Large impact of temporal/spatial correlations on per-session performance measures: single and multiple node cases

Citation
M. Conti et al., Large impact of temporal/spatial correlations on per-session performance measures: single and multiple node cases, PERF EVAL, 41(2-3), 2000, pp. 83-116
Citations number
19
Categorie Soggetti
Computer Science & Engineering
Journal title
PERFORMANCE EVALUATION
ISSN journal
01665316 → ACNP
Volume
41
Issue
2-3
Year of publication
2000
Pages
83 - 116
Database
ISI
SICI code
0166-5316(200007)41:2-3<83:LIOTCO>2.0.ZU;2-7
Abstract
The problem of evaluating the end-to-end performance in an ATM environment is known to be difficult and largely open. Since applications may have very different quality of service (QoS) requirements, a performance measure ass ociated with a random cell would not necessarily reveal the impact of multi plexing on the QoS of a specific application. For this reason, it is import ant that the traffic stream of interest be tagged and its distortion due to multiplexing be evaluated; for analysis purposes the co-existing traffic i s aggregated and forms the background traffic. In the past studies the back ground traffic has been assumed to be uncorrelated and be diverted after a single multiplexing stage. The objective in this work is to evaluate the impact of temporal and spatia l correlations on the end-to-end performance of a tagged traffic stream. Su ch correlations can be significant due to temporal correlation in the backg round traffic or partial commonality in the routing path (background traffi c is not necessarily diverted). A binary queuing activity indicator (QAI) i s proposed in this work to provide for a simple mechanism to capture these correlations. Results derived for various delay metrics of a tagged traffic stream associated with single and multiple nodes show the substantial impa ct of the spatial/temporal correlations. Furthermore, they suggest that the proposed QAI indicator is capable of capturing such correlations. (C) 2000 Elsevier Science B.V. All rights reserved.