Q-PASTE Architecture

Static STELA requires MAC layer modification while dynamic STELA requires both MAC and application layer modifications, both at the mobile client side. Although cross- layer solutions normally indicate that extra overhead is introduced as cross-layer informa- tion exchange is needed, their complexity is significantly decreased when applied in the layered model. Upper layer information, i.e. application layer in Q-PASTE, contained in a control packet is passed down to the lower layer, i.e. MAC layer in Q-PASTE, via the transport layer and the network layer. Moreover, only two control packets are needed in Q- PASTE to pass information about session information and the multimedia streaming status. Therefore the resulting overhead is not significant.

Q-PASTE involves modifications at both the client device and service gateway. Q- PASTE consists of two major components: an application layer module Packet/ApplicaTion manager (PAT) and MAC layer module STELA.

PAT is employed at the application layer of both gateway and client device. The client side PAT keeps track of all active application sessions and informs STELA ,the MAC layer

module, through a cross-layer information flow whenever an application session ends or a new session starts. On the other hand, the gateway side PAT collects real time information about the number of packets ready to be transmitted to the clients and does not allow data releasing until the packets can be shaped into new bursts.

Although data scheduling can also be fulfilled at the server/client side, the packet shap- ing function of PAT is implemented as part of gateway functions in our energy efficient solution due to several reasons. First, client side shaping requires extra cost in terms of both hardware and software complexity, as burst should be formed before being received by the WNIC and thus an extra receiver unit should be installed. Second, if continuous data is manipulated at the server side to form bursts, the client node might be communi- cating with several servers at the same time, and therefore needs to wake up for each burst from each server. Moreover, as a burst is released as soon as the buffer size reaches the threshold, extra delay is introduced when packets from individual servers are considered separately as they will wait longer before being delivered. On the contrary, a service gate- way could be either a media gateway which already exists on the market or a simple router that has built-in application layer control. The gateway can act as a centralized controller and gather data from all the servers for the client. This approach is the most efficient in terms of both energy saving and QoS level. The architecture of Q-PASTE shown in Figure 4.21 is composed of gateway side and client side components. The two components are described in details next.

4.5.1.1 Gateway Side Component Architecture

The Q-PASTE service gateway is composed of classic components at all layers except application and MAC layers where PAT and STELA modules were added. A cross-layer information flow is employed at both client side and service gateway to pass information from application layer to MAC layer. At the client side, real time session information is carried in the flow, while at the service gateway the information about fast start phase involved in traffic shaping is carried in the flow. Additional modules are included at the

Figure 4.21 Q-PASTE architecture overview

application layer: Traffic Shaper, Buffering Timer and the MAC Notifier, as shown in Figure 4.22.

The Traffic Shaper is used to deliberately gather a group of packets into a burst before releasing them to the client. The input of this module, as shown in Figure 4.23 includes the packets received from the network, and the buffering timer which is restarted for every

period of tbfafter the last data release. The output is the decision of whether to release the

buffered data for transmission or not.

The MAC Notifier module, as illustrated in Figure 4.24, monitors the traffic shaping process and gathers information about the time elapsed from the beginning of data transfer. The output of the module is a control message sent cross-layer to the MAC layer. This message has set a new header field, FS END. The default value of FS END is 0. And the field is set to 1 for the first packet generated when the observed elapsed time exceeds

Figure 4.22 Gateway side architecture of Q-PASTE at application layer

Figure 4.23 Traffic shaper in Q-PASTE

the fast start period Tf s. The value of FS END is set back to 0 for the rest of the control

packets. The information is passed down from the application layer to the MAC layer, so that the MAC layer module can take corresponding actions. The process of setting the FS-END value is repeated for every multimedia clip.

Figure 4.24 MAC notifier in Q-PASTE

4.5.1.2 Client Side Component Architecture

The Q-PASTE client side architecture, illustrated in Figure 4.21, includes classic como- nents at all layers except application and MAC layers where two modules are added respec- tively: PAT and STELA. Additionally, a cross-layer information module is also employed. PAT and STELA were described in details before, where as the cross-layer information module provides information from application layer to MAC layer.

A particular component at the client side is the playout buffer, as shown in Figure 4.25. The buffer is used to alleviate the degradation caused by unwanted changes in the data rate. Packets are temporarily stored at the client buffer in order to smooth out any potential bandwidth variation, and are pulled out and then decoded and displayed by the multimedia player. The fill rate is the rate the data enters the client buffer and the drain

rateis the rate the playout removes data from the client buffer. The fill rate varies during

the playback process not only because the network condition fluctuates, but also due to the traffic shaping scheme employed by the service gateway. The fill rate reaches its peak during the fast streaming phase when packets are directly relayed to the client without buffering. The drain rate is related to the encoding rate of the corresponding multimedia content. By bufferring data, the receiver is able to smooth out any temporary variations in the received data rate.

Figure 4.25 Playout buffer in Q-PASTE