Layered queueing is a small and elegant extension to queueing networks, to model complex resources that are held in groups when a program executes. Upper-layer servers include the use of lower layers in their service times. The interpretation of results (such as bottleneck identification) is different in layered systems, whether they are solved by analytic techniques or by simulation, or even viewed from measurements.

Layered queueing occurs in all real systems, although the layered effects may be small enough to be ignored, giving the usual "flat" models.

The model concept originated in the "active server" model of Woodside (1984), which was a systematic layered application of the surrogate delays of Jacobson and Lazowska (1983). Contributions have come from a broad group of researchers. Developments include 

• the ":rendezvous network" (Woodside et al 1986. 1995) 
• the "method of layers" (Rolia and Sevcik, 1995), 
• the solution method of Ramesh and Perros (1998), and 
• the generalized layered model of Franks and Woodside (1998) with parallel activities. Also:
• Kurasagi and Kino treated two layers (1999), 
• Kahkipuro developed a model especially for UML (2000), and 
• Menasce treated two layers again, one for software and one for hardware resources (2002).

The Carleton University group that publishes this page has been developing techniques to solve layered models and also to build them and to exploit their results. While our work is fully described here, we want to spread the interest in this approach and include other ideas, tools, and results.