Computing Reviews
Today's Issue Hot Topics Search Browse Recommended My Account Log In
Review Help
Search
Service overlay networks: SLAs, QoS, and bandwidth provisioning
Duan Z., Zhang Z., Hou Y. IEEE/ACM Transactions on Networking11 (6):870-883,2003.Type:Article
Date Reviewed: Apr 21 2004

A service overlay network (SON) is a way of providing end-to-end quality of service (QoS). The SON provider purchases bandwidth from the individual network domains to build a logical infrastructure on top of existing networks. Users do not have to negotiate with multiple network providers. Instead, they pay the SON for the value-added services.

This paper focuses on the so-called pipe service-level agreement (SLA) model, in which the route between source and destination is predetermined. The bandwidth provisioning problem tries to determine how much bandwidth is needed for each link, such that the QoS requirements of users can be fulfilled, while the overall revenue of the SON provider is maximized.

First, the static provisioning problem is solved. It is determined a priori how much bandwidth is needed for a relatively long period of time. Because traffic demands fluctuate, a certain amount of over-provisioning is needed. Once the over-provisioning parameter is fixed, the optimal amount of bandwidth for each link can be derived, simply as the ratio of average traffic demand on that link and the link utilization that should not be exceeded to ensure QoS, corrected by the over-provisioning parameter.

Next, the authors try to obtain the over-provisioning parameter, assuming that a QoS violation leads to a penalty for the SON provider. An exact solution is difficult to obtain, so an approximate solution is derived. Because QoS violations depend on traffic demand patterns, the solution is evaluated for different traffic models (self-similar M/G/∞ traffic, and a model based on real Internet traffic measurements). First, this evaluation is performed for the simplest case (a single link), but then more complex topologies (tree, meshed tree) are studied, with varying traffic loads.

The mathematical formulation becomes more complex. As a consequence, an intuitive interpretation of the results is not always obvious. It is clear that in the case where too much bandwidth is over-provisioned, bandwidth cost will be higher than the potential penalty reduction. However, the authors also claim that the amount of over-provisioning must be substantial, otherwise the penalty reduction will be too small to offset the additional bandwidth cost. This seems counter-intuitive, especially because a linear cost function is used.

Finally, dynamic bandwidth provisioning is considered. In this case, the SON still has to determine how much bandwidth should be reserved statically, but it can now allocate additional bandwidth dynamically at a higher cost (if this cost were not higher, there would be no reason for static provisioning). Again, an approximate model is constructed first, but now it is followed by an adaptive heuristic, which takes online traffic measurements into account. Because of the small delay between measurement and bandwidth adaptation, it is still possible that QoS will be violated occasionally for very short periods.

Reviewer:  F. Put Review #: CR129478 (0410-1174)
Bookmark and Share
  Reviewer Selected
 
 
Network Management (C.2.3 ... )
 
 
Network Communications (C.2.1 ... )
 
 
Pricing And Resource Allocation (K.6.2 ... )
 
 
Network Architecture And Design (C.2.1 )
 
Would you recommend this review?
yes
no
Other reviews under "Network Management": Date
Network management standards
Black U., McGraw-Hill, Inc., New York, NY, 1992. Type: Book (9780070055544)
Jun 1 1993
Network management
Held G., John Wiley & Sons, Inc., New York, NY, 1992. Type: Book (9780471927815)
Mar 1 1993
IBM Systems Journal (v.31 n.2)
Hoffnagle G. (ed)  IBM Systems Journal 22:1992. Type: Journal
May 1 1994
more...

E-Mail This Printer-Friendly
Send Your Comments
Contact Us
Reproduction in whole or in part without permission is prohibited.   Copyright 1999-2024 ThinkLoud®
Terms of Use
| Privacy Policy