On the way home Wednesday evening, I stopped by Fry's Electronics, and purchased a water cooling system for MUSE and a USB print server. Interesting fact: there are a lot of print servers on the shelves, but virtually all of them are either parallel or USB 1.1, which means that you can send your print job at 100 megabits per second, and then the print server forwards it to the printer using two tin cans and a piece of string. The one I bought was the only USB2 print server I could find there.
Upon returning home, I opened up the water cooling system box and discovered that I had an interesting problem on my hands. It is sold as a closed, preloaded system: the cooling block/pump which attaches to the CPU is already hooked up to the radiator and filled with cooland. It is not intended to be detached by the end-user. That's OK if you have a mutant case which can take a 120mm fan (although I wonder how effective the radiator is when you're blowing hot air from inside the case through it), but I don't. Furthermore, there are very few openings in the back of the Thermaltake case which are large enough for either the cooling block or the radiator to pass through. The most obvious one is the power-supply opening, if I temporarily dismount the power supply. I might also be able to remove one of the case fans and obtain a suitable opening that way. Both of these options and probably any others will require permanent modification of the case, which I'd rather not do, but right now I have no other ideas. We'll see what I come up with on Sunday when I can actually experiment with all the pieces in question. Oh, and I have no idea how I'd go about mounting the radiator and fan outside the case — having it dangling from its hoses seems like a bad idea.
Anyway, with most of my machines down due to the hot weather, that was an ideal time to redo the power distribution, so I did. Originally I had a neat layout of what machines attached to which UPSes, but my plans have changed several times since then, and there have been ad-hoc additions where I just used the most easily available power connection. Fixing the situation was conceptually easy, it mostly just involved rerouting cables. Now the small UPS at the top powers the networking gear and the router/NAT laptop; the upper large UPS will power the new fileserver and a couple other machines; and the lower large UPS will power MUSE and my main machine.
As part of redoing the power, I pondered the plight of all the small USB peripherals which I have accumulated. They had been sort of sitting around on top of various cases, but the new arrangement wasn't very conducive to that. The new fileserver would be too narrow to sit anything on, and the Thermaltake case was too tall. There was also the matter of the printer. I eventually decided that reality says that I won't be doing streaming audio broadcasts while I'm still in this apartment, at least not on a regular basis — I'm just not going to have the time. That meant that I could remove the audio gear box, freeing up the shelf immediately above the larger machines, which made space for the USB peripherals and the printer.
Since the printer now had a permanent home in easy reach of its host machine, I'd just spent $70 on a print server that I didn't need. Fortunately, I hadn't even broken the shrinkwrap, and Fry's is good about returns. (It went back Thursday morning.)
While everything was still down, I decided to follow up on my resolution to rationalize my FDDI ring, especially since I'd identified what appeared to be a damaged fiber on Tuesday night. I started by simply removing all existing fiber, and in the process did the lightbulb test on the damaged fiber and confirmed that it was bad. (Optical fibers are easy to test: point one end at a bright light source like a bulb, and look at the other end. If you see a little speck of light, the cable is probably good. If you don't, it's definitely bad.) After that, I sat down and made a diagram of the machines on the ring and what kinds of connectors they had (there are three flavors of connectors which are normally used for FDDI, and all three are represented on my ring), then derived a list of what kinds of cables I'd need.
Have you ever had what I call a "contagious failure"? Thing A doesn't work, so you check it, and it's bad. You grab thing B, check it, and it's fine, but when you put it in place, it doesn't work either. Take it out, and thing B is bad, but suddenly thing A seems to be fine... except that it doesn't actually work when you put it in place. And you know that the stuff on either side is OK because it works fine when you connect it directly together. It usually means you're overlooking something which is changing.
In my case, it was a bad adapter that was causing all this pain, one that was required to convert between connector flavors. The cable would be fine, then I'd put the adapter on, and suddenly it would be bad. It took me a while to make the connection because I wasn't paying attention to whether the adapter was on or not. When I finally made myself stop and asked, "Something must be changing, what is it?", then it suddenly became obvious. A quick lightbulb test on the adapters revealed that one showed a little spot of light as expected, and the other didn't. A replacement adapter, and everything was happy.
Well, almost. I'd discovered the original "bad" fiber because I saw a flashing light on a card when poking around behind the machines. When I shut down one of them for the night later on, I saw a flashing light again...
Because FDDI is meant to be daisy-chained, that means that a ring can be broken if a machine is turned off. A dual-ring configuration, like I have, can survive one machine being turned off, but breaks down if two or more are off. As a result, any piece of FDDI equipment which is intended for dual attachment also has a connector for a bypass switch. Bypass switches are optomechanical relays: when the equipment supplies power, the relay changes position and connects the cables between the switch and the equipment to the ring connections. When it doesn't supply power, the relay goes back to its off position which connects the ring connections to each other. Signals from the machine in one direction now pass directly through to the machine in the other direction, no electricity required.
Of the six machines on my ring, the three which may be on or off all have bypass switches. The machine that I'd just shut down was in the middle of that group. I knew that everything on either side of it was fine because while that machine was on, the ring was complete. I also knew that its bypass switch was sticky... a quick lightbulb test revealed that it had gone from sticky to stuck. When the machine was powered off, one of the relays went back to the off position, but the other one did not.
I have a spare switch, but this failure is a bit annoying because the switches I have were never meant for use with the equipment I have. Not only do they have the wrong kind of optical connector on them, requiring the use of adapters, they also have the wrong kind of electrical connector as well. Instead of the RJ11 (phone-jack style) which most of my FDDI equipment uses, they have a DIN connector. Not even a miniDIN, which some FDDI equipment uses, but rather a full-size DIN connector which nothing FDDI that I've ever seen has used. That means that replacing the failed switch will require lopping off the existing electrical connector and splicing a new one one, hoping that I get all the connections right.
I'll do that Sunday.