Now that isn't new. New are some manifestations I've been dealing with recently.

Late last year I decided to build me a comp system for the living room, for file server and film watching purposes. A box there needs to be reasonably quiet. I got me a VIA EPIA MII-12000 board which is tiny and has lots of gadgetry onboard (sound, graphics, compactflash, cardbus, usb, firewire, tv-out and so on), 512M RAM, an 80G disk and a DVD burner.

 2004_09_18-epia-start.jpg 2004_09_18-epia-finish3.jpg

The first setup was housed in a gutted SparcStation5 chassis, but despite being relatively elegant that didn't cut it: the chassis is too low for big slow-turning and quiet fans and things got either quite hot without any fans (the MII can almost run fanless even with the factory-supplied heatsink) or too noisy with the small fans on. Not good.

So I resorted to the low-tech approach and built version 2 in an IKEA cardboard box (with aluminium foil glued to the sides), with two slow quiet fans: one 12db sucker on the side and the reduced-rpm fan in the 90W PSU. This works pretty well now, and is certainly quiet enough for the living room.

 2004_10_30-mediabox-run2-done.jpg  2004_10_30-mediabox-run2.jpg

After a fair software odyssey I got the TV-out to work, DVD::Rip and transcode like me now and xine does its job reasonably well, too.

However, I want an IR remote control setup. No fun using the keyboard even with a long cable... The board has an IRDA-capable chip, and I got a cheap HP printer IR transceiver that apparently can be mangled into usability as transceiver on a PC.

Well, I must have made some mistake soldering (although I can't figure out what exactly has gone wrong): the stinking thing only heated up on connection and didn't receive or emit any useful magic waves. I think I've fried it. Bugger this, time to scrounge up the parts for building a lirc-compatible transceiver for the serial port then.

My next set of problems concerns the wavelan I'm using to connect all the hardware at home. WEP is a joke, so I use Vtun as a VPN tunnel on top of that. Knowing that Vtun is not very secure, I'm quite tempted to replace it with something better.

One thing I wanted to try is WPA or 802.1X with dynamic WEP keys with a lifetime of less than 5 minutes. This would let me get rid of Vtun; all interactive stuff runs over ssh anyway, email is SSLified: it's mainly proxy access and nfs I run over the wavelan that need protection.

So I upgraded the firmware of my access point (DWL-900AP+...which is not an overly impresive piece of equipment) with the version that allegedly does WPA and 802.1X. And started wondering why the two wireless stations wouldn't talk to each other anymore. Everything was talking fine to things behind the access point, just not within the wavelan itself.

Lots of debugging later I discovered that the stinking POS mangles the ARP responses it sends onto the wavelan: the ethernet frame has the access point as destination address, not the requester - who naturally then ignores the ethernet frame. Stupid! The same ARP responses sent on the wired etherenet are fine, which doesn't exactly aid debugging greatly. The eventual "fix"? The access point is EOLd already, no luck with any fixed firmware, so I added manually published ARP entries to my two wavelan boxes. Bah.

Then I'm seeing pretty weird MTU problems. 1464 bad and makes lotsa fragments, 1463 good. Vtun encapsulates things in UDP with no frills, so packets upto 1472 should be fine (28 bytes UDP + IP header) but they aren't. Weird.

[ published on Fri 14.01.2005 01:00 | filed in brainfarts | ]
Debian Silver Server
© Alexander Zangerl