It is worthy to stress on the assumptions made, since the list will lead us to the series of following papers that simply try to relax on these assumptions and therefore achieve higher accuracy.
1 Arrival Rate
This assumption is crucial for the mathematical analysis since it simplifies the problem. Nevertheless in any case, this arrival process describes the way the packets are arriving at a distant network host, and therefore it is not straightforward to model without realizing a possible undergoing application. However, the type-1 class of sleep mode is also working when there is no application going on.
Some authors assume different arrival models without, but nevertheless, succeeding in reasoning these models. In an Erlang distribution is used as a highly customizable distribution.
A different model is proposed using multiple interrupted Poisson processes (IPPs). It would be a challenge to model the arrival distribution, because this would give insight into the hierarchical case of packet level sleep mode and flow level sleep mode. It is foreseen that the operation of the sleep mode algorithm can take place in two different states, the one between toggling on and off an application in the terminal (flow level) and the other under an ongoing application where packets have long interarrival durations (packet level). It is expected that this kind of modeling will unveil a balance between the gain of the algorithm in these two different layers.
2 Outgoing Traffic
Another reasonable assumption is that only the incoming traffic is considered. However, prove that taking into account the outgoing traffic improves significantly the access delay. This improvement actually stems from suppression of unavailability periods which in turn raises the power consumption and lowers the delay. The same result is found by Xiao in a parallel work.
3 Sleep Mode Setup Time
As explained in the previous section, the terminal must exchange MOB_SLP-REQ messages before initiating the sleep mode algorithm. This message transaction lasts for a period durating a small number of frames. It is proposed that this period must be subtracted from the interarrival time of the incoming packets to correctly compute the access delay. In this paper, the resulting interevent times are called packet residual interarrival times.
This time offset is usually small enough to neglect in case of interarrival times between two application streams. However, the effect could be dominating in case of using the sleep mode algorithm during an ongoing flow. In this case, the interarrival times would be generally smaller and comparable to this time difference. In this perspective, the sleep mode setup time remains an important issue.
4 Energy Cost of State Switching
Switching on and off the transceiver is known to consume a fixed amount of extra energy. The authors model this switching cost to compute the final energy consumption. However, modern specification sheets for WiMAX transceiver provide low consumption idle states where the switching to active state is immediate and consumes negligible amounts of energy. Therefore, it seems that modeling the cost of state switching is not worthwhile.
5 Multiclass Scenario
Most of the papers in the literature so far focused on PSC type-1 class sleep mode. As explained above, this class refers to terminals that have only low QoS ongoing flows or no flows at all. However, considering a realistic scenario, one would have to take into account multiple connections of several QoS levels and include possible network management operations and therefore, combine many different sleep mode algorithms (one of type-1 and possibly many of type-2 and type-3) for only one terminal. These several algorithms, running in the same time, define the common availability periods of the terminal which in turn define the final unavailability periods. The terminal is allowed to go to sleep only on that commonly accepted unavailability periods (periods that the attention of the terminal is not required by the BS).
Take into account two different type of sleep mode classes. However, their work is focussed on selecting one of the two possible classes depending on delay and energy consumption optimality. Therefore, the task to analyze the performance of multiclass scenario remains open.
What is more important is the type-3 class. This class is used by the BS itself to organize multicast communications, locate the MS, and perform periodic ranging. Locationing is especially needed in mobile IEEE 802.16e networks to keep track of the mobile distance from the BS and perform handover operations. Including mobility into sleep mode modeling seems to require the addition of at least one extra sleep algorithm of type-3. Mobile scenarios are always multiclass. This assumption seems to be very important for the final results since high speed mobility tends to impose very frequent location updates. Then, periodic ranging is also very important and sometimes requires an extra PSC instance. It is interesting to investigate the effect of management operations in power consumptions of WiMAX terminals through a hybrid multiclass scenario.
6 Full Model Analysis
Most of the published work focuses on estimating the expectation of access delay and energy consumption. The mean value is indeed an intuitive characteristic and clearly valuable for comparisons and testing. The possibility of computing the variance of these measures. On the other hand, derivation of high-order moments or even of the distribution of these measures can be possible for simple interarrival models and could prove to be insightful.
7 Queueing
The analysis described above as well as most of the literature assumes that once the terminal shifts to awake mode it immediately serves the packet. However, by assuming nonzero service time, and for high-rate arrivals it is expected that queueing phenomena will occur. Specifically that the sleep mode algorithm can be modeled by a server with vacations and a M/G/1/K queue.
