I've recently experienced an issue at a couple client locations:

AP site:

RB532 (v3.6) + Senao 8602, Maxrad 13 dBi @ 120 deg vpol sector, 20 mhz g-only @ ch11 (noise levels low)

Affected clients:

* RB133c (v3.6) based client, Senao/Engenius 8602, Rootenna 14, installed in an attic shooting through a vent surrounded by metal insulation, 3 blocks away (txpower reduced to 10 dBm, verified problem still exists at default)

* Netgear WGT634U running OpenWRT w/ Madwifi, sitting in a customer's windowsill ~100M from AP (set to 5 dBm xmit, problem also exists at default txpower)

This AP has 5-8 other clients connected at all times, all clear LOS that get consistent 54/54 connections.

Both of these clients have excellent signal. The RB133/8602 setup shows -58 dBm both ways when set to default txpower. The Netgear shows -53 dBm down/-71 up.

The problem we're seeing is if I leave all rates enabled (in B/G or G-only modes) these clients choke down to nearly zero (sub-25kbit average) throughput, and the TX CCQ as seen from the AP drops down to 3-10%. Soon as I disable the higher speeds they show 85%+ consistently and the client throughput is much higher.

This can only mean theres something very wrong with the rate control algorithm Mikrotik is using. I've noticed in the past adjusting the rate control algorithms used in madwifi (Linux driver for Atheros devices) could greatly improve performance under suboptimal conditions. My suggestion would be to allow multiple options for rate control or at least allow the customer to adjust the thresholds for the stock rate control algorithm. Locking the 'station' device's rates doesn't seem to stop the AP from selecting the wrong rates.

Any ideas?