QoS Criteria for Energy-Aware Switching Networks

Modeling of Clos Switching Structures with Dynamically Variable Number of Active Switches in the Spine Stage

“…The results of the analytical calculations are shown in the form of solid lines. All results are presented in relation to the traffic offered to one AU in an output link of MSF: The study confirms high accuracy of the PGBMTC method in modeling MSFs with a variable number of switches in the middle stage, considerably higher than the accuracy of the method proposed in [10]. For the example considered in the article, the relative error of the method (in relation to the results of the simulation and averaged for the entire considered range of offered traffic) was approximately 5%, whereas, in the case of the method proposed in [10], it amounted to 30%.…”

“…A decrease in the number of switches in the internal (spine) stages of a switching fabric is followed by the creation of new structures of MSF, that is, switching fabrics with compression, which in consequence leads to a necessity to consider blocking the switches of the first stage in the model. Until then, this type of blocking was considered exclusively for the case of single-service networks [22,23] and will be analyzed in MSFs, according to the best belief of the authors, for the first time (the simplified model of a multiservice network with a variable number of switches of the middle stage, proposed by the authors in [10], did not take the blocking probability for the switches of the first stage into consideration). The results obtained on the basis of the proposed model can be used to design energy-aware, variable structures of switching fabrics and networks of data centers that can operate in any traffic condition, retaining the assumed QoS parameters.…”

“…where [P s (i)] k−u is the distribution of free links in the LAG with the capacity (k − u) component links (Equation (10)). Formally, we can then write…”

“…where the distribution [P s (i)] k is determined on the basis of (10) The application of the truncated distribution [P * s (i)] k results from the additional assumption of one free link in NFAG that is adopted in Equation (26). It is a consequence of the operation of the algorithm to control setting up connections in MSF (Section 3), according to which an event of the internal blocking can only occur when there is at least one free link in a given direction.…”

“…This means that a certain number of switches in the switching fabric will be temporarily deactivated, that is, "removed" from the network structure. At the same time, however, this method for achieving the first criterion limits the possibility of achieving the other optimization criterion since switching off a certain number of switches will consequently lead to an increase in the blocking probability [9,10]. This problem will be solved in the present article through an introduction of the appropriate limitations related to the assumed values of QoS parameters (in particular with the blocking probability) that, regardless of the load, are not to exceed the adopted boundary values.…”