My Little Red Server
As the power consumption of my current server (Cartman, an AMD Duron running @ 800 MHz) got a bit out of hand (fat electricity bills and all) I was delighted when a colleague pointed me at www.mini-itx.com, where the new motherboard form factor as proposed by VIA was showcased. I immediately fell in love with the Epia board, and as someone ordered a large quantity of the things I immediately jumped the bandwagon to get one myself.
Of course I would not want to go the standard route and get a standard Cubid case, however nice they are, but devised a case mod myself. I found a nice little modem in my pile of junk electronicks, and decided that it should house my Epia board. But how to power it? Even though the power supply requirement are modest, ATX-wise, and the Cubid only has a 12 V, 60 watt power brick with it, it still is a lot to make from a linear (non-switching) supply. I found a nice 5 V/15 A switcher at work, which would nicely supply the +5 and +3.3 volts needed by Epia. But it also needs +12 volts, and so does the harddisk. I eventually found an old wall wart from a trashed scanner. Rated at 1.2 amps it would uncomplaining supply the 2 amps needed by the harddisk (Seagate Barracuda IV 40 GB), but I did not trust the peak current handling of the little board. So I mounted it on another PCB, and filled it up with reservoir electrolytics to cater for that. I ended up with 8 times 2700 µF, or over 20 mF. That's a lot! I don't know if I really need that large amount, but I just wanted to be sure and besides, the ex-wall wart does not complain... Lucky for me Epia doesn't require negative voltages!
As always, I ran into trouble. Epia would startup, but refuse to initialize the Ethernet adapter. With an ATX supply it would work fine. So a trick was needed: I had to supply +5V_SB (standby voltage, made out of the +12 volts with a 78L05, as the main +5 V is unstable with no load) and then switch Epia out of standby so it would enable the main power supply. After some searching I found two nice power FETs in my junk box, a HUF75343G (rated at a whopping 70 A!) and a IRF9620 PFET. The latter is used for the +12V, the HUF for the 3.3 and 5 volt rails. These semiconductors are not critical, as long as they can handle the 8 amps or so needed by Epia (and thus have a nice low R(DS)on) it's OK. Now the Ethernet interface came up properly. Oh, and the CPU fan was waaaay too noisy, so it now runs on 5 volts. With the fan in the top of the case it needn't do all that much anyway.
There turned out to be some more power supply issues: as the 15 A 5 V switcher is apparently not "hard" enough, and the 3.3 V regulator feeds the mobo directly I had a few cases of win2k freezing, where the HD LED would be full on and the display was utterly frozen. I guess that the memory got corrupted, which is deadly for any OS, and to counter this problem added some nice big electrolytics across the 3.3 and 5 V outputs, close to the mobo power connector. This seemed to have solved this issue. As of writing, Epia had run flawlessly for over four days. The capacitors were 105 °C units salvaged from scrap power supplies, the 3.3 V one is 2700 µF/6.3 V and the 5 V is 4700 µF/6.3 V. The 105 °C types have the low ESR needed for this kind of application. I decided that the 12 V supply had enough capacitance :-)
A New Case!
That left the matter of the case. At first I only wanted to use the modem box to house everything, but a good friend of mine persuaded me to let him modify the box so it would look good. So he took it to a friend of us both, Maestro, who has a load of milling equipment and other gear an electronicks engineer is afraid of... They had a lot of fun drilling, milling and boring away, doing the things a mechanical engineer can be proud of!
After the mechanical work was completed my friend spray-painted the case fire-engine red, mounted the top fan and the fingerguard and he was done. So I picked it up at Maestro's place and immediately put everything in place. The harddisk, which initially was meant to be mounted upside-down on the top cover, now had to be mounted on standoff brackets, floating over the ATX switch board. The power switch came next to the power cord receptacle, and the power and HD LEDs were pushed in their holes in the front panel. The fan in the top cover runs on about 4.5 volts, producing ample airflow to cool the innards. I did have to remove the HD IDE plug bracket (still present in the pic below) to accomodate the fan. I had fucked up the measurement, but hey, I'm an electronicks engineer... Time to fire it up!
It works pretty smooth, and sits there being beautiful on top of my Monstercomputer. I didn't connect keyboard etc to it, just a network cable, and use remote control to do stuff on it. Right now it serves a Shoutcast stream, a website, an FTP server, and also does my Usenet and P2P downloading. And it's much more silent than my old server! Oh, and the Netbios name is Epiaja, which is Dutch and means something like "Epi-yeah right"
So yeah, right, I ended up with an unstable system. I seem to have a subscription on this particular mode of failure, that I would just do some browsing or whatever, and it would freeze on me. Aaaagh! History repeats itself! But what is the cause? I tried every fix known to man but to no avail. Then it dawned that my power supply could be the cause of my troubles. I hooked up a scope to the +12V line, and saw the supply voltage drop with about a volt with harddisk activity. I connected a heavy duty resistor across it and BANG! my system froze. Or rather, it froze silently... So I decided to ditch the wall-wart kludge and delve into my stack of electronic junk. I came up with a Pace digibox, which featured a TOP200 switcher with a nicely chambered open bobbin for its transformer. The power supply section had a width that would fit exactly next to Epia. I attacked the board with a saw and severed the section. Then I took out the transformer to study it. The +5 V output had 6 windings in parallel of 3 turns each, so to make 12 volts I would need 7 turns. And then 8 windings, to cater for the current. Rewire the whole shebang, add electrolytics and inductors, rewire the drive for the optocoupler to adjust the output voltage to 12 volts and presto! A new PSU!
The new switcher will deliver more than 4 amps under screeching protest, but with a modest load (1 amp for Epia and 2 A peak for my Seagate) it would not complain, and run cool. In the end it turned out that the actual source of the voltage drop was not the wall-wart PSU but the switching FET. After replacing it with an IRF9540 and moving all parts to the power interface board beneath the harddisk all was fine.
Power Supply Quirks - Will It Ever End?
Or is this another case of RTFM? When looking into the power requirements for Epia I found that +5V_SB was specified as min 0.1 A, max 1.5 A and peak 2.0 A. And have all this supplied by a tiny 78L05? I don't think so! Exit 78L05. In any case, the reason for the 100 mA regulator (see above) was that the 15 A supply would pulse without load. This proved to be no problem for Epia, as apparently the standby voltage is sufficient to switch on with the power button. Also, Epia is always ON anyway. With load, the 5 V supply is stable. I hope. If not, I'll ditch it. Probably for another phat Pace kludge. Time will tell...
All this work for nothing! In the end it turned out to be Billy Gatez' fault! The freezing problems (that seem abundant according to reports across Internet) appear to be a feature built into Windows 2000. After I did a "format c:" and installed Red Hat Linnuks 8.0 on my lovely box all was fine. At time of writing it has been up for 53 days:
So, yes, I made another Pace PSU mod for the 5 V supply. It proved to make no difference at all. By this time I was desperate. Inspired by my colleagues @ work I decided to install Linux instead. I dreaded this as I would have to figure out how to convert my win32 apps to Linux. But then it also was a challenge. So I set to it, Apache web server was easy, the FTP server needed some work to staighten out, but Pure FTP deamon works like a charm now. The only thing not going is the audio streaming as I couldn't get an mp3 encoder for raw audio input. They are available, so this is just a matter of doing more searching.
Also I changed my phan grille for a flashy "radioactivity" fingerguard. OK, it only works for real big fingers, but what the heck:
Copyright © Zappy TV 2002, 2003