Cloud Resource Management Methods Based on maximization of benefits and differential pricing

In February 11, I read a paper on cloud resource management: flexible use of cloud resources through profit maximization and price discirmination (icde 2011). Take notes:

This paper proposes a virtual machine allocation strategy based on marginal cost and marginal revenue. Each user sets a budget function, which specifies the hourly payment amount agreed by the user corresponding to the different response time (execution time) of the task. The task response time is related to the number of virtual machines assigned to the user. Increasing the number of virtual machines reduces the response time, increases the user's payment amount, and increases the benefits of cloud service providers. However, the task response time is not linearly related to the number of virtual machines. When the number of virtual machines is large, the newly added virtual machines will not bring great benefits. Each virtual machine has a certain cost. By combining the task response time with the number of virtual machines and the cost of each newly added virtual machine, you can decide how to reasonably allocate the task to the User Virtual Machine On The IAAs cloud to maximize the benefits of the cloud service provider. The details are as follows:

Take the number of VMS as the abscissa. With the increase in the number of virtual machines, the performance is getting smaller and smaller, that is, the response time is getting smaller and smaller, so the benefit is a curve with a decreasing slope and the corresponding marginal benefits (Mr) is a declining curve. In addition, we simplify the system overhead model. Assuming that the cost of each virtual machine is fixed, the cost is a straight line from the origin, and the marginal cost (MC) is a horizontal line. The abscissa corresponding to the intersection of marginal benefits and marginal costs is the number of virtual machines that maximize the benefits of cloud service providers.

The above methods maximize the benefits of cloud service providers in a single user environment. In a multi-user environment, virtual machines belonging to multiple users compete for resources on physical machines, so the performance of a single virtual machine will decrease. The performance of virtual machines with the same number of users will also decrease compared with the previous ones, therefore, the benefit curve is lower and smoother than before. The marginal benefit curve is also lower. The marginal cost curve remains unchanged, so the intersection of the two changes (left shift ).

In addition, when multiple users submit jobs at the same time, we should consider the total input of all users rather than just one user. In this way, even if the benefits of a few users are not very large, the benefits will always be maximized. The algorithm proposed in this article is based on the maximum amount that each user can pay. That is, the number of virtual machines allocated to each user is proportional to the maximum amount that a user can pay.