To better understand experimental environment we introduce additional indicators
about HO strategy performances. Those indicators underline differences between HP- and
SP-HO strategy and does not depend on predictor behavior.
Table 2 shows performance indicator obtained in 6 different simulations of
similar time length (about 30 minutes). The indicators are:
- s/ap, average time length between two consecutive WiC re-association;
- EBcloseHO%, percentage of empty buffer with close HO;
- EBnotHO, number of empty buffer without re-association.
We define two consecutive HOs
close if they happen with less than 3.2 s in the
between. Downloader and Player execute concurrently; so starting from an empty buffer
we will get a full buffer within
200 KB / [(1500-1000) Kbit/s] = 3.2 s.
After an HO buffer is always almost empty. Then if the next HO happens at less
than 3.2 s, buffer becomes completely empty certainly. For this reason we do not
consider as error if there is not available data in the buffer after two close HOs.
Note we consider two close HOs as the result of bouncing effect. So EBcloseHO%
is an indicator of HO strategy regularity; grater is EBcloseHO%, more irregular
is a particular HO strategy because bouncing effect happens often.
We introduce experimental results about a traditional fixed size (not adaptive) buffer to
underline dependency between above indicators and only HO-strategy, without any
dependency with HO predictor. Actually different HO strategies carry out different
performances, but different HO predictors does not affect performance values.
| |
Handover
Strategy |
s / AP |
EBcloseHO% |
EBnotHO |
| Fixed size |
HP |
18 |
18.5 |
1 |
| SP |
26 |
5.9 |
0 |
| Actual RSSI |
HP |
19 |
9.0 |
0 |
| SP |
29 |
3.5 |
0 |
| GM-filtered RSSI |
HP |
17 |
18.2 |
1 |
| SP |
29 |
1.3 |
1 |
Table 2. Performance indicator about experimental environment.
HP strategy always performs much more re-associations than SP strategy.
HP performs a re-association every time a visible AP RSSI overcomes the currently
associated AP RSSI, plus Hysteresis Handover Threshold (HHT), instead of SP that
delays re-associations according to Fixed Handover Threshold (FHT). HP-HOs
happen more frequently than SP-HOs. For this reason
s/ap
is considerably greater in SP than HP and vice versa
EBcloseHO% is
considerably greater in HP than SP. Moreover useless HP-HO due to RSSI fluctuations
increases re-association number. HHT value is 2 dB; a greater HHT lowers
number of useless re-associations but it also delays
too much wireless client re-association, reducing HP-strategy proactive behavior.
On the contrary SP-strategy is not affected from bouncing effect because SP-HO
happens when the currently associated AP RSSI is really low (less then -80 dB = FHT).
Note occasionally it happens buffer is completely empty without a HO. It is due
to multithread architecture of our proposed solution.
It could happen Player process executes too much and Download process just a bit.
If it happens just after an HO, when buffer is almost empty, buffer becomes
completely empty even if at the moment there is not a HO.