## A smart hill-climbing algorithm for application server configuration (2004)

### Cached

### Download Links

- [www.stat.purdue.edu]
- [www.www2004.org]
- [www2004.org]
- [www.ra.ethz.ch]
- [www.iw3c2.org]
- DBLP

### Other Repositories/Bibliography

Venue: | 13th Int. Conf. on WWW |

Citations: | 27 - 0 self |

### BibTeX

@INPROCEEDINGS{Xi04asmart,

author = {Bowei Xi and Cathy H. Xia and Zhen Liu and Li Zhang and Mukund Raghavachari},

title = {A smart hill-climbing algorithm for application server configuration},

booktitle = {13th Int. Conf. on WWW},

year = {2004},

pages = {287--296}

}

### Years of Citing Articles

### OpenURL

### Abstract

The overwhelming success of the Web as a mechanism for facilitating information retrieval and for conducting business transactions has led to an increase in the deployment of complex enterprise applications. These applications typically run on Web Application Servers, which assume the burden of managing many tasks, such as concurrency, memory management, database access, etc., required by these applications. The performance of an Application Server depends heavily on appropriate configuration. Configuration is a difficult and error-prone task due to the large number of configuration parameters and complex interactions between them. We formulate the problem of finding an optimal configuration for a given application as a black-box optimization problem. We propose a Smart Hill-Climbing algorithm using ideas of importance sampling and Latin Hypercube Sampling (LHS). The algorithm is efficient in both searching and random sampling. It consists of estimating a local function, and then, hill-climbing in the steepest descent direction. The algorithm also learns from past searches and restarts in a smart and selective fashion using the idea of importance sampling. We have carried out extensive experiments with an online brokerage application running in a WebSphere environment. Empirical results demonstrate that our algorithm is more efficient than and superior to traditional heuristic methods. Categories and Subject Descriptors

### Citations

8169 |
Genetic Algorithms
- Goldberg
- 1989
(Show Context)
Citation Context ... are in the above spirit, attempting to find near-optimal or best-possible solutions instead of global optima. These include simulated annealing [14], random recursive search [18], genetic algorithms =-=[6]-=-, Tabu search [5], and hillclimbing [15, 1]. The first three are generally applicable as they require little aprioriknowledge of the problem. When the objective function has an explicit form, Hill-cli... |

4143 |
Artificial Intelligence: A Modern Approach, Pearson Education, 2nd edition
- Russell, Norvig
- 1995
(Show Context)
Citation Context ... find near-optimal or best-possible solutions instead of global optima. These include simulated annealing [14], random recursive search [18], genetic algorithms [6], Tabu search [5], and hillclimbing =-=[15, 1]-=-. The first three are generally applicable as they require little aprioriknowledge of the problem. When the objective function has an explicit form, Hill-climbing could quickly reach an optimal point ... |

4019 |
Stochastic Relaxation, Gibbs distributions and the Bayesian restoration of images
- Geman, Geman
- 1984
(Show Context)
Citation Context ...reached. The way in which the temperature is decreased is known as the cooling schedule. When the cooling schedule is controlled appropriately, the algorithm is guaranteed to achieve a global optimum =-=[4]. -=-Despite its many successful applications, using simulated annealing efficiently is a bit of an art — convergence can be slow. Recursive random search [18] utilizes pure random sampling. The algorith... |

3884 | Optimization by simulated annealing
- Kirkpatrick, Gelatt, et al.
- 1983
(Show Context)
Citation Context ...uristic commonly used to solve global optimization problems, especially in the presence of many false minima. It was motivated by the annealing process for a material to reach the thermal equilibrium =-=[7]-=-. A simulated annealing optimization starts with a Metropolis Monte Carlo simulation at a high temperature. This means that a relatively large percentage of the random steps that result in an increase... |

538 |
A comparison of three methods for selecting values of input variables in the analysis of output from a computer code
- McKay, Conover, et al.
- 1979
(Show Context)
Citation Context ... of efficiency”. Since the dimension of the problem we are dealing with is usually high, naive sample methods can become very expensive. We therefore rely on the Latin Hypercube Sampling (LHS) schem=-=e [10]-=-. LHS is considered to be an extremely efficient space-filling sampling strategy for handling high dimensions. It is considered more powerful than pure random Monte Carlo sampling. The basic idea of L... |

200 |
The Parallel Genetic Algorithm as Function Optimizer
- Muhlenbein, Schomisch, et al.
- 1991
(Show Context)
Citation Context ...o better evaluate and understand the performance of different algorithms, we assume the black-box function has an explicit form. We use one of the standard benchmark functions, the Rastrigin function =-=[13], which is defi-=-ned as: N� f(x) =N · β + (x 2 i − β · cos(2πxi)). (4.1) i=1 Figure 4 provides a 3D view of the Rastrigin function with N =2 and β =0.8 on the range [−1, 1] × [−1, 1]. Note that this ben... |

87 | Using MIMO Feedback Control to Enforce Policies for Interrelated Metrics With Application to the Apache Web Server
- Diao, Gandhi, et al.
- 2002
(Show Context)
Citation Context ...an approach that combines queueing theory and control theory for response time regulation, the approach can be used to handle a limited and small number of parameters. Using feedback control systems, =-=[2, 9]-=- studied the problem of regulating system performance within specified QoS value. The approach works well for a small number of tuning parameters with some linear dependency assumptions. We formulate ... |

87 |
Uncertainty and Sensitivity Analysis Techniques for Use in Performance Assessment for Radioactive-Waste Disposal," Reliability Engineering and System Safety
- Helton
- 1993
(Show Context)
Citation Context ...provides a better coverage of the parameter space and allows a significant reduction in the sample size to achieve a given level of confidence without compromising the overall quality of the analysis =-=[3]-=-. Given the above advantages of LHS, we use LHS in our Smart Hill-Climbing algorithm whenever there is a need to do random sampling. In the following sections, we shall show that by using LHS instead ... |

61 | Business-oriented Resource Management Policies for E-Commerce Servers,” Performance Evaluation
- Menascé, Almeida, et al.
- 2000
(Show Context)
Citation Context ... runs. There has been little previous work on the optimal configuration for Web application servers. In the context of HTTP servers, which have less complex interactions than Web Application Servers, =-=[11]-=- describes an Apache implementation that manages web server resources based on maximizing revenue. This approach requires substantial modifications to the Apache resource management schemes. [8] uses ... |

57 | Learning evaluation functions to improve optimization by local
- Boyan, Moore
- 2001
(Show Context)
Citation Context ... find near-optimal or best-possible solutions instead of global optima. These include simulated annealing [14], random recursive search [18], genetic algorithms [6], Tabu search [5], and hillclimbing =-=[15, 1]-=-. The first three are generally applicable as they require little aprioriknowledge of the problem. When the objective function has an explicit form, Hill-climbing could quickly reach an optimal point ... |

56 |
Tabu search, in: Modern heuristic techniques for combinatorial problems
- Glover, Laguna
- 1993
(Show Context)
Citation Context ... spirit, attempting to find near-optimal or best-possible solutions instead of global optima. These include simulated annealing [14], random recursive search [18], genetic algorithms [6], Tabu search =-=[5]-=-, and hillclimbing [15, 1]. The first three are generally applicable as they require little aprioriknowledge of the problem. When the objective function has an explicit form, Hill-climbing could quick... |

47 | Queueing model based network server performance control
- Sha, Liu, et al.
- 2002
(Show Context)
Citation Context ...s. [8] uses layered-queueing modeling to model business process applications. The method basically requires thorough knowledge about the software architecture and can be expensive and time-consuming. =-=[16]-=- describes an approach that combines queueing theory and control theory for response time regulation, the approach can be used to handle a limited and small number of parameters. Using feedback contro... |

33 | A Recursive Random Search Algorithm for Large-scale Network Parameter Configuration
- Ye, Kalyanaraman
- 2003
(Show Context)
Citation Context ...uristic search algorithms are in the above spirit, attempting to find near-optimal or best-possible solutions instead of global optima. These include simulated annealing [14], random recursive search =-=[18]-=-, genetic algorithms [6], Tabu search [5], and hillclimbing [15, 1]. The first three are generally applicable as they require little aprioriknowledge of the problem. When the objective function has an... |

22 | Online response time optimization of apache web server
- Liu, Sha, et al.
- 2003
(Show Context)
Citation Context ...an approach that combines queueing theory and control theory for response time regulation, the approach can be used to handle a limited and small number of parameters. Using feedback control systems, =-=[2, 9]-=- studied the problem of regulating system performance within specified QoS value. The approach works well for a small number of tuning parameters with some linear dependency assumptions. We formulate ... |

17 |
The deployer’s problem: Configuring application servers for performance and reliability
- Raghavachari, Reimer, et al.
- 2003
(Show Context)
Citation Context ...oper must configure an application server so that it can manage, for example, the concurrency of an enterprise application appropriately. In general, configuration is a difficult and error-prone task =-=[12] d-=-ue to the large number of configuration parameters and complex interactions between them — an application server may have more than a hundred parameters that can be modified. Examples of the paramet... |

13 | for Large-scale Network Parameter Configuration - Ye, ARecursiveRandomSearchAlgorithm - 2003 |

8 | Modeling and
- Petland, Liu
- 1999
(Show Context)
Citation Context ...ers, [11] describes an Apache implementation that manages web server resources based on maximizing revenue. This approach requires substantial modifications to the Apache resource management schemes. =-=[8]-=- uses layered-queueing modeling to model business process applications. The method basically requires thorough knowledge about the software architecture and can be expensive and time-consuming. [16] d... |

5 |
Simulated Annealing and Adaptive
- Romeijn, Smith
- 1994
(Show Context)
Citation Context ...number of experiments. Many heuristic search algorithms are in the above spirit, attempting to find near-optimal or best-possible solutions instead of global optima. These include simulated annealing =-=[14]-=-, random recursive search [18], genetic algorithms [6], Tabu search [5], and hillclimbing [15, 1]. The first three are generally applicable as they require little aprioriknowledge of the problem. When... |

3 | Improving Hit and Run for Global Optimization - Zabinsky, Smith, et al. - 1993 |