03/18/98 - Confirmed to start at Taos
03/18/98 - Not only start at Taos, but go right to my first contract at sony.
My saved up funds are running out. Even if we continued to pursue the T1 solution, the hardware will cost ~2k and the monthly service 1.2k, which is unacceptable as there are people like jbot in the valley getting 2.5Mbit ADSL for only .35k/month.
We have no choice but to consider a wireless T1 solution - connecting to someone in the valley.
The wonderful service of ml.org provides us
a stable domain name for our unstable ip addresses. Now we have mail,
basic web service, and the ability to get to our machines at home from
anywhere. Yea! It's very s l o w, however, 400-800 ms turnaround time. Ugh.
Going up and down confuses the ip_masquarading code in linux 2.0.30 terribly. Ultimately we just live with a conventional web proxy and ssh into the main box to get to the outside world.
Fired off an email to ricochet telling them of our amazing success, they don't beleive us, greg put up our chart of the line of sight modems we detected.
Greg and I run around installing redhat errata everywhere, add a very restrictive set of tcp wrappers, and breathe a sigh of relief that the hack wasn't worse. It's unlikely he can bother us again, but we are extra paranoid for a few weeks.
Monica's much needed contribution puts us in a spot where we can actually contemplate buying something.
After finding out how much the basic boards and antennas cost (about 1k for each side) Greg and I feel that the charge for the dedicated boxes (3k ea) is highway robbery and we launch into a exploration of what wireless cards were supported by Linux drivers. Two - the old standby wavelan card, and a much newer card, the arlan 655, from Airolan. [ ed: the arlan card is now marketed as the 2300 series ]
After debating the issue with the various salesreps (Jim Bradfield was very helpful), and their technical people, I discover that wavelan depends on the tcp/ip layer to handle retransmits. Ugh. The mosquitonet project's research said that this affected tcp/ip quite adversely in typical wireless packet loss situations. The Arlan 655 does link layer error correction - only under very extreme conditions does it drop packets. That decides the issue and we place an order with Jim Bradfield of Network Solutions.
I get a telescope to survey our domain from the back porch. We can barely make out the Marriot sign directly in the center of our field of view. Extrapolating that from the map we extract from mapquest we figure we are in really good shape to connect to one of two pre-existing sites that had good bandwidth.
I'm dubious. It's an neat hack. But will it work?
Several days hacking followed. Greg gets a 486SLC box for 200 bucks that sort of works. It turns out the ide interface is non-standard and we can't upgrade the hard disk, anyway, so it becomes 'pinion'. We get it to boot and route packets but the arlan driver doesn't work.
Grab the latest arlan driver from Elmer's site, compile it, get it installed,
figure out what a registrationMode and spreadingCode are, and we can send
packets from upstairs to downstairs. YES!
At this point we felt our problems were solved. Just grab some cable, install the antenna on the roof, go down to jbot's or eric's and we'd be in business.
Boy were we wrong.
Greg and I must've skipped the part in Boy scout school that taught about magnetic variance. Our compass was off by 15 degrees and with that both of our sites end up being obscured by mountains.
The RP-TNC can be obtained through any Amphenol dealer, its part number is '31-5677'. The connectors are roughly $5 each, however the dealer may require a minimum order (say $50). We got ours through from:
PO Drawer 99111
Fort Worth, TX 76199-0111
Ultimately we mail ba.internet and the svlug. All sorts of people respond that are willing to help - but only one - A ham by the name Everett Basham - has a real clue about our connectivity problems.
He sneered at our cable and showed me some of his prized heliax. He loans me some good LMR400 for a basic test and lo and behold, we can work.
So we reconfigure the router for the correct 'virtual' network. It works, router to router, but refuses to route packets. Much headscratching and many trips to Everett's roof later I realize it is because the firewall code later down in the config file is still set to firewall packets to the old network.
We leave the router up on the roof and happily ftp stuff back and forth until the router hangs. We reset the router, twiddle the spreadingcode down to 1Mbit/sec, change the MTU to 586, beat the heck out of it some more and call it a night.
- Repeat as nessessary until you realize the real source of the problem.
Similarly, standing in front of the antenna is to be avoided.
Our dinky little 486SLC 25 (pinion) can transfer files back and forth at about 44KBytes/sec. I figure theres lots of overhead dealing with this little 8 bit ISA card and am happy we get even this much performance from the thing.
Greg whips out his visa card and picks up a more capable pentium box, with PCI network cards. Our only hope is that the radio card didn't die as well.
A fresh rebuild later and the new pinion comes up perfectly.
A week goes by while pacbell attempts to address our ADSL problems...
I crawl up on the roof at dawn in my bathrobe and realalign the antenna by eye... it works again!! - and god himself is spitting down on the house, I can barely see the trees at the end of our lot.
I start writing up this report in the hope that it would be useful to others. It absolutely pouring rain, and my ping times to the valley occilate between 12 and 17 ms. Greg said he knew when he sent that mail out about our great wireless uptime something bad would happen. I said: 'don't do that, then'.
11/10/98 - I stay home from work (sick) and take a stab at finishing this report. Of course it's pouring rain and the antenna gets knocked around so back up on the roof I go. While on the roof, I have a revelation - despite all the hassles and expense, the ups and the downs, this project has been FUN, and I wouldn't have missed mucking around with all this technology for the world.
This system costs only a little more a month than hybrid cable, offers full bidirectional near-T1 bandwidth, has lower latency than anything, and ultimately will cost much less per month than the major alternatives (ISDN, 56K, T1) in our location.
12/05/98 - --- the end ---
I'd like to express my thanks to the following people:
To Elmer Joandi and Victor Jennings - thanks for the arlan driver!
To the members of the linuxrouter project - Your insanity is reciprocated!
To Linus Torvalds - Thanx for such a kewl OS!
To Jim Bradfield, who's patience with my questions was unmatched
And to greg, my roomate, who took over when I got discouraged.
Author: Michael Taht