Skip to main content

Advances, Systems and Applications

Cost-based hierarchy genetic algorithm for service scheduling in robot cloud platform

Abstract

Service robot cloud platform is effective method to improve the intelligence of robots. An efficient cloud service scheduling algorithm is the basis of ensuring service quality and platform concurrency. In this paper, Hierarchy Genetic Algorithm of robot service(RHGA) is presented to solve the scheduling problem with the crucial constraints. Firstly, the limitations and attributes of the cloud service robots and cloud services are presented and boiled down to an important optimization goal. Secondly, three factors (i.e. evolutionary factor, hunting factor and parent similarity) are integrated with RHGA to promote the efficiency of small-scale service invocations and improve the performance of large-scale service invocations on the platform. Finally, a series of experiments are conducted on several service scheduling algorithms, including four traditional efficient algorithms and two state-of-art algorithms. The experimental results demonstrate that the RHGA can enhance the performance on small-scale service scheduling and ensure its excellent ability in large-scale service scheduling. Moreover, the empirical studies also prove that our proposal has a better performance in service scheduling completion time and cost-savings with comparison to other methods.

Introduction

Cloud computing technology has changed the application development mode of traditional industries. Previously, the industrial environment and terminal can only rely on the local area network or terminal computing power for intelligent computing. Cloud computing platform provides various services such as Infrastructure as a Service (IaaS), Platform as a Service (PaaS) and Software as a Service (SaaS) [1] for terminals or systems through the network. Robot, as a typical intelligent terminal, has long been limited by its own computing power, which affects the real-time calculation of dense data.

Subsequently, an increasing number of scholars are addicted to the combination of robotics and cloud computing. In 2010, Kuffner [2] firstly proposed the robot cloud platform conceptually to offload the modules of complex data processing, artificial intelligence and model training. The robot only serves as the terminal for information collection and task execution.

The robot cloud service platform changes the development mode of the robot, transfers the robot intelligent algorithm that were previously deployed on the robot to cloud platform. Through WiFi or Bluetooth, robot can remotely use the AI cloud service to improve their intelligence, such as face recognition, scene recognition, voice recognition, voice interaction and other cloud services in Cloud service pool(CSP) through protocols. However, the heterogeneity of robots (different sensor parameters, different robot structures and different protocols) perplexes the robot cloud platform. The cloud services are not fully compatible with different types of robots, which is an important reason blocking the development of robot cloud platform. Therefore, it is very important to explore a general robot cloud service platform to satisfy the needs of different robots.

However, the robot cloud service robot platform is designed for plenty of robots rather than a few. Robot cloud platform will publish a certain number of AI cloud services. When cloud platform publishes a large number of robot cloud services, the system needs to prepare data storage, message bus, file access and permission control related to cloud services, which will consume a large amount of computing resources. On the contrary, if the cloud service is published too little, the platform cannot satisfy the requirements of the robot for cloud services. Therefore, reasonable scheduling of cloud services to maximize resource utilization is very meaningful.

Through the analysis of cloud service scheduling workflow, cloud services are deployed on virtual machines with different configurations. For different properties of IO and CPUs, cloud services have different information processing capabilities, such as the indicators of real-time and availability. In other words, the cost of robots calling cloud services is not consistent. This study considered the problem of how to schedule the cloud service to minimize the cloud service response time, that is, to improve the response speed of the platform to the instructions called by the robot.

There are many methods to schedule cloud services. Common non-heuristic cloud service scheduling algorithms such as First Come First Serve (FCFS), Min-Min, Max-Min, Round Robin, Weighted Round Robin, RASA, Segmented Min-Min algorithm are static algorithms following a predefined approach of scheduling the cloudlets in the computing environment as stated by Chaudhary et al. These algorithms have poor parameter dynamic performance and high computational resource consumption. The swarm intelligence-based cloud service scheduling is adaptive, intelligent, collective, random, decentralized, self-collective, stochastic and is based on biologically inspired mechanisms than the other conventional mechanisms. Commonly used such as genetic algorithm, particle swarm algorithm, ant colony algorithm, artificial bee colony algorithm, etc. Most of these heuristic algorithms are random search methods that simulate natural biological evolution or group social behavior. Since algorithms usually do not rely on gradient information, they have been proved to generate near optimal solutions for task scheduling problem. However, their scheduling overhead increases vastly as the number of tasks or number of resources increases, and the optimization rate of the solution will gradually decrease. Multiplicity of solutions also slows down the convergence rate of the genetic algorithm.

Genetic algorithm (GA) is a heuristic search algorithm that simulates natural selection and heredity, has better robustness. Compared with other swarm intelligence algorithms, the operation of gene selection, crossover and mutation makes the genetic algorithm have stronger ability to search the optimal solution. Meanwhile, this strong search ability also reduces the convergence speed of the algorithm. In this paper, an improved genetic algorithm is proposed to guarantee the diversity of solutions and improve the convergence speed of the algorithm. The main contributions of this paper are as follows.

(i). The framework of cloud service platform of service robot is proposed, which can meet the service call of heterogeneous service robot. By analyzing the workflow of service call of robot, a cost-based cloud service scheduling model is established.

(ii). Multi-layer population hunting strategy is proposed. By hunting inferior individuals at different levels of the population and using the optimal individuals to supplement the population, the quality of the population increases, which accelerate the convergence of the algorithm.

(iii). Dynamic updating method is introduced to improve the traversal range of the algorithm to the spatial solution set. The probability of individual variation is adjusted according to the fitness value in the variation stage.

(iv). Parental similarity-based method is proposed to determine gene similarity. For parents with high similarity, multi-point crossover can enhance individual diversity, which can accelerate the search for a global solution.

The remainder of this paper is organized as follows. Section 2 introduces the Related Work about cloud service robots and cloud-based scheduling algorithms. In Section 3, we consider the characteristics of service robots and service cloud platform to form a final cost model with constraints. Section 3 proposes the proposed RHGA with three novel factors, i.e. evolutionary factor, hunting factor, and parent similarity. Section 4 reports the experimental analysis of RHGA and verifies the effectiveness of our proposal on service scheduling completion time and cost-savings, followed by concluding remarks and introducing future work in Section 5 briefly.

Related Work

Cost and intelligence are the key problems in the development of service robots. Since Kuffner proposed the combination of cloud computing and robot in 2011, the robot cloud service platform has gradually received extensive attention from scholars at home and abroad. A typical robotic platform is the European “RoboEarth” project, which aims to build a system of shared knowledge among robots. After that, many cloud robotic frameworks such as DAvinCi [3], Robot-Cloud [4], C2TAM [5], Rapyuta [6], RCC [7] and IAPcloud [8] have been proposed. DAvinCi is based on Robot Operate System(ROS), distributed system architecture of hadoop and the Cloud computing model of Map/Reduce, provide the proxy service for robot with binding the server cluster of Hadoop and ROS. Robot-Cloud is designed for the integration and collaborative control of heterogeneous robots. This framework is implemented with components including the cloud controller, ROS master, storage unit, Map-reduce cluster and robot cluster, in which each robot has installed ROS and the robotic service is published through the master node in the cloud controller. C2TAM is a visual SLAM system based on a distributed framework. Map optimization and storage are performed in the cloud, and the camera’s tracking algorithm is run on the local computer. Rapyuta is an open source cloud robotics framework. Each robot connected to the RoboEarth Cloud Engine will have a system level clone on the cloud giving them the ability to move their heavy computation into the cloud. In order to solve the resource sharing problem, Zhihui Du et al. built the robot cloud computing center(RCC), which provided the information sharing platform to improve the multi-robot task management coordination capacity. Song presented the new control platform of IAPcloud has significant advantages in solving the problem of collaborative control among heterogeneous robots and their auxiliary devices, declining the programming difficulties of cloud robotic control systems and shortening the development and deployment cycle of the application. The above cloud robot architecture can unload the complex computing and storage of the robot ontology to the cloud, which can extend the performance of the robot ontology beyond the physical hardware. However, the compatibility of heterogeneous service robots and communication protocols and platforms still needs further research.

The robot cloud platform provides diversified AI cloud services for robots, improves the intelligence of robots and reduces the cost of robots. Considering the cost of service invocation by robots caused by the quality difference of platform services, many scholars have studied the service scheduling of robots. Service scheduling problems in the robot cloud belongs to a category known as NP-hard problems. The two most common kinds of algorithms which aim to produce near optimal solutions within the polynomial time for such problem are used frequently: exhaustive algorithms and intelligent evolution optimization algorithms [9]. Traditional typical scheduling methods, such as Min-Min algorithm, Max-Min algorithm, First Come First Service (FCFS) algorithm, Round Robin (RR) algorithm, can realize cloud service scheduling [10]. However, these algorithms are based on certain rules or expert experience, unable to adapt to the dynamic scheduling of complex robot cloud service system. Intelligent evolutionary algorithm is always a popular algorithm in service scheduling research, e.g. particle swarm optimization (PSO) [11], genetic algorithm (GA) [12], artificial bee colony (ABC) [13] etc. Moreover, Due to the imitation of natural evolution, genetic algorithm has stronger search ability and robustness than other heuristic algorithms [10, 14, 15]. The slow convergence speed and easy to fall into local optimum also trouble the genetic algorithm. At present, a typical idea is algorithm fusion, in which other machine learning algorithms with fast convergence speed are combined with traditional GA to improve the convergence speed. Ahmad [16] proposed a Hybrid GA-PSO algorithm, which reduced the total number of iterations of the task by searching the solution process by stages. However, the diversity of solutions did not improve greatly, and the total cost remained unchanged. Mohan [17] presented a novel hybrid meta-heuristic scheme harmony-inspired genetic algorithm (HIGA) to improve population quality and accelerate algorithm convergence. Divya [18] proposes Hybrid Genetic-Gravitational Search Algorithm (HG-GSA), and uses the gravitational search algorithm to optimize the population generated by the genetic algorithm and improve the speed of the algorithm to force the optimal solution. Compared with traditional GA, fusion algorithm reduces the overall number of iterations. However, the algorithm complexity within each iteration increases, and the overall cost range changes little.

Some scholars introduce specific strategies in the process of crossover and mutation of traditional genetic algorithms to improve the solving rate of the optimal solution. Considering the scheduling characteristics of cloud services, Xiong [19] incorporated Johnson’s rule into the traditional GA algorithm and designed a new mutation operation to improve the convergence speed of the algorithm. Amir [20] proposed a hybrid genetic algorithm based on cost and time constraints to schedule cloud service. Through the adaptive crossover of single, double and multiple points of the parent population, the quality of the population was improved and the convergence of the algorithm was accelerated. Henrique [21] proposed a hybrid genetic algorithm based on cost and time constraints to implement workload flow scheduling, and improved the quality of individual population through the adaptive crossover of single, double and multiple points of parent population.

In order to address the limitations of existing scheduling solutions, we present the RHGA, a scheduler based on a Genetic Algorithm. In summary, (a) RHGA proposes a hierarchical structure in the hunting stage to improve the proportion of the high-quality population and speeds up the convergence of the algorithm. (b) Adaptive mutation rule is proposed to improve the ability of traversing the solution set. (c) Multi-point crossover fusion strategy based on the similarity between subpopulations and parents is proposed to speed up the search efficiency of the optimal solution.

In addition, our laboratory (Cloud Intelligent Robotics Laboratory) has devoted to the cloud service robots to study the scheduling algorithms [22,23,24] in recent years. As far as we know, these cloud applications indicate that cloud robots could expand their knowledge and skills through the cloud and accomplish the work much more efficiently. However, most of the existing works focus on the architecture of the platform rather than the cloud service scheduling algorithm, which makes it worth exploring a more efficient scheduling algorithm. A short introduction to the cloud robot platform can be found in the url (https://www.bilibili.com/video/BV1P4411P7yP or https://youtu.be/RhYhykr6AM0).

System Model and Problem formulation

Cloud platform model and assumption

In this paper, the robot cloud platform between robots and platform is based on the former framework that we have presented in [25], which composes of Gateway Layer, Interface Layer, Service pool and Algorithm Layer. The brief architecture of the proposed CIRCP (Cloud Robotics Intelligent Cloud Platform) is shown in Fig. 1. Service robots can utilize any internet communication tools to apply for services according to the JSON format, e.g. Bluetooth, WIFI, and 5G. A simple service scheduling process is as follows. Firstly, service robots send service requests through Interface Layer. Next, Gateway Layer performs identity authentication and with the help of scheduler to invoke service from Service Pool which is based on the Algorithm Layer. Finally, service robots complete the corresponding function via the called service. There are still one main issues in both service robots and cloud platform need to be considered, that is, the final cost of service robots and the cloud platform.

Fig. 1
figure 1

The main architecture of robot service cloud platform

As for various service robots, the purpose of their design is also very different, which means that the service processing ability and self-state can be significantly different [4]. A brief service invocation finite-state machine is displayed in Fig. 2. On the one hand, the communication cost between service robots and cloud platform are different, which is defined as remote-communication cost. On the other hand, the cost varies with the hardware of service robots which is called intra-communication cost. In general, this part of the cost is counted as the expenditure of service preparation cost, including S2 state.

Fig. 2
figure 2

Service invocation finite-state machine

In terms of the robot cloud platform, service scheduler invokes services from Service Pool. A brief workflow of service invocation on the platform is presented in Fig. 3. Three critical factors need to be considered. (1) Different cloud services are deployed on different servers. (2) The service processing time is determined by many factors, e.g. IO, CPU etc. In short, resources utilization is considered an important issue, including HardDisk and CPU. (3) TSeveralhe cost of the specific service invocation sequence can be various. Here, we pay more attention to the essential issues, and ignore the communication cost within the cluster due to the infinite internal bandwidth. Thus, the cost of that scheduler transfers the service to the actuator is negligible.

Fig. 3
figure 3

A brief workflow of service invocation

Problem formulation

Based on the cloud platform model, several constraints exist in the service robot cloud platforms, such as non-uniform interface standards, the heterogeneity of robots, high time cost of communicating with the platform. Nevertheless, the most critical factor in service scheduling is the communication cost [10, 15, 26]. In this paper, both side cost, including the service robots and the cloud platform in the cloud, turns into one main objective optimization problem. On the one hand, the final cost is a fundamental optimization goal of the two parties. Furthermore, cost of both also has the same nature toward final cost, and is simplified into one goal to improve processing efficiency. On the other hand, the adjusted proportional coefficient according to the weight makes it easier to deal with practical problems. These indicators express as the following three definitions.

Definition 1 (Robot cost): As for robot communication, a robot could be seen as an end-user device with two essential problems that need to be considered. One is the intra-communication \(K_{n}\) that denotes the communication time of i-th robot with i-th service on the platform. The final of cost of robot communication \(C_r\) shows in Eq. (1)

$$\begin{aligned} C_{r}=\sum _{t=1}^{M}\left( \sum _{i=1}^{N} c_{i} r_{i i}+\sum _{i=1}^{N} e_{i} k_{t i}\right) \end{aligned}$$
(1)

Where \(c_i\) and \(e_i\) are the cost of the corresponding service invocation for remote and intra communication. \(r_{ti}\) is calculated by Eq. (2). \(k_{ti}\) is expressed in Eq. (3).

$$\begin{aligned} r_{t i}=\frac{L_{i}}{R P_{t}} \end{aligned}$$
(2)
$$\begin{aligned} k_{t i}=\frac{L_{\ {data_i }}+L_{r e q_{-} i}}{B_{t}} \end{aligned}$$
(3)

Where \(RP_{t}\) represents the processing power of different robots. \(L_{i}\) is the length of service, which is also treated as the main QoS constraint on the robot side. \(L_{data_{-}i}\) and \(L_{req_{-}i}\) denote the length of service transferred in the robot. \(B_{t}\) is the identifier for bandwidth of robots. Moreover, \(r_{ti}\) is the cost of the i-th service \(S_{i}\) called by the t-th robot \(R_{t}\). The cost matrix of robot calling cloud service is shown in Eq. (4).

$$\begin{aligned} CostMap={ \left[ \begin{array}{cccc} r_{11} &{} r_{12} &{} \cdots &{} r_{1i}\\ r_{21} &{} r_{22} &{} \cdots &{} r_{2i}\\ \vdots &{} \vdots &{} \ddots &{} \vdots \\ r_{t1} &{} r_{t2} &{} \cdots &{} r_{ti}\\ \end{array} \right] } \end{aligned}$$
(4)

In addition, as the robot cloud service must be registered in the cloud at first, the CostMap could be calculated in advance for various robots and saved in their storage. Thus, the entire computing process could be accelerated.

Definition 2(Platform cost): Different cloud services are deployed on different servers, occupies resources of specific services influence the various cost of specific service invocation. Moreover, different cost also depends on different service invoking sequences. Finishing time is a critical factor of the cloud service. the total cost \(P_{r}\) of the specific service invocation sequence for all the robots can be formulated in Eq. (5).

$$\begin{aligned} P_{r}=\sum \limits _{i=1}^N{{c_i}^{\prime }{S_{pi}}+}\sum \limits _{i = 1}^N {k_i^{\prime }{S_{mi}}} \end{aligned}$$
(5)
$$\begin{aligned} s.t.\quad r \in R \end{aligned}$$
(6)

Where \(c_{i}^{\prime }\) is the cost of service invocation on the specific server. \(K_{i}\) is the cost of a specific service finishing time. \(S_{pi}\) is the occupied resources of specific service invocation on a specific server. \(S_{mi}\) is a specific service finishing time. As for the service invocation sequence, which is shown in Eq. (7), it can have different length and concurrency.

$$\begin{aligned} \textrm{T}S_{n} = (< {S_{p1}},{S_{m1}}> ,< {S_{p1}},{S_{m1}}> , \cdots , < {S_{pi}},{S_{mi}} > ) \end{aligned}$$
(7)

In addition, \(S_{i}\) is composed of \(S_{mi}\) and \(S_{pi}\) on the platform.

Definition 3 (Total cost): When a large scale of robots invoke the services, the robot communication time and service finishing time has the same effect on cost. That is, effective scheduling algorithm can shorten the time of communication and execution. In order to make efficient and timeless scheduling, we reform the above formula as one objective optimization, which is shown in Eq. (8). The main goal is to find the best scheduling strategy to minimize the total cost and reduce the total scheduling time.

$$\begin{aligned} Obj:\textrm{Minimize}(TC = \alpha C_{r} + \beta P_{r}) \quad {\alpha ,\beta \in [0,1]} \end{aligned}$$
(8)

s.t. \(\alpha + \beta = 1\); \(Rstate \in S4, S5\)

Where TC is the total cost of the service invocation process. \(\alpha \) and \(\beta \) are the coefficient of robot cost and platform cost. The robot state (Rstate) should at least meet S4 or S5. Moreover, the server must have the ability to handle several services at the same time.

Hierarchy Genetic Algorithm of robot service

GA is a search heuristic that mimics the process of natural selection by implementing some of its basic mechanisms, such as reproduction, crossover, and mutation. In this paper, evolutionary factors, hunting factors and parental similarity are introduced to improve the convergence speed of the algorithm.

Main framework of algorithm

The evolutionary algorithm has been widely used for solving the NP-Hard problem within a reasonable time. Hence, in this work, we propose an improved evolutionary algorithm RHGA based on the genetic algorithm. RHGA is inspired by nature’s hunting laws, and different efficient and effective solutions could be found at the same time. The basic structure of RHGA is shown in algorithm 1, in which the num of services is considered as k. Thus, the problem turns into a search in a k-dimensional search space. Moreover, all possible sets must be mapped in the search space.

In algorithm 1, main ideas are as follows.

(a) Sequence Representation. A chromosome is used to deal with a service request sequences. Those who have low scores are better individuals than the high.

(b) Initialization. A population for one service scheduling sequence is generated randomly in lines \(1-2\) and will be evaluated using Eq. (8) in line 3. Thus, the early best individual can be saved.

(c) Three cultural factors: next step. In lines \(5-8\), cultural effects are applied via three important factors: evolutionary factor (\(\delta \)), hunting factor (hkf), and parent similarity (ps). Thus, we can obtain the optimal solution within the shortest or reasonable number of iterations. Finally, the best service scheduling strategy will be applied to the cloud platform. Thus, the total time complexity of the algorithm is \(O(MG*({P_{rank}} + {H_{select}})*(P/2) + P*({F_{last}} - {F_{current}}))\). Where MG represents the number of iterations, \({F_{last}}\) and \({F_{current}}\) denote the last and current population in algorithm 2, respectively, \({P_{rank}}\) is the ranked population size, and \({H_{select}}\) is the steps of selection process in algorithm 3, P is the identifier for the population size.

figure a

Algorithm 1 Main Architecture of the RHGA (SE, Obj, MG, SC);

It is worth noting that three cultural factors play an important role in sequence selection. Thus, a detailed explanation will be given in the next sub-sections.

Evolutionary factor

To keep the diversity of the population, the population evolutionary factor is introduced in algorithm 2. With the slightly adaptive changing rate of the mutation, more superior individuals could be generated. As the generated probability (rand) is lower than selective probability (rsp), the last and the current chromosome are chosen to generate the offspring. The mutation rate will be changed under the condition that the initial threshold (th) is greater than \(\delta \). In the real world service scheduling, both GA and PSO have a poor performance when the number of services invocation is small. Algorithm 2 could alleviate or even eliminate the problem, and a better result than them is shown in Section 5. Thus, we could derive that the complexity of this part is \(O({F_{last}} - {F_{current}})\).

figure b

Algorithm 2 Population evolutionary factor

Hunting factor

Inspired by the predator hunting strategy in nature, we propose a three-level hierarchy selection strategy in algorithm 3. The survival order of population is reordered according to their fitness in lines 1-2, firstly. Then, line 3 divides them into three levels according to a certain ration. Finally, in lines 4-8, different hunting factors (hkf) are set for each level. In addition, the inspiration could be described by the biological phenomena in the nature. When predators hunt their prey, in the prey population, individuals whose fitness is low are hunted by predators much more easily. However, in some real circumstances, those who have high fitness and a better survivor technique may be hunted in the course of their being chased to some extent, while the probability of this phenomenon is much lower than the formers (those who have a low chance of survival). Thus, for each level, the individuals whose survival probability (sp) is lower than the hunting of factor will be killed while others survive. Finally, the best individual replication strategy is applied for filling the rest of the population after all the operations.

In terms of this strategy, two main time-consuming issues need to be considered, that is, the sorting process and the selection process. Therefore, the time complexity of this part is up to \(O({P_{rank}} + {H_{select}})\).

figure c

Algorithm 3 Predator hunting strategy

Parents similarity

A novel crossover operation to generate new offspring is proposed in algorithm 4. In order to make the population become more convergence and superiority, the similarity of parents is used to perform crossover selection. In lines 1-6, when the selection probability (dey), based on a Gaussian probability model, is lower than the odds of crossover selection (os) and the cosine distance between parent p1 and parent p2 is lower than the value of similarity (ss), each parent will be chosen to give its own genetic randomly to generate only one child. As it runs twice, two children will be generated and compared with the previous parents to find the best individuals. In other conditions, children will be generated by using a single point of intersection crossover. The algorithm could guarantee to produce more good individuals and improve the efficiency of service scheduling. Moreover, the complexity of this part is O(P/2) .

figure d

Algorithm 4 Update the population

Experiment

In this section, three comparative experiments are conducted to verify the validity and performance of our proposed RHGA with comparison to some existing efficient algorithm strategies, Max-Min algorithm, Min-Min algorithm, First Come First Service (FCFS), Round Robin (RR), Dynamic Non-Linear Particle Swarm Optimization (DNPSO) [27], and Genetic Algorithm based Task Scheduling (GATS) [28].The empirical results show that RHGA can have a better performance on service scheduling with respect to the cost of robot communication time. In addition, all units of time are in milliseconds.

Experimental environment

The proposed RHGA and existing algorithms (DNPSO, GATS, Max-Min, etc.) are implemented in CloudSim [29]. The CloudSim is a universal cloud computing simulation platform that can simulate a real cloud. Moreover, the experimental environment is shown in Table 1, where there is mainly a single desktop computer and a set of IDEA development tools.

Table 1 Experimental environment

All the specific algorithm parametric values are summarized in Table 2, where the threshold of the population evolutionary factor is set to 0.3, three hunting factors from top to bottom are 0.1, 0.3, and 0.6, and the threshold of similarity of parents is set to 0.6. The number of services ranges from 100 to 10000. A thousand iterations of each algorithm will be implemented as well as run 30 times.

Table 2 The specific algorithm parameters

The cloud platform simulation parameters and corresponding values are shown in Table 3. Each host has 1000M/s network bandwidth. The MIPS rating is 1000 for each core of hosts with Linux operating system. The length of a service ranges from 1000ms to 2000ms.

Table 3 The cloud platform simulation parameters

Contrastive experiment

Experiment on number of services

In this experiment scenario, we set the 10 datacenters as a cluster on the cloud platform, 25 populations, and a different number of services. Each algorithm runs 30 times and obtains the results, including the maximum value, minimum value, and the mean value, which is shown in Table 4 (see the attached data file). Although DNPSO and GATS have a better performance than others on the average scheduling time, except for RHGA. However, when the number of services is small, GATS and DNPSO scheduling performance are not always better than the non-evolutionary algorithm. Even in some cases, e.g. the number of services is 100, they have a poor performance in the minimum value or the maximum value of total scheduling time. The reason may be that the results of these evolutionary algorithms are stochastic to some extent. Thus, GATS and DNPSO can’t always have superior advantages.

Table 4 The results of experiments on number of services

As per the results in Table 4, our proposed RHGA has a better performance than other algorithms in all three aspects. The maximum value of scheduling time is lower than others’ minimum value of scheduling time, e.g. when the number of services is 6000, the maximum value of scheduling time of RHGA is 4.34E+05ms, while the minimum of DNPSO is 4.48E+05ms, and GATS is 4.43E+05ms.

To make the empirical studies straightforward, we report the scheduling time chart in Fig. 4 (small-scale). With respect to the problem of large-scale access, a numerical difference function shows in Eq. (9) to achieve a more comprehensive presentation, and the result presents in Fig. 5 (large-scale).

$$\begin{aligned} \Delta = S{T_{fuc}} - S{T_{RHGA}} \end{aligned}$$
(9)

Where \(\delta \) is the numerical difference, indicates the scheduling time of the specific algorithm, i.e. RR, Max-Min and DNPSO, etc. \(ST_{RHGA}\) is regarded as the baseline for other algorithms.

Fig. 4
figure 4

The scheduling time (small-scale)

Figure 4 shows the scheduling time in small-scale service invocations. RHGA could overcome the problem of high time cost when the platform schedules a small number of services. Besides, RHGA is more robust than other algorithms for the cloud platform and robot to obtain a more stable service sequence. This is mainly because the evolutionary factor could ensure that the population has enough optimal solutions when a small number of service invocation accesses. In addition, it can be seen from Fig. 5, the time gap between RHGA and other algorithms is becoming larger as the number of services increases up to 9000 services. Due to the limited computing resources that set in the experiment, when the services continue to increase, the time gap is smaller than before. While RHGA is algorithm is still more time-saving than other algorithms. Thus, it can be inferred that the diversity and accuracy of the solution set will be improved in large-scale access by using the hunting factor and parent similarity. According to the above analysis, RHGA can enhance the performance on small-scale service scheduling and ensure its excellent ability in large-scale service scheduling. Additionally, RHGA is robust than the compared algorithms and easy to obtain a more stable service sequence.

Fig. 5
figure 5

The numerical difference of scheduling time (large-scale)

Experiment on number of Datacenters

To further evaluate the impact of cluster size in the cloud service scheduling problem and verify that whether RHGA could have higher performance on service scheduling when suitable cluster size is set, we set the size of the population to 25, the number of datacenters (i.e. the cluster size) to 10, 30, 50 and 100. The number of services is 100, 1000, 10000, and used as a static condition here. Each algorithm runs 30 times and takes the average, and the results report in Fig. 6.

According to the following empirical figures, as the number of datacenters increases, the average scheduling time of all the above algorithm decreases due to the sufficient computing resources. However, DNPSO and GATS may have lousy performance as the number of datacenters increases. As can be seen in Fig. 6(b)(c)(d), when the number of services is 100, DNPSO has the worst performance, while GATS also has not much of advantages in the average time. The reason for this problem may exist in two aspects. On the one hand, the randomness of these meta-heuristic algorithms and the lack of a better mechanism to save outstanding individuals that could lead to the loss of excellent solutions. On the other hand, when the size of clusters becomes larger, the compared non-evolutionary algorithms could have enough computing power and obtain the best results. After that, as the number of services increases, DNPSO and GATS gradually gain their advantages in searching solutions. Moreover, referring to Fig. 6, it is observed that RHGA has considerably performance above the rest empirical algorithms in the experimental process. These observations prove that the three critical factors integrated with RHGA significantly promote the efficiency of robot cloud service scheduling.

Fig. 6
figure 6

The average scheduling time of various clusters

The relation between the number of services and the average scheduling time of RHGA is plotted in Fig. 7. As can be seen in the figure, when the number of services goes up, the average scheduling time as well as the gap of scheduling time between the different sizes of clusters increase, and the average time of makespan reached a maximum in1000 services between 30 clusters and 10 clusters. The above results signal that scheduling time can have a notable improvement due to the computing power improvement. However, the tremendous cost, analyzed in the next experiment, of building a cloud platform with many computing resources, should be considered. Also, the observations in Fig. 7 suggest that the best trade-off size for clusters is 30.

Fig. 7
figure 7

The average scheduling time of RHGA on services

Experiment on cost

To investigate the total final cost of service scheduling, we set the population size to 25, the number of datacenters to 10, 30, 50 and 100, and the number of services to 100, 1000, 5000 and 10000. Here, the average cost on a service scheduling is measured in dollars. Each algorithm runs 30 times to obtain the average results that display in Fig. 8. As can be seen from these figures, the average cost goes up when the number of datacenters increases. Also, it could be observed that RHGA remains cost-savings than other algorithms with the increasing datacenters. In contrast, as the number of datacenters becomes larger, a considerable amount of energy and an enormous cost for cloud providers will be needed, which leads to the rise of the average cost. However, the experimental results confirm that RHGA could ensure service scheduling efficiency as well as minimize the cost-savings. Additionally, RHGA can guarantee performance in the whole process via three factors.

Fig. 8
figure 8

The average cost of various clusters

Figure 9 shows the relation between cluster number and the average cost of RHGA. As shown in the figure, the average cost of RHGA increases when the number of clusters becomes larger with a fixed size of scheduling services. Moreover, the observations indicate that a vast number of services could make a high cost on the cloud. The tremendous cost may be mainly due to the increasing cloud virtual machines and high cost of cluster maintenance. Thus, it appears that RHGA could keep a better balance between the number of clusters and the efficiency of service scheduling.

Fig. 9
figure 9

The average cost of RHGA on clusters

Conclusion

This work presents an improved service scheduling algorithm of RHGA to minimize the final cost of service scheduling as well as reduce the scheduling time in the service robot cloud platform. The main contributions of this research are as follow. By considering the characteristics of service robots and cloud service instances, all the constraints and characteristics are boiled down to a single-objective optimization problem to alleviate the urgent deficient in communication cost of two sides. Then, three factors, i.e. evolutionary factor (\(\delta \)), hunting factor (hkf), and parent similarity (ps), are integrated with RHGA to promote the performance on small-scale service scheduling and ensure its excellent ability in large-scale service scheduling. Specifically, the results in the first experiment suggest that RHGA is robust than the compared six algorithms, i.e. FCFS, Max-Min, Min-Min, Round-Robin, (DNPSO) PSO, and (GATS) GA. Thus, it is more likely to be applied in more complex service scheduling to obtain a more stable service sequence. At last, the empirical results confirm that RHGA could ensure service scheduling efficiency as well as minimize the cost-savings. Additionally, it could make a trade-off between the number of clusters and the efficiency of service scheduling.

In the future, the RHGA will applied for more applications in real-world scenarios. Besides, more constraints and problems will be considered to optimize and solve in our research, e.g. the balance between the number of clusters and service scheduling efficiency.

Availability of data and materials

The data used to support the findings of this study are available from the corresponding author upon request.

Abbreviations

IaaS:

Infrastructure as a Service

PaaS:

Platform as a Service

SaaS:

Software as a Service

RHGA:

Hierarchy Genetic Algorithm of robot serivice

References

  1. Zhu X, Chen C, Yang LT, Xiang Y (2015) Angel: Agent-based scheduling for real-time tasks in virtualized clouds. IEEE Trans Comput 64(12):3389–3403

    Article  MathSciNet  MATH  Google Scholar 

  2. Kuffner J (2010) Cloud-enabled humanoid robots. In: 2010 10th IEEE-RAS International Conference. IEEE, p 16–27

  3. Arumugam R, Enti VR, Bingbing L, Xiaojun W, Baskaran K, Kong FF, Kumar AS, Meng KD, Kit GW (2010) Davinci: A cloud computing framework for service robots. In: 2010 IEEE international conference on robotics and automation. IEEE, pp 3084–3089

  4. Doriya R, Chakraborty P, Nandi G (2012) Robot-cloud: A framework to assist heterogeneous low cost robots. In: 2012 International Conference on Communication, Information & Computing Technology (ICCICT). IEEE, pp 1–5

  5. Riazuelo L, Civera J, Montiel JM (2014) C2tam: A cloud framework for cooperative tracking and mapping. Robot Auton Syst 62(4):401–413

    Article  Google Scholar 

  6. Hunziker D, Gajamohan M, Waibel M, D’Andrea R (2013) Rapyuta: The roboearth cloud engine. In: 2013 IEEE international conference on robotics and automation. IEEE, pp 438–444

  7. Du Z, Yang W, Chen Y, Sun X, Wang X, Xu C (2011) Design of a robot cloud center. In: 2011 Tenth International Symposium on Autonomous Decentralized Systems. IEEE, pp 269–275

  8. Zheng S, Lin Z, Zeng Q, Zheng R, Liu C, Xiong H (2018) Iapcloud: A cloud control platform for heterogeneous robots. IEEE Access 6:30577–30591

    Article  Google Scholar 

  9. Abd Elaziz M, Xiong S, Jayasena KPN, Li L (2019) Task scheduling in cloud computing based on hybrid moth search algorithm and differential evolution. Knowl-Based Syst 169(APR.1):39–52

  10. Mavrovouniotis M, Li C, Yang S (2017) A survey of swarm intelligence for dynamic optimization: Algorithms and applications. Swarm Evol Comput 33:1–17

    Article  Google Scholar 

  11. Kennedy J (2011) Particle swarm optimization. In: Proc of 1995 IEEE Int Conf Neural Networks, (Perth, Australia), IEEE, 4(8). pp 1942–1948

  12. Goldberg DE (1989) Genetic algorithm in search optimization and machine learning. Addison Wesley xiii(7):2104–2116

  13. Karaboga D, Basturk B (2007) A powerful and efficient algorithm for numerical function optimization: artificial bee colony (abc) algorithm. J Glob Optim 39(3):459–471

    Article  MathSciNet  MATH  Google Scholar 

  14. Balusamy B, Sridhar J, Dhamodaran D, Krishna PV (2015) Bio-inspired algorithms for cloud computing: A review. Int J Innov Comput Appl 6(3–4):181–202

    Article  Google Scholar 

  15. Ragmani A, El Omri A, Abghour N, Moussaid K, Rida, M, (2017) An intelligent scheduling algorithm for energy efficiency in cloud environment based on artificial bee colony. In: 2017 3rd International Conference of Cloud Computing Technologies and Applications (CloudTech), IEEE, p 1–8

  16. Manasrah AM, Ba Ali H (2018) Workflow scheduling using hybrid ga-pso algorithm in cloud computing. Wirel Commun Mob Comput 2018, 1(1):1–16

  17. Sharma M, Garg R (2020) Higa: Harmony-inspired genetic algorithm for rack-aware energy-efficient task scheduling in cloud data centers. Eng Sci Technol Int J 23(1):211–224

    Google Scholar 

  18. Chaudhary D, Kumar B (2019) Cost optimized hybrid genetic-gravitational search algorithm for load scheduling in cloud computing. Appl Soft Comput 83(1):1-16

    Google Scholar 

  19. Xiong Y, Huang S, Wu M, She J, Jiang K (2017) A johnson’s-rule-based genetic algorithm for two-stage-task scheduling problem in data-centers of cloud computing. IEEE Trans Cloud Comput 7(3):597–610

    Article  Google Scholar 

  20. Iranmanesh A, Naji HR (2021) Dchg-ts: a deadline-constrained and cost-effective hybrid genetic algorithm for scientific workflow scheduling in cloud computing. Clust Comput 24(2):667–681

    Article  Google Scholar 

  21. Shishido HY, Estrella JC, Toledo CFM, Arantes MS (2018) Genetic-based algorithms applied to a workflow scheduling algorithm with security and deadline constraints in clouds. Comput Electr Eng 69:378–394

    Article  Google Scholar 

  22. Fengyu Z, Lei Y, Rui S, Tian T, Hongxing C (2017) A novel building and scheduling method of cloud platform services for robot. Robot 39(1):89–98

    Google Scholar 

  23. Chen H, Zhou FY, Tian T (2015) Design of soa interface model in service robot cloud computing platform. J Shandong Univ 45(4), 31–39

  24. Liu J, Zhou F, Yin L (2019a) Design of a service robot cloud service platform. In: 2019 4th Asia-Pacific Conference on Intelligent Robot Systems (ACIRS), IEEE, p 124-128

  25. Liu J, Zhou F, Yin L, Wang Y (2019) A novel cloud platform for service robots. IEEE Access 7:182951–182961

    Article  Google Scholar 

  26. Gamal M, Rizk R, Mahdi H, Elnaghi BE (2019) Osmotic bio-inspired load balancing algorithm in cloud computing. IEEE Access 7:42735–42744

    Article  Google Scholar 

  27. Chang J, Hu Z, Tao Y, Zhou Z (2018) Task scheduling based on dynamic non-linear pso in cloud environment. In: 2018 IEEE 9th International Conference on Software Engineering and Service Science (ICSESS), IEEE, p 877-880. https://www.researchgate.net/publication/331846461_Task_Scheduling_Based_on_Dynamic_Non-Linear_PSO_in_Cloud_Environment

  28. Agarwal M, Srivastava GMS (2016) A genetic algorithm inspired task scheduling in cloud computing. In: IEEE International Conference on Computing, Communication and Automation (ICCCA),IEEE, p 364–367. https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=7813746

  29. Alshammari D, Singer J, Storer T (2017) Does cloudsim accurately model micro datacenters? In: 2017 IEEE 10th International Conference on Cloud Computing (CLOUD), IEEE, p 705–709

Download references

Acknowledgements

We want to thank the authors of the literature cited in this paper for contributing useful ideas to this study. The authors would like to thank the anonymous reviewers for their helpful insights and suggestions which have substantially improved the content and presentation of this paper.

Funding

This work is supported by The National Key R & D Program of China (Grant No. 2017YFB1302400), National Natural Science Foundation of China (Grant No.61773242, No.61803227), Major Agricultural Applied Technological Innovation Projects of Shandong Province (SD2019NJ014).Intelligent Robot and System Innovation Center Foundation(2019IRS19). In addition, the authors thank the anonymous reviewers for providing valuable comments to improve this paper.

Author information

Authors and Affiliations

Authors

Contributions

L. Yin contributed on the design of RHGA. F. Zhou presented the performance evaluation. J.liu designed the experiment. M. Gao and M. Li participated in the design and optimization of framework. All authors have read and approved the final manuscript.

Corresponding author

Correspondence to Fengyu Zhou.

Ethics declarations

Competing interests

The authors declare that they have no competing interests.

Additional information

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Yin, L., Liu, J., Zhou, F. et al. Cost-based hierarchy genetic algorithm for service scheduling in robot cloud platform. J Cloud Comp 12, 35 (2023). https://doi.org/10.1186/s13677-023-00395-w

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doi.org/10.1186/s13677-023-00395-w

Keywords