I had this idea today since many of us are always trying to put really fast computers on tall towers and often times fall prey to heat problems and reliability. The RB532's of today aren't fast enough to run an NStreme AP. And it's far too expensive to mount 4 or 5 P4's on a tower just for redundancy sake (I hate having a single AP computer that is the primary source for failure on a tower). We're always struggling trying 533mhz VIA C3's and such, when a RB532 CAN do it but can't do it and everything else too.

So I propose a "dumb" software AP running on an RB112 or RB532 - one that uses a faster, ethernet connected "controller" Mikrotik that could be mounted indoors in the tower control room, that can control multiple "dumb" AP's. This could be a way to move towards a GPS synchronized AP NStreme cluster. I would hope that it would be possible to use 2 controllers to support a failover arrangement as well. This could also be used for backhauls as well - so everything comes into the controller computer as raw data from these APs & links, and it takes care of all of the heavy processing tasks, while the dumb AP's do nothing other than bridge packets between themselves and the actual controller.

I'm not proposing to make a new distribution, but rather a simple package set, on the dumb AP, system + wireless-remote, designed to do nothing more than send raw data (not IP but actual 802.11 raw data) back and forth between a faster computer mounted below.

On the controller, system + wireless-controller + all the other packages, that is designed to take care of authentication, encryption, packet aggregation, routing, bridging, conn-track, etc (all of the CPU intensive tasks), and acts as if it were just another interface (RemoteWLAN as an idea). So you can have a tower with 6 AP's on it, and the machine below will list 6 "virtual" remote AP's above, and do everything for those APs.

Any thoughts on this?

Sounds like ARUBA wireless switching work!

Regards
Lutz

Now that i see that Aruba stuff, it seems like even a better idea.. too much burden is placed on the AP for processing, it seems like having a remotely installed processing unit might allow even a RB112 to get full throughput instead of needing bigger computers up top. I did a BW test from a P4 2.8 to a RB112 via ethernet and was able to get a 65Mbps UDP 10s average P4>RB112, says to me maybe if the software was doing less on the RB112 that maybe a RB532 could do even better. I could envision a wireless package that takes 802.11-encapsulated packets from the ethernet power, and sends it out the wireless directly. Same thing in recieves - just send the raw packet down the ethernet to the main system. Throw in a GPS sync and TDD and now it's carrier class (esp. if you could use some form of redundant switches at the bottom so that if one controller fails, another can take over)...

Maybe a bit easier would be a wireless card that has the main part in the pc indoors , and just sends up IF on a few hundred MHz to the rest of it mounted on the tower ?

interresting idea... sounds like something I would like to test should it ever be implemented...

currently we do the "next best thing" to that, we put a 532 with a SR5 on the tower (set to 330Mhz) turn on nstream and polling, then bridge wlan1 to a vlan (specific to that AP) tied to ether1

at the base of the tower, we have a single P4 machine, with 2-4 ethernet ports (depends on how many backhaul's are at the tower) we create a vlan for each AP, and then add a PPPoE server for each vlan, using the same profile for each server. this makes it so that one address pool is shared by all the sectors on the tower.

VLAN setup... (there are usually 4 vlan/PPPoE config's per tower, 3 sectors, and a omni for failover, i removed the rest as it's just redundant) PPPoE Server Sertup This config makes our AP's dumb for all routing purposes, the only thing they have to proces is the actual wireless link (and nstream)... it helps, we don't have any routing overhead or anything. I can get a fully loaded (25 cpe's to a sector is our max) rb532, it runs at 25-35% cpu on average, with an average load of 3-9mbps going across it...

Well I'm looking to support VOIP (4-24 lines per customer), and high speed internet (2Mbps-10Mbps per customer) for between 10-15 customers per AP. I have already done what you've suggested, except I just route the data back to my main server via backhaul (and use basic IP rather than PPPoE). I don't want to have a system that can't handle the full speed of NStreme + Polling, and overclocking RB532's seems to be a bandaid in my opinion. The WAR boards have enough horsepower for the right price, but no MT on those.

The WAR boards have enough horsepower for the right price, but no MT on those.

didn't MT say they're not going to support the IXP?

that would be the solution soo many people have asked for... there are countless posts about wanting a 500MHz+ board...

MT didn't mentioned anything precise. But they should not wait any longer,
think they are one of the last ones around. All the other
vendors (software + hardware solutions) around doing nearly the same
(one of them is the one you mentioned with the board) have already
switched to IXP.

I do not think that IXP is the only solution, but at least its a very powerful
one compared to the MIPS MT is using at the moment.

seandsl
--

Nearly 4 months ago a upcoming more powerul RB was hinted at on the forums here in a few threads---but we have heard nothing since then...

We need a more powerful CPU and support for 3-4 high-powered 400mw cards, and then I think we will be off to the races!

SMA