| |
|
Fluid model & Perturbation Analysis
Performance analysis on Resilient Packet Ring model
The Resilient Packet Ring (RPR, IEEE 802.17) is a new high speed backbone technology for metropolitan area networks. Compared with its predecessors(SONET and Gigabit Ethernet), RPR inherits the high speed ring networks' merits: it supports multiple classes of services and different working modes, which diversifies its functionality, the automatic topology discovery and advertisement of station capabilities allow systems to become operational without manual intervention, the dynamic bandwidth allocation increases the network's efficiency, the compatibility of different physical layers also ensures the broad use of RPR in the near future. Meanwhile, in the system performance, RPR excels in its high utilization and global fairness: with the technology of spatial reuse, the bandwidth of the RPR networks is highly utilized, and the fairness algorithm guarantees the fair service allocations to each node.
Been proved to be a good approximation for the high speed networks, fluid model provided very useful heuristics and accurate information for the controlling and scheduling of communication networks, moreover, it makes the traditional queuing networks analysis approach easier and more tractable.
In this research, we will use the fluid model to approximate the RPR networks. The spatial reuse and the fairness algorithm implemented in IEEE Draft P802.17/D3.2 will be the reference based on which our fluid model is established. Starting from the Single Queue Mac Design of RPR, we assumes the flow traffics to be well defined random processes, and the parameters of the fairness algorithm be the one that is under control, and our goal is to analyze the effect of those parameters on the performances of the corresponding fluid model, i.e., the perturbation analysis of the performance metrics with respect to the fairness algorithm parameter. The analysis will be extended to the Dual Queue Mac Design, which is the more general working mode of resilient packet ring.
Fluid approximation of call center with dynamic priority
M. Chen, J.Q. Hu, and M.C. Fu, Fluid Approximation and Perturbation Analysis of a Dynamic Priority Call Center, Boston University Technical Report, Department of Manufacturing Engineering, Boston University, 2004
Reference:
Some theoretical analysis on the priority fluid model
Perturbation analysis of FIFO multiclass fluid model
Other literatures: See Professor Cassandras website
The fluid network analysis at Umass
Back to top > |
|
|
| |
|
Complex System
My research work is focus on the fluid model and fluid networks of high speed internet. The following are some of my viewpoints on the research:
1. On what scale should we look at the queueing models of high speed internet.
Traditionally, queueing model is used widely to analyze the network, in which packets(or jobs) are transmitted discretely. With the increasing complexity of the high speed network, tracking each packet and analyzing the packet's performance became either improperate or efficient. Fluid model was introduced to solve this problem. The basic idea is that, instead of considering each jumps caused by the packet arrival or departure, we smooth the sample path by regarding the sample path as a continuous function. In fact, if we shift our view on the sample path from fine scale to a large scale in terms of the time horizon, every sample path, no matter the complexity of the network, is continuous. In the modern high speed network, fluid model is indeed proved to be a very good approximation for the traditional queueing models.
2. How the fine scale influence the large scale behavior.
In fact, the fluid model is an approximation of discrete queueing model. However, in many cases, the fluid model ignore the details of discrete queueing model. Then what characteristics have been reflected on the corresponding fluid model? What has been obscured? How the behaviors of fluid model is influenced?
Some key events in discrete model,
3. Pattern formation of fluid models
4.Multiple states: Empty, Busy, Overflow... Back to top > |
|
|
