From: Jouni Malinen (jkmaline_at_cc.hut.fi)
Date: 2002-04-05 18:05:30 UTC
On Tue, Apr 02, 2002 at 11:47:25PM -0500, Brian Capouch wrote:
> I hope my question here hasn't already been beaten to death:
>
> Could you discuss for the list a bit the comparative benefits of doing
> AP each of the two different ways?
Well, I'll try to explain my viewpoint on this..
> Some have said the tertiary firmware requires paying Intersil, others
> have said not.
The firmware itself is freely available, but only from Intersil premier site; and access to that site requires buying, e.g., Prism2 reference design ($20k or something like that). AP manufacturers can apparently distribute the firmware with their devices without extra licensing fees, but I'm not sure about the details of this. Anyway, it looks like at least Intersil is not releasing it freely to anyone. Whether someone else that has access to premier site could do this seems to be on quite grey area.
> Some have said the firmware approach is more efficient and faster,
> others say not.
This depends.. AP firmware uses MAC processor on the wireless LAN card for management functions and bridging between associated stations. Consequently, the host processor does not need to reserve processing power for this. Assuming that the processors on Prism2 chipset are efficient enough to not be bottlenecks, firmware approach would be at least as efficient as anything done on the host side. For low cost access points, it is certainly useful to be able to use processors with smaller performance requirements (i.e., which cost less).
I'm not sure, whether current chipset performance is a bottleneck for anything. It might be with WEP since it seems that using WEP reduces throughput. If the host processor has extra processing power, it could have better performance, e.g., with WEP (which would of course had to be implemented in the driver first ;-).
As far as performance of the AP implemented using Host AP mode is concerned, I see no problem in reaching the same level as with firmware-based APs when using modern host CPUs. I have been able to match, and even exceed, the performance of firmware-based APs in my tests using Host AP mode. In addition, the maximum throughput was more or less the theoretical maximum of 802.11b.
However, I do not see performance as the motivation for using Host AP (assuming that using Host AP mode does not reduce performance considerably). The main reason for me is the additional freedom that is available when implementing management functions in the host driver.
Changing how firmware-based AP works, would, in many cases, require having a possibility to modify the firmware. This, however, is next to impossible for anyone outside Intersil (or whoever else has the programming environment and source code the firmware). It would require laborious reverse engineering of the MAC processor code base and firmware binary..
On the other hand, using Host AP mode has, at least limited, documentation that can be acquired (although it still requires NDA). There are still some firmware-based limitations on what can be done, especially on the low-level 802.11 control functions. However, quite many changes that could be done in Host AP mode would require firmware modifications on firmware-based AP implementation.
Especially for research needs, it is most helpful to be able to modify the low level operations oneself. It might also be quite useful outside research work, not to have to rely on one company modifying the firmware if something needs to be changed.
Of course it would be quite nice to have free, open source firmware for 802.11 station and access point functionality to play with, but I do not think this is available anywhere--at least not for any modern, off-the-shelf hardware platform.
-- Jouni Malinen PGP id EFC895FA