Abstract
Todays cloud deployed applications are mostly multi-tiered. Usually, the first tier consists of an Application Service Providers' (ASPs) web servers, the second tier has application servers and the third tier contains database servers. Tiered architectures are often difficult to evaluate in terms of performance. Existing performance models are very effective in finding the mean performance measures. However, metrics such as response-time percentiles are of greater importance to the end-users since it is more desirable to reduce the variability of a system’s response time, rather than minimizing the mean response time. In this work, a multi-tier application is modeled as an open queuing network of 3-tiers and the response-time percentiles are estimated using discrete event simulation. Here, we assume that each tier is replicated into a number of copies and each copy runs on a separate Virtual Machine (VM). Although simulation models are computationally more expensive as compared to analytical models, they are much more general. The simulation model of this work can be used as decision support for ASPs in order to determine the optimal configuration of VMs for a given workload such that a required response-time percentile is within a given threshold. In this work, Simpy, a discrete event simulation framework, has been used. The results show that as the number of VMs are increased in a 3-tier open queueing network, the overall system performance (i.e. percentiles and mean response times) does not necessarily become better. The results further show that different system configurations containing the same number of VMs, yield different performance depending on the replication level in different tiers.
Highlights
Application service providers (ASPs), which provide web services to customers have to ensure that their services are always live, ensure minimum delay and adhere to Service Level Agreement (SLA) requirements
Most application service providers have a configuration consisting of 3 tiers where the first tier consists of web servers, the second tier consists of application servers and the third tier consists of DBMS servers [13]
The model in this paper allows application service providers to simulate their requirements and arrive at an ideal configuration or number of Virtual Machine (VM) they would need for their operations, which in turn they could request to be purchased from a cloud service provider
Summary
Application service providers (ASPs), which provide web services to customers have to ensure that their services are always live, ensure minimum delay and adhere to Service Level Agreement (SLA) requirements. Application service providers request virtual machines from cloud service providers, for deploying their web applications to satisfy SLA requirements for their own customers. Application service providers typically can have their resources modeled as a queuing network composed of a set of single-server queues. These single server queues can be thought of as a configuration of 2-3 tiers. Approximations are made for response time distributions An example of such an approximation technique is to divide the network into single server queues. Approximation techniques on obtaining response time distribution for queuing networks with failures of VMs in tiers are presented as in [16] and [12]
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have