Improving the responsiveness of internet services with automatic cache placement
Proceedings of the 4th ACM European conference on Computer systems
| Hi-index | 0.00 |
Internet service utilities host multiple server applications on a shared server cluster. A key challenge for these systems is to provision shared resources on demand to meet service quality targets at least cost. This dissertation presents a new approach to utility resource management focusing on coordinated provisioning of cluster resources, including CPU, memory and storage bandwidth. Our approach is model-based: it incorporates internal models of service behavior to predict the value of candidate resource allotments under changing load. The model-based approach enables the system to achieve important resource management goals, including differentiated service quality, performance isolation, improved content caching, and proactive allocation of surplus resources to meet performance goals. This dissertation makes significant steps in adaptive resource provisioning for web service utilities. (1) It presents a feedback-controlled framework for adaptive provisioning in an Internet data center. (2) It introduces models for predictive resource provisioning, including an algorithm and framework supporting models for proactive prediction of resource requirements. Using load estimates, our model-based approach estimates the effectiveness of the resource allotments. (3) It incorporates server energy, cache memory and storage bandwidth as primary target resources for the adaptive provisioning algorithm. (4) It presents an algorithm for determining cache hit rate based on memory size for static web services with Zipf-like object popularity distributions. It presents an algorithm to parameterize the model from on-line observations. (5) It introduces the notion of provisioning to meet customer demand using energy consumption as a component of this decision process. (6) It illustrates the effects of caching on the characteristics of request streams arriving at the enterprise edge. It also shows how caching affects request distribution policies. Combined, these contributions expand the understanding of memory as a provisioned resource, and advance the state of provisioning systems to incorporate pluggable resource models for flexible and accurate adaptive resource allocation.