Handoff time measurement
Sun Nov 3 05:42:59 PST 2002
Jean Tourrilhes wrote:
>>Any one knows how much time it takes the handoff in a very simple WLAN
>>network with 2 APs (running the Prism2 driver as Master mode) and using
>>a station (as Managed mode) with also running Prism2?
>>I refer to the MAC (layer-2 handoff), without using MobileIP or anything
>>else. I have tried to measured it, but the results I have got are very
>>high: between 1 and 2 seconds!!! When it should be around a few hundred
>>milliseconds (200 at much, maybe). Of course, the handoff time depends
>>of the beacon period that in my case is the default, thus 100 ms.
> That sounds like the proper number. Few hundred ms is totally
>unrealistic, I don't know where you go this idea.
Passive scanning is slow, I agree. That's why active scanning was
introduced. The fact that active scanning might not be properly supported
affects the quality of commercial products. I recommend to read this paper to have an experimental understanding of the MAC layer handoff in 802.11: http://www.it.kth.se/~aep/publications/measurements-kanter-maguire-escudero.pdf
The results shows that the measured handoff time was 157 ms.
Anyway, note that handoff time can be reduced with passive scanning if a
shorter beacon period is used. The drawback is a reduction of the channel
capacity due to higher signalling traffic. It all depends on the scenario.
If your users move fast, then it maybe worth to reduce the beacon period.
>Not also that you don't want to trigger excessive handoffs, so
>because of fading and interference (which may last in the order of
>seconds), you will use a long term average to trigger handoff (like
>SNR), which may delay the start of handoff in some situations. If you
>try to trigger handoff faster (let say, every time you miss a beacon),
>your system will become very unstable (so worthless).
Sure, this is true in general. But in my case I want to trigger handoff
only to distribute the load (thus, disassociation stations from the
APs), so I do not have that stability problem.
> Good luck...
I need it ;)
More information about the Hostap