A GPU cloud server allocation method based on user needs

Abstract

The present invention discloses a GPU cloud server allocation method based on user demand, which relates to the technical field of GPU cloud servers. The steps include: a client receives a user's account login and sends the user's account information to a service management module; the user type to which the user belongs is determined according to the user's account information; the service management module traverses the resource configuration information of all GPU cloud servers connected to it; the service management module calculates the user's service quality index value K according to the user type and resource configuration information. n ; In all K n Among the values, select the maximum K n value, according to the maximum K n The value corresponds to the pricing method M, GPU model T, server location R, cloud service provider C and GPU cloud server hourly price P, and this is used as the optimal resource configuration information. By assigning various calculation factors, the present invention can allocate the optimal GPU cloud server to users according to different user types.

Description

Technical Field

[0001] The present invention relates to the technical field of GPU cloud servers, and in particular to a GPU cloud server allocation method based on user demand. Background Art

[0002] With the rise and development of cloud services at home and abroad, cloud services have become one of many production solutions today, and cloud service providers with Internet product thinking provide a variety of different pricing factors to influence the final price of the server. One of the most important links in cloud gaming cost control is the price inquiry strategy of GPU cloud servers. Since different user scenarios bring different product requirements, different price inquiry strategies are needed to control the cost of GPU cloud servers to achieve the best input-output ratio.

[0003] At present, cloud gaming is just getting started, and the infrastructure of GPU cloud servers has just been completed. In addition to cloud gaming developers and operators, GPU cloud servers are also purchased by a large number of companies engaged in AI-related businesses. Therefore, there will be periodic shortages of GPU cloud servers in the market, and the shortages of different models among different cloud service providers are different. Therefore, the traditional method of bulk purchasing of cloud servers of a single model in a single region can no longer meet the current market demand.

[0004] Since the user experience time of cloud gaming is mostly 1-2 hours, it is impossible to directly buy out a large number of GPU cloud servers for a long time. Therefore, real-time switching of GPU cloud servers has become a high-priority demand. This has also led to the problem of different startup costs due to the fragmented pricing models of cloud service providers.

[0005] Therefore, in order to solve the above problems, it is necessary to establish a new GPU cloud server pricing strategy that can not only meet the needs of users, but also meet the appropriate cost control and sufficient server margin, and also meet the end-user experience. Summary of the invention

[0006] In order to solve one of the above technical problems, the present invention provides a GPU cloud server allocation method based on user needs. By assigning various calculation factors, the optimal GPU cloud server can be allocated to users according to different types of users.

[0007] To solve the above technical problems, the present invention provides the following technical solutions: a method for allocating a GPU cloud server based on user needs, applied to a GPU cloud server system, wherein the GPU cloud server system comprises a client, a service management module and a GPU cloud server connected in sequence; the method for allocating a GPU cloud server based on user needs comprises the following steps:

[0008] Step S1: The client receives the user's account login and sends the user's account information to the service management module;

[0009] Step S2: The service management module receives the user's account information and determines the user type of the user according to the user's account information;

[0010] The user types include: cost-sensitive users, rigid-need users, enterprise users, large online gathering users, and single application users;

[0011] Step S3: The service management module traverses the resource configuration information of all GPU cloud servers connected to it, wherein the resource configuration information includes: GPU server surplus A, pricing method M, GPU model T, server location R, cloud service provider C, and GPU cloud server hourly price P;

[0012] The service management module calculates the user's service quality index value K according to the user type and resource configuration information. n ;

[0013] Step S4: In all K n Among the values, select the maximum K n value, according to the maximum K n The value corresponds to the determined pricing method M, GPU model T, server region R, cloud service provider C and GPU cloud server hourly price P, and this is used as the optimal resource configuration information;

[0014] Step S5: The service management module allocates the corresponding GPU cloud server to the user based on the optimal resource configuration information.

[0015] Furthermore, the service quality index value K of the user is calculated n , the calculation formula is:

[0016] K n =[(X1*A m )+(X2*M s )+(X3*T d )+(X4*R g )+(X5*C j )] / (X6*P n );

[0017] Among them, X1, X2, X3, X4, X5 and X6 are weight coefficients, which are assigned according to the user type; A m=A1, A2, A3. When the GPU cloud server system scans that the GPU server surplus that can be provided by a single cloud service provider C is less than or equal to 20, set A1=1 / 3. When the GPU server surplus scanned is greater than 20 and less than 60, set A2=2 / 3. When the GPU server surplus scanned is greater than or equal to 60, set A3=1.

[0018] M s =M1, M2, that is, the pricing method M includes preemptible instance M1 and pay-as-you-go M2. Assign values ​​to M1 and M2: M1=1, M2=2 / 3;

[0019] T d is one or more of T1, T2, and T3, that is, the GPU model T includes one or more of the T4 model T1, the A10 model T2, and the V100 model T3, T1=1, T2=2 / 3, and T3=1 / 3;

[0020] R g is one or more of R1, R2, and R3, that is, the region R where the server is located includes one or more of the first region R1, the second region R2, and the third region R3, and R1, R2, and R3 are assigned values, R1=1, R2=2 / 3, and R3=1 / 3;

[0021] C j is one or more of C1, C2, and C3, that is, the cloud service provider C includes one or more of the first cloud service provider C1, the second cloud service provider C2, and the third cloud service provider C3, and C1, C2, and C3 are assigned values, C1=1, C2=2 / 3, and C3=1 / 3;

[0022] In K n and P n In, K n The number of is determined by the pricing method M, GPU model T, server region R, and cloud service provider C. That is, the value of n is the number of pricing methods M * the number of GPU models T * the number of server regions R * the number of cloud service providers C. After the GPU cloud server system determines a single pricing method M, a single GPU model T, a single server region R, and a single cloud service provider C, the GPU cloud server system can scan and obtain the corresponding GPU server surplus A. According to the GPU server surplus A, A can be obtained. m In addition, after a single pricing method M, a single GPU model T, a single server location R, and a single cloud service provider C are determined, the hourly price P of the GPU cloud server provided by the single cloud service provider C is also uniquely determined;

[0023] The service management module calculates all K according to all the pricing methods M, GPU models T, server regions R and cloud service providers C. n value.

[0024] Furthermore, when calculating all K n Before setting the value, assign values ​​to X1, X2, X3, X4, X5 and X6 according to the user type, as follows:

[0025] When a user is classified as a cost-sensitive user, X1 is set to 10, X2 is set to 10, X3 is set to 10, X4 is set to 10, X5 is set to 10, and X6 is set to 50, that is, K n =[(10*A m )+(10*M s )+(10*T d )+(10*R g )+(10*C j )] / (50*P n );

[0026] When the user is classified as a rigid demand user, X1 is set to 10, X2 is set to 10, X3 is set to 10, X4 is set to 50, X5 is set to 10, and X6 is set to 10, that is, K n =[(10*A m )+(10*M s )+(10*T d )+(50*R g )+(10*C j )] / (10*P n );

[0027] When a user is classified as an enterprise user, X1 is set to 10, X2 is set to 10, X3 is set to 20, X4 is set to 30, X5 is set to 10, and X6 is set to 10, that is, K n =[(10*A m )+(10*M s )+(20*T d )+(30*R g )+(10*C j )] / (10*P n );

[0028] When the user is classified as a large online gathering user, X1 is set to 40, X2 is set to 10, X3 is set to 15, X4 is set to 15, X5 is set to 10, and X6 is set to 10, that is, K n =[(40*A m )+(10*M s )+(15*T d )+(15*R g )+(10*Cj )] / (10*P n );

[0029] When the user is classified as a single application user, X1 is set to 10, X2 is set to 10, X3 is set to 30, X4 is set to 10, X5 is set to 10, and X6 is set to 30, that is, K n =[(10*A m )+(10*M s )+(30*T d )+(10*R g )+(10*C j )] / (30*P n ).

[0030] After adopting the above technical solution, the present invention has at least the following beneficial effects: the present invention performs K n The calculation of K n The corresponding resource configuration information is the optimal GPU cloud server required by the user; the present invention calculates the optimal resource configuration information according to different demand factors, which solves the existing problem that different regions, different GPU models, different cloud service providers and different pricing methods lead to a large service quality gap under the same cost in different user scenarios; this embodiment does not simply pursue the lowest cost, but obtains the GPU cloud server configuration list with the highest service quality within a given budget according to different user scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 This is a flowchart of the steps of a GPU cloud server allocation method based on user needs in the present invention. DETAILED DESCRIPTION

[0032] It should be noted that, in the absence of conflict, the embodiments in the present application and the features described in the embodiments may be combined with each other. The present application is further described in detail below in conjunction with the drawings and specific embodiments.

[0033] like Figure 1 As shown, the present invention discloses a GPU cloud server allocation method based on user needs, which is applied to a GPU cloud server system. The GPU cloud server system includes a client, a service management module and a GPU cloud server connected in sequence; the GPU cloud server allocation method based on user needs includes the following steps:

[0034] Step S1: The client receives the user's account login and sends the user's account information to the service management module;

[0035] Step S2: The service management module receives the user's account information and determines the user type of the user according to the user's account information;

[0036] The user types include: cost-sensitive users, rigid-need users, enterprise users, large online gathering users, and single application users;

[0037] Step S3: The service management module traverses the resource configuration information of all GPU cloud servers connected to it, wherein the resource configuration information includes: GPU server surplus A, pricing method M, GPU model T, server location R, cloud service provider C, and GPU cloud server hourly price P;

[0038] The service management module calculates the user's service quality index value K according to the user type and resource configuration information. n , the calculation formula is:

[0039] K n =[(X1*A m )+(X2*M s )+(X3*T d )+(X4*R g )+(X5*C j )] / (X6*P n ); where X1, X2, X3, X4, X5 and X6 are weight coefficients, and these weights need to be set according to the user type; A m Including A1, A2, A3, ..., A m In this embodiment, m=3 is preferred, that is, A m =A1, A2, A3. When the GPU cloud server system scans that the GPU server surplus provided by a single cloud service provider C is less than or equal to 20 (the surplus is insufficient), A1 is set to 1 / 3. When the GPU server surplus is scanned to be greater than 20 and less than 60 (the surplus is normal), A2 is set to 2 / 3. When the GPU server surplus is scanned to be greater than or equal to 60 (the surplus is sufficient), A3 is set to 1. If the GPU server surplus is scanned to be 0, A=0, and there is no need to calculate the K value. This situation will basically not occur.

[0040] In this embodiment, it is preferred to set s=1, 2, that is, M s =M1, M2, that is, the pricing method M includes preemptible instance M1 and pay-as-you-go M2. Assign values ​​to M1 and M2: M1=1, M2=2 / 3;

[0041] Example T dOne or more of T1, T2, and T3, that is, the GPU model T includes one or more of T4 model T1, A10 model T2, and V100 model T3. T1, T2, and T3 are assigned values, T1=1, T2=2 / 3, and T3=1 / 3. If the cloud server system does not find the GPU model T configured in the cloud server system under the region R where one of the servers of a cloud service provider C is located, T=0, and there is no need to calculate the K value. This situation will basically not occur.

[0042] This embodiment R g is one or more of R1, R2, and R3, that is, the region R where the server is located includes one or more of the first region R1, the second region R2, and the third region R3, and R1, R2, and R3 are assigned values, R1=1, R2=2 / 3, and R3=1 / 3;

[0043] This embodiment C j is one or more of C1, C2, and C3, that is, the cloud service provider C includes one or more of the first cloud service provider C1, the second cloud service provider C2, and the third cloud service provider C3, and C1, C2, and C3 are assigned values, C1=1, C2=2 / 3, and C3=1 / 3;

[0044] In K n and P n In, K n The number of is determined by the pricing method M, GPU model T, server region R, and cloud service provider C. That is, the value of n is the number of pricing methods M * the number of GPU models T * the number of server regions R * the number of cloud service providers C. After the GPU cloud server system determines a single pricing method M, a single GPU model T, a single server region R, and a single cloud service provider C, the GPU cloud server system can scan and obtain the corresponding GPU server surplus A. According to the GPU server surplus A, A can be obtained. m In addition, after a single pricing method M, a single GPU model T, a single server location R, and a single cloud service provider C are determined, the hourly price P of the GPU cloud server provided by the single cloud service provider C is also uniquely determined;

[0045] The service management module calculates all K according to all the pricing methods M, GPU models T, server regions R and cloud service providers C. n value;

[0046] In calculating all K n Before setting the value, assign values ​​to X1, X2, X3, X4, X5 and X6 according to the user type, as follows:

[0047] When a user is classified as a cost-sensitive user, X1 is set to 10, X2 is set to 10, X3 is set to 10, X4 is set to 10, X5 is set to 10, and X6 is set to 50, that is, K n =[(10*A m )+(10*M s )+(10*T d )+(10*R g )+(10*C j )] / (50*P n );

[0048] When the user is classified as a rigid demand user, X1 is set to 10, X2 is set to 10, X3 is set to 10, X4 is set to 50, X5 is set to 10, and X6 is set to 10, that is, K n =[(10*A m )+(10*M s )+(10*T d )+(50*R g )+(10*C j )] / (10*P n );

[0049] When a user is classified as an enterprise user, X1 is set to 10, X2 is set to 10, X3 is set to 20, X4 is set to 30, X5 is set to 10, and X6 is set to 10, that is, K n =[(10*A m )+(10*M s )+(20*T d )+(30*R g )+(10*C j )] / (10*P n );

[0050] When the user is classified as a large online gathering user, X1 is set to 40, X2 is set to 10, X3 is set to 15, X4 is set to 15, X5 is set to 10, and X6 is set to 10, that is, K n =[(40*A m )+(10*M s )+(15*T d )+(15*R g )+(10*C j )] / (10*P n );

[0051] When the user is classified as a single application user, X1 is set to 10, X2 is set to 10, X3 is set to 30, X4 is set to 10, X5 is set to 10, and X6 is set to 30, that is, K n =[(10*A m )+(10*M s )+(30*Td )+(10*R g )+(10*C j )] / (30*P n );

[0052] Step S4: In all K n Among the values, select the maximum K n value, according to the maximum K n The value corresponds to the determined pricing method M, GPU model T, server region R, cloud service provider C and GPU cloud server hourly price P, and this is used as the optimal resource configuration information;

[0053] For example, if the user type is determined to be an enterprise user, then K n =[(10*A m )+(10*M s )+(20*T d )+(30*R g )+(10*C j )] / (10*P n ), A m =A1, A2, A3, M s =M1, M2 (preemptible instance M1 and pay-as-you-go instance M2), T d =T1, T2 (only two models are set, T4 model T1 and A10 model T2), R g =R1, R2 (in this example, the server region R is selected as the first region R1 and the second region R2), C j =C1, C2 (only two cloud service providers C are set, including the first cloud service provider C1 and the second cloud service provider C2, namely Alibaba Cloud C1 and AWS Cloud C2);

[0054] Assign values ​​at the same time: A1=1 / 3, A2=2 / 3, A3=1, M1=1, M2=2 / 3, T1=1, T2=2 / 3, R1=1, R2=2 / 3, C1=1, C2=2 / 3; n=s*d*g*j=2*2*2*2=16, that is, K n =K1, K2, K3, …, K 14 , K 15 , K 16 , P n =P1, P2, P3, …, P 14 , P 15 , P 16 ;

[0055] K1=[(10*A m)+(10*M1)+(20*T1)+(30*R1)+(10*C1)] / (10*P1). After the service management module determines the four conditions of preemptible instance M1, T4 model T1, first region R1, and Alibaba Cloud C1, it queries the corresponding GPU server surplus and GPU cloud server hourly price P1. At this time, the service management module scans that the GPU server surplus is 32 (normal), so it assigns A m =A2=2 / 3, the service management module finds out that the hourly price of the GPU cloud server is P1=3.25; therefore K1=[(10*2 / 3)+(10*1)+(20*1)+(30*1)+(10*1)] / (10*3.25)≈2.3590;

[0056] K2=[(10*A m )+(10*M2)+(20*T1)+(30*R1)+(10*C1)] / (10*P2). After the service management module determines the four conditions of pay-as-you-go M2, T4 model T1, first region R1, and Alibaba Cloud C1, it queries the corresponding GPU server surplus and GPU cloud server hourly price P2. At this time, the service management module scans that the GPU server surplus is 74 (sufficient surplus), so it assigns A m =A3=1, the service management module finds out that the hourly price of the GPU cloud server is P2=7.92; therefore K2=[(10*1)+(10*2 / 3)+(20*1)+(30*1)+(10*1)] / (10*7.92)≈0.9680;

[0057] K3=[(10*A m )+(10*M1)+(20*T2)+(30*R1)+(10*C1)] / (10*P3), the service management module checks the four conditions of preemptible instance M1, A10 model T2, first region R1, and Alibaba Cloud C1, and queries the corresponding GPU server surplus and GPU cloud server hourly price P3. At this time, the service management module scans that the GPU server surplus is 27 (sufficient surplus), so it assigns A m =A2=2 / 3, the service management module finds out that the hourly price of the GPU cloud server is P3=5.80; therefore K3=[(10*2 / 3)+(10*1)+(20*2 / 3)+(30*1)+(10*1)] / (10*5.80)≈1.2069;

[0058] By analogy, K4=[(10*A m)+(10*M1)+(20*T1)+(30*R2)+(10*C1)] / (10*P4)=[(10*2 / 3)+(10*1)+(20*1)+(30*2 / 3)+(10*1)] / (10*3.15)≈2.1164;

[0059] K5=[(10*A m )+(10*M1)+(20*T1)+(30*R1)+(10*C2)] / (10*P5)=[(10*1)+(10*1)+(20*1)+(30*1)+(10*2 / 3)] / (10*8.78)≈0.8732;

[0060] K6=[(10*A m )+(10*M1)+(20*T1)+(30*R1)+(10*C2)] / (10*P6)=[(10*1)+(10*2 / 3)+(20*2 / 3)+(30*1)+(10*1)] / (10*14.2)≈0.4930;

[0061] K7=[(10*A m )+(10*M2)+(20*T1)+(30*R2)+(10*C1)] / (10*P7)=[(10*1)+(10*2 / 3)+(20*1)+(30*2 / 3)+(10*1)] / (10*8.92)≈0.7474;

[0062] K8=[(10*A m )+(10*M2)+(20*T1)+(30*R1)+(10*C2)] / (10*P8)=[(10*1)+(10*2 / 3)+(20*1)+(30*1)+(10*2 / 3)] / (10*12.16)≈0.6031;

[0063] K9=[(10*A m )+(10*M1)+(20*T2)+(30*R1)+(10*C2)] / (10*P9)=[(10*1)+(10*1)+(20*2 / 3)+(30*1)+(10*2 / 3)] / (10*27.6)≈0.2415;

[0064] K 10 =[(10*A m )+(10*M1)+(20*T2)+(30*R1)+(10*C2)] / (10*P 10)=[(10*2 / 3)+(10*1)+(20*2 / 3)+(30*2 / 3)+(10*1)] / (10*9.6)=0.625;

[0065] K 11 =[(10*A m )+(10*M1)+(20*T2)+(30*R1)+(10*C2)] / (10*P 11 )=[(10*1)+(10*1)+(20*1)+(30*2 / 3)+(10*2 / 3)] / (10*11.62)≈0.5737;

[0066] K 12 =[(10*A m )+(10*M1)+(20*T2)+(30*R1)+(10*C2)] / (10*P 12 )=[(10*1)+(10*1)+(20*2 / 3)+(30*2 / 3)+(10*2 / 3)] / (10*26.4)≈0.2272;

[0067] K 13 =[(10*A m )+(10*M1)+(20*T2)+(30*R1)+(10*C2)] / (10*P 13 )=[(10*1)+(10*2 / 3)+(20*1)+(30*2 / 3)+(10*2 / 3)] / (10*13.16)≈0.4813;

[0068] K 14 =[(10*A m )+(10*M1)+(20*T2)+(30*R1)+(10*C2)] / (10*P 14 )=[(10*1)+(10*2 / 3)+(20*2 / 3)+(30*1)+(10*2 / 3)] / (10*32.6)≈0.2045;

[0069] K 15 =[(10*A m )+(10*M1)+(20*T2)+(30*R1)+(10*C2)] / (10*P 15 )=[(10*1)+(10*2 / 3)+(20*2 / 3)+(30*2 / 3)+(10*1)] / (10*15.6)≈0.3846;

[0070] K 16 =[(10*A m)+(10*M1)+(20*T2)+(30*R1)+(10*C2)] / (10*P 16 ) = [(10*1) + (10*2 / 3) + (20*2 / 3) + (30*2 / 3) + (10*2 / 3)] / (10*36.7) ≈ 0.1544;

[0071] Obviously, the above K1≈2.3590 is the largest, that is, the optimal GPU cloud server resource configuration information allocated to the enterprise user is: preemptible instance M1, T4 model T1, first region R1 and Alibaba Cloud C1;

[0072] Step S5: The service management module allocates the corresponding GPU cloud server to the user based on the optimal resource configuration information.

[0073] This embodiment provides K for different user types. n The calculation of K n The corresponding resource configuration information is the optimal GPU cloud server required by the user; this embodiment calculates the optimal resource configuration information according to different demand factors, which solves the existing problem that different regions, different GPU models, different cloud service providers and different pricing methods lead to a large gap in service quality at the same cost in different user scenarios; this embodiment does not simply pursue the lowest cost, but obtains the GPU cloud server configuration list with the highest service quality within a given budget according to different user scenarios.

[0074] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various equivalent changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A GPU cloud server allocation method based on user demand, applied to a GPU cloud server system, wherein the GPU cloud server system comprises a client, a service management module and a GPU cloud server connected in sequence, characterized in that: The following steps are involved: Step S1: The client receives the user's account login and sends the user's account information to the service management module; Step S2: The service management module receives the user's account information and determines the user type of the user according to the user's account information; The user types include: cost-sensitive users, rigid-need users, enterprise users, large online gathering users, and single application users; Step S3: The service management module traverses the resource configuration information of all GPU cloud servers connected to it, wherein the resource configuration information includes: GPU server surplus A, pricing method M, GPU model T, server location R, cloud service provider C, and GPU cloud server hourly price P; The service management module calculates the user's service quality index value K according to the user type and resource configuration information. n ; Step S4: In all K n Among the values, select the maximum K n value, according to the maximum K n The value corresponds to the determined pricing method M, GPU model T, server location R, cloud service provider C and GPU cloud server hourly price P, and this is used as the optimal resource configuration information; Step S5: The service management module allocates the corresponding GPU cloud server to the user according to the optimal resource configuration information; The service quality index value K of the user is calculated n , the calculation formula is: K n =[(X1*A m )+(X2*M s )+(X3*T d )+(X4*R g )+(X5*C j )] / (X6*P n ); Among them, X1, X2, X3, X4, X5 and X6 are weight coefficients, which are assigned according to the user type; A m =A1, A2, A3. When the GPU cloud server system scans that the GPU server surplus that can be provided by a single cloud service provider C is less than or equal to 20, set A1=1 / 3. When the GPU server surplus scanned is greater than 20 and less than 60, set A2=2 / 3. When the GPU server surplus scanned is greater than or equal to 60, set A3=1. M s =M1, M2, that is, the pricing method M includes preemptible instance M1 and pay-as-you-go M2. Assign values ​​to M1 and M2: M1=1, M2=2 / 3; T d is one or more of T1, T2, and T3, that is, the GPU model T includes one or more of the T4 model T1, the A10 model T2, and the V100 model T3, T1=1, T2=2 / 3, and T3=1 / 3; R g is one or more of R1, R2, and R3, that is, the region R where the server is located includes one or more of the first region R1, the second region R2, and the third region R3, and R1, R2, and R3 are assigned values, R1=1, R2=2 / 3, and R3=1 / 3; C j is one or more of C1, C2, and C3, that is, the cloud service provider C includes one or more of the first cloud service provider C1, the second cloud service provider C2, and the third cloud service provider C3, and C1, C2, and C3 are assigned values, C1=1, C2=2 / 3, and C3=1 / 3; In K n and P n In, K n The number of is determined by the pricing method M, GPU model T, server region R, and cloud service provider C. That is, the value of n is the number of pricing methods M * the number of GPU models T * the number of server regions R * the number of cloud service providers C. After the GPU cloud server system determines a single pricing method M, a single GPU model T, a single server region R, and a single cloud service provider C, the GPU cloud server system can scan and obtain the corresponding GPU server surplus A. According to the GPU server surplus A, A can be obtained. m In addition, after a single pricing method M, a single GPU model T, a single server location R, and a single cloud service provider C are determined, the hourly price P of a GPU cloud server provided by a single cloud service provider C is n It is also the only one that is certain; The service management module calculates all K according to all the pricing methods M, GPU models T, server regions R and cloud service providers C. n value.

2. The method for allocating GPU cloud servers based on user demand according to claim 1, characterized in that: In calculating all K n Before setting the value, assign values ​​to X1, X2, X3, X4, X5 and X6 according to the user type, as follows: When a user is classified as a cost-sensitive user, X1 is set to 10, X2 is set to 10, X3 is set to 10, X4 is set to 10, X5 is set to 10, and X6 is set to 50, that is, K n =[(10*A m )+(10*M s )+(10*T d )+(10*R g )+(10*C j )] / (50*P n ); When the user is classified as a rigid demand user, X1 is set to 10, X2 is set to 10, X3 is set to 10, X4 is set to 50, X5 is set to 10, and X6 is set to 10, that is, K n =[(10*A m )+(10*M s )+(10*T d )+(50*R g )+(10*C j )] / (10*P n ); When a user is classified as an enterprise user, X1 is set to 10, X2 is set to 10, X3 is set to 20, X4 is set to 30, X5 is set to 10, and X6 is set to 10, that is, K n =[(10*A m )+(10*M s )+(20*T d )+(30*R g )+(10*C j )] / (10*P n ); When the user is classified as a large online gathering user, X1 is set to 40, X2 is set to 10, X3 is set to 15, X4 is set to 15, X5 is set to 10, and X6 is set to 10, that is, K n =[(40*A m )+(10*M s )+(15*T d )+(15*R g )+(10*C j )] / (10*P n ); When the user is classified as a single application user, X1 is set to 10, X2 is set to 10, X3 is set to 30, X4 is set to 10, X5 is set to 10, and X6 is set to 30, that is, K n =[(10*A m )+(10*M s )+(30*T d )+(10*R g )+(10*C j )] / (30*P n ).

Images