[KLUG Advocacy] re: OT:uPNP

[Mike Williams]

>
>
>From: Adam Tauno Williams <awilliam at whitemice.org>
>  
>
>>>> >You can disbable NetBIOS on a per-interface basis
>>>      
>>>
>>> Unfortunately, that won't work, as I'd like to have this machine be a 
>>> bit of a file server as well.  
>>    
>>
>
>To the external network?
>  
>
Good point.  Hadn't occurred to me that dialing would be triggered by 
the outside interface, not the inside one.

>
>>> I suppose.  Half the time it wouldn't dial when I punched up a web or 
>>> ftp session either, though.  On-demand is likely to be a problem, 
>>> though, because there will be two Windows boxes behind it, and Windows 
>>> boxes have a nasty habit of having "update software" (some of which 
>>> blurs the line between legitimate update checks and spyware) that calls 
>>> home regularly.  I think it will be more trouble than it's worth 
>>> configure complicated rules explaining what IPs or protocols trigger 
>>> dialing and which ones don't.
>>    
>>
>
>This is simplified by using a proxy;  simply setup a squid proxy on the
>edge-device.  If the squid server requires authentication than all this
>automatic gunk is automatically broken, which is beautiful.  And you get
>a performance kick for free.  HTTP requests blocked by the proxy won't
>invoke demand-dial since they will never reach the external facing
>interface.
>  
>
Squid being the reccommended package?  One more thing for me to figure 
out for this little project.  Is there a way to make browsers properly 
autodetect that they need a proxy?  I've seen stuff like that on the 
client side but have no idea how to set up a server that way.  Also, how 
many obscure protocols work through a proxy, and what will have to be 
routed through using regular NAT?

>>>>>> >>> Option b might be worth trying if I can't manage option a, which 
>>>>>> >>> involves dumping the modem for DSL.  The modem is a piece of junk 
>>>>>> >>> anyway.  External, so it's a real modem (not winmodem) but it only 
>>>>>> >>> manages a connection above 31.2k about once every other February 29th.
>>>>>          
>>>>>
>>>> >Seems normal, all the 56k noise is total nonsense; it is just this site of 
>>>> >impossible to ram 56k over a phone line - a T1 channel with clear-channel 
>>>> >encoding only does 64k,  add in the DAC/ADC process and the unbalanced 
>>>> >impedence of a POTS pair....  I've seen phone lines from buildings where I 
>>>> >could see the CO from the roof clock in at under 28k.
>>>      
>>>
>>> The wiring's good.  
>>    
>>
>
>I'm talking about the wiring buried in the dirt or dangling between dead
>branchless trees.
>
I know.  Overall, it is good enough to do 56k (40k or so really) if the 
hardware at both ends is compatible.