<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.0 TRANSITIONAL//EN">
<HTML>
<HEAD>
<META HTTP-EQUIV="Content-Type" CONTENT="text/html; CHARSET=UTF-8">
<META NAME="GENERATOR" CONTENT="GtkHTML/3.6.2">
</HEAD>
<BODY>
Hello everyone,<BR>
<BR>
I resume my old problem:<BR>
Hardware+software: 2 WRAP.1D, each with 2 miniPCIs CM9 (atheros chipset=>madwifi drivers), latest voyage linux installed on both with no additional software on top.<BR>
Test: Configuring two point to point links between the two WRAPs, and using multipath routing with iproute2, one of the WRAPs (or both sometimes) was hanging when transfering at high speed (generally after a few seconds of transfert).<BR>
<BR>
<BR>
WRAP1 ath0 (Master)-------------------(Managed) ath0 WRAP2<BR>
WRAP1 ath1 (Master)-------------------(Managed) ath1 WRAP2<BR>
<BR>
<BR>
After a few more investigations (it could have been a power issue), I randomly tried configuring the interfaces differently:<BR>
<BR>
WRAP1 ath0 (Master)-------------------(Managed) ath0 WRAP2<BR>
WRAP1 ath1 (Managed)-------------------(Master) ath1 WRAP2<BR>
<BR>
<BR>
Miracle!!! It doesn't hang anymore, and is very stable (about one week transferring with netperfs/iperfs/wgets,..... and still up)! Seems it was madwifi after all... Anyone think it could still be a power issue (maybe Master mode eats more power??)?<BR>
<BR>
Now that a stable configuration was found, it works perfectly for small speeds (wifi links negociated at 36Mbps and under) and multipath routing does its job, multiplying by 2 the throughput (for more than one TCP flow). The problem now is that I cannot get more than 40Mbit/s (using the two links) as it seems that above this speed it eats all cpu and become way over loaded. Knowing that I get around 35Mbps with a unique link (I'm talking about small distances), I should be able to get between 60 and 70Mbps with multipath routing.<BR>
<BR>
My question would be if someone thinks it could be possible (and worth the work) to optimize the distribution (using a different kernel flavor for example, or recompiling everything for 586MMX architecture make a better use of the Geode processor...) enough to be able to acheive these speeds without overloading the board.<BR>
<BR>
Thanks again very much for reading me,<BR>
<BR>
Francois.
</BODY>
</HTML>