Next:
Configuring multiserver systems with
Up:
Analysis of Multi-server Systems
Previous:
Future directions
Contents
Performance analysis of multiserver systems
Subsections
Configuring multiserver systems with multiple priority classes
Introduction
State of the art in the optimal number of servers
How many servers are best in FCFS system?
How many servers are best in priority system?
Equal mean sizes
High priority class has smaller mean
High priority class has larger mean
Response time as a function of the number of servers
Summary
New approximations for many priority classes
Motivation
Approximations for multiserver systems with multiple priority classes
The Buzen-Bondi (BB) approximation
The Mitrani-King-Nishida (MK-N) approximation
New approximation: DR-A
Comparing DR-A with BB and MK-N
The MK-N and DR-A approximations
The BB approximation
Concluding remarks
Improving traditional task assignment policies
Task assignment in server farms
State of the art in task assignment policies
Task assignment with cycle stealing
Motivation for Cycle Stealing
Modeling and analysis
Performance of task assignment with cycle stealing
Summary of results
Stability
Mean response time
Concluding remarks
Reducing switching costs in cycle stealing
Motivation
Threshold based policy for reducing switching costs in cycle stealing
Analysis
State of the art in cycle stealing
Results
Summary of results
Stability
Mean response time
Benefits of cycle stealing: wide range
Benefit of cycle stealing:
Effect of donor job size variability
Effect of thresholds
Concluding remarks
Designing robust resource allocation policies
Introduction
Static and dynamic robustness
State of the art in resource allocation policies for the Beneficiary-Donor model
Summary of results
Static robustness and mean response time of single-threshold allocation policies
Single-threshold allocation policies: T1 and T2
Stability under single-threshold allocation policies
Mean response time of single-threshold allocation policies
Case 1:
Case 2:
Static robustness of single-threshold allocation policies
Static robustness and mean response time of multi-threshold allocation policies
T1T2 policy
The adaptive dual threshold (ADT) policy
Static robustness of the ADT policy
Mean response time of the ADT policy
Dynamic robustness of threshold based policies
Concluding remarks
Takayuki Osogami 2005-07-19
