cost to use ES j

is the normalized cost value

and is proportional to the CPU usage u. (

), which can be expressed as:

은 일정 기간 동안 ES j를 사용할 때 지불하는 가장 비싼 ES의 비율로 계산된 정규화된 비용이다. 그러므로 온-오프 비용 모델에서의

는 다음과 같이 결정할 수 있다: We also consider an on-off cost model in which the cost of using an edge server is determined by whether it is used or not for a given period of time, regardless of throughput.

is the normalized cost calculated as a percentage of the most expensive ES paid when using ES j for a certain period of time. Therefore, in the on-off cost model

can be determined as follows:

가 실행되는 트랜스코딩 서버의 인덱스를

라고 하자. TS를 사용할 수 없고

0이면

가 -1로 설정된다. According to an embodiment of the present invention, each task in TS j

index of the transcoding server that is running

let's say TS is not available

if 0

is set to -1.

에 대한

및

값을 찾는 것이다: 프로세싱 용량 제약, 네트워크 범위 제약 및 네트워크 범위 제약. The goal of the present invention is to maximize the overall PWQ under the following three constraints

for

and

It's about finding values: processing capacity constraints, network range constraints and network range constraints.

를 초과해서는 안 된다. In the processing capacity constraint according to an embodiment of the present invention, the total CPU usage required to execute all tasks in TS j is the maximum processing capacity as follows

should not exceed

In network coverage constraints according to an embodiment of the present invention, each transcoding task must be executed on an edge server within network range of a channel or central transcoding server as follows:

을 초과해서는 안 된다: 1) In the cost budget constraint according to an embodiment of the present invention, the total cost incurred is the cost budget of the live streaming service provider as follows

must not exceed:

에 대한

및

값을 찾는 문제인 트랜스코딩 작업 할당 문제(TTAP)를 공식화할 수 있으며, 다음과 같다: Each task now maximizes the overall PWQ

for

and

We can formulate the transcoding task assignment problem (TTAP), which is a problem of finding values, and is as follows:

TTAP can be reduced to a multiple-choice multi-knapsack problem with assignment restrictions (MMPAR) known as NP-hard.

In the present invention, a two-step heuristic is introduced to solve this problem. First, what to execute during the transcoding job is determined in consideration of maximization of PWQ, and allocated to the transcoding server within the limitation of processing capacity and network range. Next, the tasks allocated in the first step are eliminated one by one, with minimal PWQ loss, until the cost budget constraint is met.

에 대해

의 임시 값을 보유하는 임시 이진 변수를

라고 한다. Referring to FIGS. 3 and 4 , the first step is determining a transcoding task and allocating an edge. Each task during algorithm execution

About

A temporary binary variable holding the temporary value of

It is said.

는 실행할 작업 집합에서

를 제거했을 때의 PWQ를 나타내며 아래와 같이 표현할 수 있다: variable

in the set of tasks to run

It represents the PWQ when is removed and can be expressed as:

의 모든 값은 1로 초기화되므로

는 초기에는

제거로 인한 PWQ의 변화를 나타낸다.

All values of are initialized to 1, so

is initially

Shows the change in PWQ due to removal.

대

의 비율인

를 결정할 수 있다. 이는 아래와 같다.now

big

which is the ratio of

can decide This is shown below.

의 값이 높을수록 PWQ에 미치는 영향이 크기 때문에

를 제거하는 것은 바람직하지 않다. 따라서

값이 더 높은 작업이 우선적으로 할당된다.

The higher the value of , the greater the effect on PWQ.

It is not advisable to remove thus

Jobs with higher values are assigned priority.

를 통해 각 ES j에 추가 작업을 할당하는 가능성 및 타당성을 표현했다.

의 값은 해당 ES에서 남은 프로세싱 용량과 비용을 결합한 식으로 비용 모델에 따라 다르게 정의된다. variable to determine which ES to assign the task to.

expressed the possibility and feasibility of assigning an additional task to each ES j through

The value of is defined differently depending on the cost model by combining the remaining processing capacity and cost in the corresponding ES.

값이 더 높은 작업을, 남은 프로세싱 용량이 많고 사용 비용이 낮은 ES에 할당해야 한다. 따라서

의 값은 아래와 같으며,

값이 더 높은 ES에 우선적으로 태스크를 할당한다: In the linear model according to an embodiment of the present invention, to increase the utilization of the edge server cost-effectively

Tasks with higher values should be assigned to ESs with more remaining processing capacity and lower usage costs. thus

The value of is as follows,

Tasks are assigned preferentially to ESs with higher values:

값이 더 높은 ES에 작업을 우선적으로 할당한다: In the on-off model according to an embodiment of the present invention, when using the on-off cost model, assigning one task to E generates a total fixed cost. Therefore, distributing task execution among ESs is disadvantageous. With this model, in order to get the most out of ES

Assign tasks preferentially to ESs with higher values:

(i=1,..., N^ch)을 제일 먼저 ES에 할당한 다음, 다른 비트레이트 버전들을 ES에 할당한다. 알고리즘이 끝나면 TS j에 할당된 작업의 배열

를 구할 수 있다.Fig. 3 shows the pseudocode of the step of determining the transcoding task and allocating the edge. The lowest bitrate version should always be transcoded, so

(i=1,..., N ^ch ) is first assigned to ES, and then other bitrate versions are assigned to ES. array of tasks assigned to TS j at the end of the algorithm

can be obtained.

In the step of reallocating and removing transcoding work for cost reduction of the algorithm according to an embodiment of the present invention, if the cost budget constraint is not met in the first step, the task assigned to the ES while attempting to reduce the overall PWQ as little as possible Some of them need to be removed.

). 이 작업이 할당된 엣지 서버의 PWQ 대 CPU 사용량의의 비율은 중앙 트랜스코딩 서버에 적용되는 비율보다 클 수 있다. 이 경우 제거된 작업을 중앙 트랜스코딩 서버로 재배포할 수 있으며, 이 재배포는 비용 모델에 따라 다르게 동작한다.Let A ^ES be the set of tasks assigned to ESs in the first step. (A ^ES =

). The ratio of CPU usage to PWQ of the edge servers assigned this task may be greater than the ratio applied to the central transcoding server. In this case, the removed jobs can be redistributed to the central transcoding server, and this redistribution behaves differently depending on the cost model.

를 다음과 같이 계산한다: In the linear model according to an embodiment of the present invention, since the cost of using each ES increases according to the amount of processor used, a variable expressing cost efficiency for each task.

is calculated as:

값이 가장 낮은 작업은 할당 해제 후보들을 담은 집합 A^del에 추가되고 이 프로세스는 비용 예산 제약을 만족할 때까지 반복된다.

The task with the lowest value is added to the set of deallocation candidates, A ^del , and this process is repeated until the cost budget constraint is satisfied.

는 다음과 같이 계산할 수 있다: In the on-off model according to an embodiment of the present invention, all tasks allocated to one or more ESs must be removed in order to satisfy cost budget constraints. Cost effectiveness variable for ES j

can be calculated as:

값이 가장 낮은 엣지 서버에 할당된 작업들은 집합 A^del에 추가되고 이 프로세스는 비용 예산 제약을 만족할 때까지 반복된다.

Tasks assigned to the edge server with the lowest value are added to the set A ^del , and this process is repeated until the cost budget constraint is satisfied.

으로 대체되며, A₀는 새로운 A^del에 포함된 가장 낮은 비트레이트 버전의 작업들로 구성된다. 그런 다음

값이 가장 높은 작업을 A^del에서 A₀ 로 반복적으로 이동하여 CTS의 용량 제한 내에서 전체 PWQ를 증가시킨다. 해당 단계가 완료되면 A^ES 및 A₀ 집합을 통해

및

를 설정할 수 있다. Finally, you need to decide which jobs to transcode on the central transcoding server. It is determined by the union of the working set A ₀ found in the first step and the working set A ^del deallocated from ES. To achieve this, A ^del is

, where A ₀ consists of the lowest bitrate version jobs included in the new A ^del . after that

Iteratively moves the highest-valued task from A ^del to A ₀ to increase the overall PWQ within the capacity limits of the CTS. When that step is complete, through the set of A ^ES and A ₀

and

can be set.

According to an embodiment of the present invention, N ^ch is basically set to 6000, which is the number of intermediate channels simultaneously broadcast on Twitch.tv. In the present invention, a bitrate set provided by Zencoder, a public transcoding cloud provider, was used, and according to each resolution, values of a and b in Table 1 below and the bitrate of the version to be generated, r (in Mbps), are used. The CPU usage of each task was obtained by substituting it into the regression model calculation formula, arb.

Table 1 is a parameter that determines the CPU frequency (GHZ) required to transcode a 1080p 2,750kbps version to a different bitrate version for live streaming.

<Table 1>

값은 Twitch.tv에서 얻은 값인 각각 0.399와 14260로 설정했다.The VMAF value is modeled as a normal distribution, the VMAF value for the version at each bitrate is obtained through measurements, the access probability of each channel is modeled as a gamma distribution, and k and

The values were set to 0.399 and 14260 respectively, the values obtained from Twitch.tv.

의 평균값과

의 분산을 갖는 정규분포를 사용했다.

값은 독립적으로 설정되며, 각 사례에서 가장 인기 버전의 우세를 결정한다. 본 발명에서는 중간 비트일수록 인기도가 높을 경우를 디폴트로 하여,

를 1,

=Nver/2 를 사용했다.To express the relative popularity of different versions,

and the average value of

A normal distribution with a variance of .

The values are set independently and determine the dominance of the most popular version in each case. In the present invention, by default, the higher the popularity of the middle bit,

1,

=Nver/2 was used.

We used actual cellular network location data from Melbourne, Australia to determine the location of our edge servers. In addition, the CPU characteristics of the edge server were based on Asus RS620SA-E10-RS12, Dell PowerEdge R7525, HPE Apollo XL225n Gen10 Plus, Fujitsu Primergy RX4770 M6, and Lenovo Technology ThinkSystem SR665 released in 2020 or 2021.

를 얻었으며 이는 0.63에서 1 사이의 값이었다. 본 발명에서는 0.63에서 1 사이에서 랜덤으로

를 선택했다.By default, it uses a linear cost model. Normalized cost values from six cost models (c4, c5n, c5a, c5, c6gn, and c6g) associated with computing resources operated by Amazon Web Services.

was obtained, which ranged from 0.63 to 1. In the present invention, randomly between 0.63 and 1

has chosen

In the present invention, six benchmark combinations of ES allocation and task selection were compared with the proposed transcoding task allocation scheme. The six benchmark combinations are RA (random ES allocation)+AP (version transcoding of all popular channels), RA+HPF (most popular task first), LUF (least utilized ES first)+AP, LUF+ These are HPF, LCF (least cost ES first)+AP and LCF+HPF. The algorithm of the transcoding task assignment scheme proposed in the present invention is called TDA+CR.

5 is a diagram comparing popularity-weighted quality according to cost limitation between a transcoding job allocation technique according to an embodiment of the present invention and a conventional technique.

로 사용하는 비용에 대한 비용 예산의 비율이며 다음과 같다:

When N ^ch is 6000 and N ^ES is 100, the effect of cost budget on total PWQ was investigated. For this purpose, the C ^ratio variable was introduced. This is the maximum utilization rate of ES j

It is the ratio of the cost budget to the cost of using it as:

TDA+CR, a technique proposed in the present invention, always achieves the best performance, generating 1.06 to 2.13 times (an average of 1.45 times) more PWQs than other methods. The difference tends to increase as the C ^ratio decreases, indicating that the proposed technique is effective in coping with the reduced cost budget.

6 is a diagram comparing popularity-weighted quality according to the number of channels between a transcoding job allocation technique according to an embodiment of the present invention and a conventional technique.

When the C ^ratio is 50% and the N ^ES is 100, the effect of the cost budget on the total PWQ was investigated. The proposed technique, TDA+CR, achieves 1.08 to 1.97 times (average 1.44 times) more PWQ than other methods.

In particular, it achieves a PWQ that is 1.08 to 1.19 times (average 1.14 times) higher than LCF+HPF, the highest among benchmark methods.

As the number of channels increases, the higher bitrate version is transcoded less, so the PWQ drops. This fall is better because the proposed technique is much smaller than the comparative benchmark scheme.

7 is a diagram comparing popularity-weighted quality according to the number of edge servers between a transcoding job allocation technique according to an embodiment of the present invention and a conventional technique.

As expected, the total PWQ increases with the increase in the number of edge servers. This is because more processing capacity is available for transcoding.

The proposed technique obtains 1.05 to 2.03 times (average 1.45 times) more PWQ than other methods. This difference increases as the number of edge servers decreases, which shows that the proposed technique works more effectively than other techniques when the total processing capacity of edge servers is low.

8 is a diagram comparing popularity-weighted quality in an on-off cost model according to cost limitation between a transcoding job assignment technique according to an embodiment of the present invention and a technique in the prior art.

의 각 값을

로 설정했다. All results presented so far were obtained using a linear cost model. Therefore, we conducted experiments on an on-off cost model. For a fair comparison with the linear cost model

each value of

set to

Also, since the LCF method allocates the next task based on current cost usage, it cannot be directly used for on-off cost models that do not change in this model. Therefore, the present invention introduces LCF-ONOFF, which selects the cheapest ES and allocates tasks until the ES is fully utilized.

9 is a diagram comparing popularity-weighted quality of a transcoding task assignment technique according to an embodiment of the present invention in a linear cost model and an on-off cost model.

When using the on-off cost model, when N ^ch is 6000 and N ^ES is 100, the effect of cost budget on total PWQ was investigated. The proposed technique achieves the best performance even in the on-off cost model, obtaining 1.06 to 1.98 times (average 1.34 times) more PWQ than other methods. This gap is more pronounced when cost budgets are low.

A linear cost model always has a higher PWQ than an on-off cost model. This difference is between 0.24% and 6.63% (mean 2.34%) and increases with decreasing C ^ratio . This is because the on-off cost model forces the use of a relatively small number of ESs, whereas the linear cost model allows more frequent allocation of the best ES for a particular task. Through these experimental results, it was confirmed that the proposed method is always excellent.

The devices described above may be implemented as hardware components, software components, and/or a combination of hardware components and software components. For example, devices and components described in the embodiments may include, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable array (FPA), It may be implemented using one or more general purpose or special purpose computers, such as a programmable logic unit (PLU), microprocessor, or any other device capable of executing and responding to instructions. A processing device may run an operating system (OS) and one or more software applications running on the operating system. A processing device may also access, store, manipulate, process, and generate data in response to execution of software. For convenience of understanding, there are cases in which one processing device is used, but those skilled in the art will understand that the processing device includes a plurality of processing elements and/or a plurality of types of processing elements. It can be seen that it can include. For example, a processing device may include a plurality of processors or a processor and a controller. Other processing configurations are also possible, such as parallel processors.

Software may include a computer program, code, instructions, or a combination of one or more of the foregoing, which configures a processing device to operate as desired or processes independently or collectively. The device can be commanded. Software and/or data may be any tangible machine, component, physical device, virtual equipment, computer storage medium or device, intended to be interpreted by or provide instructions or data to a processing device. can be embodied in Software may be distributed on networked computer systems and stored or executed in a distributed manner. Software and data may be stored on one or more computer readable media.

The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer readable medium. The computer readable medium may include program instructions, data files, data structures, etc. alone or in combination. Program commands recorded on the medium may be specially designed and configured for the embodiment or may be known and usable to those skilled in computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic media such as floptical disks. - includes hardware devices specially configured to store and execute program instructions, such as magneto-optical media, and ROM, RAM, flash memory, and the like. Examples of program instructions include high-level language codes that can be executed by a computer using an interpreter, as well as machine language codes such as those produced by a compiler.

As described above, although the embodiments have been described with limited examples and drawings, those skilled in the art can make various modifications and variations from the above description. For example, the described techniques may be performed in an order different from the method described, and/or components of the described system, structure, device, circuit, etc. may be combined or combined in a different form than the method described, or other components may be used. Or even if it is replaced or substituted by equivalents, appropriate results can be achieved.

Therefore, other implementations, other embodiments, and equivalents of the claims are within the scope of the following claims.

Claims (8)

Prioritizing transcoding jobs performed in the central transcoding server in order of cost-effectiveness relative to Popularity-Weighted Video Quality (PWQ) in an edge computing environment through a collection server;
allocating a corresponding transcoding job to a central transcoding server or an edge server according to network coverage and calculation load for each transcoding job given the priority through the aggregation server;
reassigning or removing some of the work through the aggregation server to the central transcoding server operated by a live streaming provider in order to minimize the effect of network coverage and computational load on the popularity-weighted video quality; and
Providing live video streaming to a plurality of viewer terminals through a transmission server after performing the transcoding task allocated through the collection server in the central transcoding server or the edge server.
including,
The step of giving priority to the transcoding jobs performed in the central transcoding server in order of cost-effectiveness relative to popularity-weighted video quality in an edge computing environment through the collection server,
A list of channels exists in the collection server, and video streams of each channel are captured, encoded, and uploaded;
Defines a video stream of a version with a requested bit rate from multiple viewer terminals as a combination of binary values, calculates a popularity-weighted video quality, and determines the video quality transmitted to a plurality of user terminals within processing capacity constraints and network range constraints. Allocating transcoding tasks according to popularity-weighted video quality to maximize, and then removing the assigned tasks to minimize popularity-weighted video quality loss until cost budget constraints are met.
How to assign transcoding jobs. delete According to claim 1,
The step of allocating the transcoding job to the central transcoding server or the edge server according to the network range and computational load for each transcoding job given the priority through the collection server,
Using a binary variable indicating whether each channel is within network range of the transcoding server, giving a binary value depending on whether or not the transcoding operation was performed for the lowest bitrate version of that channel.
How to assign transcoding jobs. According to claim 3,
After performing the transcoding task allocated through the collection server in the central transcoding server or the edge server, providing live video streaming to a plurality of viewer terminals through a transmission server,
If the version with the bit rate requested from the plurality of viewer terminals is not transcoding, transmitting the version with the highest bit rate among the versions lower than the requested bit rate among the versions in which the transcoding task is executed
How to assign transcoding jobs. an aggregation server that prioritizes transcoding jobs performed on the central transcoding server in order of cost-effectiveness relative to Popularity-Weighted Video Quality (PWQ) in an edge computing environment;
A central transcoding server and an edge server to which the corresponding transcoding job is allocated according to network coverage and computational load for each transcoding job given the priority through the aggregation server - the central transcoding server is the popularity weighted video Some jobs are reallocated or some jobs are removed through the aggregation server to minimize the impact of network coverage and computational load on quality; and
A transmission server that provides live video streaming to a plurality of viewer terminals after performing the transcoding task allocated through the collection server in the central transcoding server or the edge server.
including,
The collection server,
A list of channels exists, the video stream of each channel is captured, encoded, and uploaded;
Defines a video stream of a version with a requested bit rate from multiple viewer terminals as a combination of binary values, calculates a popularity-weighted video quality, and determines the video quality transmitted to a plurality of user terminals within processing capacity constraints and network range constraints. Allocating transcoding tasks according to popularity-weighted video quality to maximize, and then removing the assigned tasks to minimize popularity-weighted video quality loss until cost budget constraints are met.
Transcoding job allocation system. delete According to claim 5,
The central transcoding server and the edge server,
With a binary variable indicating whether each channel is within network range of the transcoding server, a binary value is given depending on whether or not the transcoding operation is performed on the lowest bitrate version of that channel.
Transcoding job allocation system. According to claim 7,
The transmission server,
If the version with the bit rate requested from the plurality of viewer terminals is not transcoding, transmitting the version with the highest bit rate among the versions lower than the requested bit rate among the versions in which the transcoding task is executed
Transcoding job allocation system.

Images