29.01.2015 Views

Embedded Software for SoC - Grupo de Mecatrônica EESC/USP

Embedded Software for SoC - Grupo de Mecatrônica EESC/USP

Embedded Software for SoC - Grupo de Mecatrônica EESC/USP

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

382 Chapter 28<br />

space of our failure mo<strong>de</strong>l. Another important parameter we vary is p, the<br />

probability that a packet is <strong>for</strong>war<strong>de</strong>d over a link (see Section 4). For example,<br />

if we set the <strong>for</strong>warding probability to 1, then we have a completely <strong>de</strong>terministic<br />

algorithm which floods the network with messages (every tile always<br />

sends messages to all its four neighbors). This algorithm is optimal with<br />

respect to latency, since the number of intermediate hops between source and<br />

<strong>de</strong>stination is always equal to the Manhattan distance, but extremely inefficient<br />

with respect to the bandwidth used and the energy consumed. Stochastic<br />

communication allows us to tune the tra<strong>de</strong>off between energy and per<strong>for</strong>mance<br />

by varying the probability of transmission p between 0 and 1.<br />

5.1. Target application: An MP3 enco<strong>de</strong>r<br />

MP3 enco<strong>de</strong>rs and <strong>de</strong>co<strong>de</strong>rs are ubiquitous in mo<strong>de</strong>rn embed<strong>de</strong>d systems, from<br />

PDAs and cell phones to digital sound stations, because of the excellent<br />

compression rates and their streaming capabilities. We have <strong>de</strong>veloped a<br />

parallel version of the LAME MP3 enco<strong>de</strong>r [5] (shown in Figure 28-6) and<br />

introduced the co<strong>de</strong> in our stochastically communicating simulator. The<br />

following results are based on this experimental setup.<br />

5.1.1. Latency<br />

The latency is measured as the number of rounds MP3 needs to finish encoding<br />

and it is influenced by several parameters: the value of p (probability of<br />

<strong>for</strong>warding a message over a link), the levels of data upsets in the network,<br />

the buffer overflows and the synchronization errors. Figure 28-7 shows how<br />

latency varies with the probability of retransmission p and with the probability<br />

of upsets As expected, the lowest latency is when p = 1 (the<br />

messages are always <strong>for</strong>war<strong>de</strong>d over a link) and (when there are no<br />

upsets in the network), and increases when and to the point<br />

where the encoding of the MP3 cannot finish because the packets fail to reach<br />

their <strong>de</strong>stination too often. Note that is <strong>de</strong>pen<strong>de</strong>nt on the technology used,<br />

while p is a parameter that can be used directly by the <strong>de</strong>signer in or<strong>de</strong>r to<br />

tune the behavior of the algorithm.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!