Microelectronics | IBM eServer xSeries 330
Synopsis
An old IBM server with an interesting story. I bought it back in 2008 or 2009 to store all my web content. It was decomissioned from a datacenter that received new servers. I remember I got it for a tiny fraction of its original price. It was like $ 50 or something. I don't remember what operating system it had (some flavor of Linux, for sure) but I wiped everything off and installed a copy of FreeBSD 4.11. I thought that was a good OS for an early 2000s server. It hosted all my work for around four years, and then I sold it to a friend and got an Apple server instead. He used it for some time and then retired it. As of summer 2024, my friend called me and asked me if I didn't want my server back as it had been collecting dust for the past many years. Thus, I got it back for free (thanks, by the way). As my retrocomputing hardware park has increased in the meantime, I thought it would be a good moment to introduce a vintage server in the network.
I even found a low-resolution picture of this server from when I first bought it. I remember I temporarily placed it under my old 486-based server up until I prepared the new networking infrastructure.
Anyway, let's get back to present times.
When I got back home and powered the machine, I was surprised that it booted into the same FreeBSD 4.11 OS that I installed back then. Since I completely forgot the root password, or my friend has changed it, I had to reset it first. Then I was greeted with that familiar console prompt. Overcome with nostalgia, I browsed the directory structure only to find the software packages and configurations that I initially installed some fifteen years ago. I think I spent around four hours logged on to the server and when I looked at the clock, it was already around 2 AM. I powered everything off and called it a day.
The next day, when I started the server with the intention of cleaning everything and performing a fresh install, surprise: the SCSI drive start command failed systematically. The platter motor starts when the wake command is issued, but the magnetic heads don't execute their initialization routine. Either the drive electronics are fried, or the heads are suffering from the said stiction (a contraction of static friction) phenomenon. While this is a failure mode of early drives, I tried the known trick of knocking the drive on all sides. I couldn't manage to get the head block to initialize. So much for the old SCSI drive.
Anyway, this is nothing that I cannot fix. And the thought of experimenting with retro servers was a catalyst of my wish to continue with this project.
The hardware configuration is as follows.
- Dual Pentium III / 1.4 GHz in SMP (symmetric multiprocessing) mode
- 512 Mb RAM (accepting a maximum of 4 Gb) of type ECC SDRAM at 133 MHz, registered DIMMs
- 3½ 1.44 Mb floppy disk drive
- One Ultra-160 SCSI drive of ~36 Gb (with support for an additional drive)
- 24x IDE CD-ROM drive
- Two 33 MHz / 64-bit PCI-X expansion slots
- S3 video controller with 8 Mb VRAM
- Two 10/100 Intel Ethernet controllers
- Two RS-485 advanced system management processor ports (in / out)
- One Serial port
- Two console ports (in / out)
- Single 200 W power supply
Sounds like a pretty decent hardware configuration, for a server of the early 2000s.
Maintenance
I plan to address the following issues.
- Clean the unit, both internally and externally. This server hasn't been cleaned since 2002. Well, I once blew the dust while it was in my possession.
- Replace the cooling fans as they are very loud.
- Preventively replace some of the electrolytic capacitors.
- Maintenance of the power supply unit.
- Add more memory.
- Add an industrial CF card in an IDE adapter for the operating system installation.
- Replace the dead SCSI drive with a working one, and add a second drive tray.
- Add a better network card.
There are quite a lot of operations, so let's start.
Cleaning the Server
First, I removed all internals and cleaned the case. I effectively washed it with the shower and some detergents. Next, I blew the water with compressed air and then let the case sit in the sun for a couple of hours. Finally, the case looks very clean.
This is the front panel PCB with the power on and reset button. There are some LEDs hidden behind those prisms for various functions such as power and stand-by, error state, and server monitor selection.
I installed the mainboard, which IBM calls a planar. There is an interesting contraption: somebody (appears to have been very skilled) added a VGA port and a PS/2 keyboard connector. The wires are soldered directly on the solder side, to the appropriate pins on the console out port. The adaptation works perfectly, without the need for a special splitter cable.
SCSI drives connect to the system planar board through a dedicated backplane which in turn is connected to the mainboard via a SCSI cable. The SCSI backplane has its own ID and is listed in the devices table during the boot sequence, when the controller searches and initiates the attached SCSI peripherals.
The CD-ROM is made by LG for IBM, and it's a standard component that should also fit laptops. It connects to the system planar via IDE interface. There is an interesting springy wire clip that secures the CD-ROM to the metal tray.
The floppy disk drive is located above the CD-ROM, on the same metal tray. It is fixed via a similar springy wire clip mechanism. The floppy drive connects to the system planar via a fragile flexible ribbon cable.
There are two 3.5" hot-swap SCSI hard disk drive bays. My machine is equipped with only one IBM-branded disk drive of around 36 Gb, made by Seagate. It has its own proprietary hot-swap tray, with a pass-through connector for adapting Ultra-160 SCSI drives to the IBM SCSI backplane. The other drive bay is filled with a plastic cover. The plan is to find another tray and mount two disk drives in RAID-1.
But for now, I removed the dead SCSI drive, and cleaned the hot-swap tray and the dummy filler piece.
That's it for now, in terms of cleaning.
Replacing the Fans
The cooling fans in this unit are 40x40x20 mm, and are very noisy. In reality, nobody cares about server-grade fans and their noise. My server rack is in the same room where I spend most of my time when I deal with computers or electronics. This means it has to make as less noise as it is possible. The fans are made by Nidec and are of type C34637-58, part of the BETA V series. They are rated 12 V / 0.13 A and have an airflow of 8 CFM at 8,000 rpm, producing a whooping 31.2 dB(A).
I decided to go with Noctua NF-A4x20 FLX fans. I like Noctua products and I tend to use them extensively in my projects. These are rated 12 V / 0.04 A with an aiflow of around 5.5 CFM at 5,000 rpm. These fans produce around 14.9 dB(A). Immediately, we can see the Noctua fans are weaker than their server-grade counterparts. Since my rack is open frame type, and the average temperature in my room is around 20 to 26 °C, I say the new fans will be enough.
Since the new fans have a lower maximum speed than the Nidec parts, I expect the server will complain about this and for sure will trigger error events for fans 1 through 6, in no particular order. Knowing how the BIOS of these machines work, the fans will most probably always run at their maximum rated speed as the server thinks there's a cooling issue (which is correct). Thus, I expect the front (and back) error lamps will always be lit. In addition, the individual fan error LEDs on the system planar board will be lit. I don't think there is a way to overcome the rpm monitoring routines.
There are two fans by the power conversion circuit section.
In addition, there are four fans blowing air directly to the dual CPU heatsinks.
And a view from the other side. As with any 1u server implementation, things are packed a bit tight and general case airflow is not that good.
Once I finished rebuilding the server, I did some basic cable management using plastic zip ties.
These fans have a very conveniently placed cable guide on the side with the rotor. Unfortunately, the cables don't fit that easily. I had to remove a small portion of the thermo tube in order to expose an additional 5 mm of those three cables. Then I was able to pass each one of them through that small guide. But the results are promising.
I did the same with the two fans located by the drive bays, next to the power conversion section.
The machine will be powered on sporadically for experiments and research. Thus, silence is gold.
Preventative Capacitors Replacement
Machines of this age often come packed with sub-par capacitors, most of them suffering from the dreadful early-2000s capacitor plague.
Fortunately, this one has high quality electrolytic capacitors installed. There are some very good polymer capacitors made by Sanyo, part of the OS-CON series. These should not be replaced. There are two capacitors made by Matsushita and twelve made by NCC (or UCC, if you consider the US subsidiary of Chemi-Con). Nevertheless, electrolytic capacitors that ran for a longtime in harsh environments, particularly high heat, could be preventively replaced.
As for replacements, I went with organic polymer capacitors made by Panasonic (who bought Sanyo's OS-CON division) and Nichicon. I would've used only OS-CON parts, but I couldn't find 560 uF / 25 V capacitors. The radial leaded OS-CON series stops at 20 V for capacitances larger than 470 uF.
The following list contains the parts that I replaced.
| IBM eServer xSeries 330 System Planar | ||||
|---|---|---|---|---|
| Identifier | Value | Qty | Notes | Mouser Number |
| C690-C695 | 1500 uF / 6.3 V | 6 | Polymer Capacitor | 667-6SEPC1500M+T |
| C147-C149 | 1500 uF / 6.3 V | 3 | Polymer Capacitor | 667-6SEPC1500M+T |
| C90, C109, C125 | 1500 uF / 6.3 V | 3 | Polymer Capacitor | 667-6SEPC1500M+T |
| C327, C1365 | 560 uF / 25 V | 2 | Polymer Capacitor | 647-RNL1E561MDS4 |
There is a group of six 1500 uF / 6.3 V capacitors by the first CPU assembly. Next to the voltage regulator there is a 560 uF / 25 V capacitor.
I removed the old parts and installed the new OS-CON capacitors. The capacitor in the lower-right area is also of type aluminum organic polymer, and it's made by Nichicon.
Another group of six 1500 uF / 6.3 V capacitors is located next to the second CPU assembly.
I replaced all of them with new OS-CON capacitors.
I forgot to take a picture of the original Matsushita part installed in position C327. But here is a picture of the replacement, same brand and type as C1365.
This operation took around two hours. While I don't particularly like to replace parts on multi-layer PCBs, by using the right tools, the process could become more pleasant. I used my trusty old Weller soldering station set at 375 °C, and my Engineer SS-2 desoldering pump (solder sucker).
As always, I was careful not to damage the soldermask or the solder pads. For instance, in the following picture you can see the terminals of the cluster of four new capacitors, located near the first microprocessor.
Here is a final picture of the old capacitors.
Before tossing these capacitors in the bin, let's take some notes on their current parameters.
| CAPACITOR DATA | |||||
|---|---|---|---|---|---|
| Brand | Type | Style | Rating | Capacity | ESR |
| Chemi-Con | Electrolytic | Radial | 1500 uF / 6.3 V | 1233 uF | 0.083 Ω |
| Chemi-Con | Electrolytic | Radial | 1500 uF / 6.3 V | 1247 uF | 0.082 Ω |
| Chemi-Con | Electrolytic | Radial | 1500 uF / 6.3 V | 1237 uF | 0.082 Ω |
| Chemi-Con | Electrolytic | Radial | 1500 uF / 6.3 V | 1234 uF | 0.083 Ω |
| Chemi-Con | Electrolytic | Radial | 1500 uF / 6.3 V | 1231 uF | 0.084 Ω |
| Chemi-Con | Electrolytic | Radial | 1500 uF / 6.3 V | 1245 uF | 0.083 Ω |
| Chemi-Con | Electrolytic | Radial | 1500 uF / 6.3 V | 1220 uF | 0.083 Ω |
| Chemi-Con | Electrolytic | Radial | 1500 uF / 6.3 V | 1236 uF | 0.083 Ω |
| Chemi-Con | Electrolytic | Radial | 1500 uF / 6.3 V | 1235 uF | 0.084 Ω |
| Chemi-Con | Electrolytic | Radial | 1500 uF / 6.3 V | 1229 uF | 0.082 Ω |
| Chemi-Con | Electrolytic | Radial | 1500 uF / 6.3 V | 1232 uF | 0.082 Ω |
| Chemi-Con | Electrolytic | Radial | 1500 uF / 6.3 V | 1231 uF | 0.083 Ω |
| Matsushita | Electrolytic | Radial | 560 uF / 25 V | 473 uF | 0.122 Ω |
| Matsushita | Electrolytic | Radial | 560 uF / 25 V | 489 uF | 0.115 Ω |
The measured parameters of the Chemi-Con LXZ series 1500 uF / 6.3 V capacitors are pretty consistent. Albeit being around 250 uF lower than their rated specs, the ESR is very low. And it has kept being so for the past 20+ years. On the other hand, the Panasonic FC series 560 uF / 25 V capacitors measured around 80 uF less than their rated capacitance. But ESR is still low. For the record, all measurements were made at 120 Hz. So, these are good capacitors.
I normally collect only two digits after the decimal point. But in this case, I had to type down three digits, revealing a tiny ESR value variation between the LXZ series capacitors.
Well, that's kind of expected from reputable capacitors. However, old electrolytic capacitors might fail at anytime. While they were powered on, they were constantly formed and the oxide layer was consistently maintained. Once the machine is turned off for longer periods of time, the oxide layer degrades and upon switching on, the inrush current could be so high that it could cause more serious damage to the system planar.
In the end, with the new capacitors, the reliability of the system planar has increased. A side effect is the more modern appearance.
Maintenance of the Power Supply
The power supply has never been opened nor cleaned of dust. However, it has always been powered on for the past 20+ years. Most of the time idling, as in the past decade, the server was mostly unused, but still connected to the mains. It's interesting that it still works and nothing blew up.
There are two cooling fans made by Superred, rated 12 V / 0.14 A. The model string is CHA4012CB-MA-O. These fans are also 40x40x20 mm and are noisier than the Nidec parts above. In addition, one of them has faulty bearings and produces a loud squeezing sound.
I cleaned the PCB assembly and took notes of the electrolytic capacitors so that I know what to order.
The metal case is now free of dust, ready to receive the power supply PCB assembly.
The original fans have inverted control wires. They are red(1)-yellow(2)-black(3). Noctua fans have a standard black(1)-red(2)-yellow(3) configuration. I had to carefully invert the Noctua fan pins inside the connector.
The cooling fans smoketest passed with success, thus on to the next step.
Given this PSU was always-on, I decided to replace the electrolytic capacitors as part of the preventative maintenance process. The original parts are made by (Nippon / United) Chemi-Con and held up pretty well. There are a few 10 uF / 50 V green sleeve capacitors made by Lelon that I've often seen in hardware this old. While I don't particularly trust this brand, it's also true that I haven't seen a failed one so far. In addition, the sole high-voltage ripple filter capacitor is made by Rubycon. Back then, manufacturers were more serious when they chose suppliers for electronic parts.
Anyway, I will replace all of them with new capacitors. I made a list of what is due for replacement.
| IBM eServer xSeries 330 Power Supply | ||||
|---|---|---|---|---|
| Identifier | Value | Qty | Notes | Mouser Number |
| C | 1 uF / 50 V | 3 | Film Capacitor | 80-R82CC4100AA60J |
| C | 3.3 uF / 50 V | 1 | Electrolytic Capacitor | 232-50ML3R3MEFC4X5 |
| C | 10 uF / 50 V | 5 | Polymer Capacitor | 80-A759BQ106M1HAE105 |
| C | 47 uF / 25 V | 4 | Polymer Capacitor | 80-A755KS476M1EAAE25 |
| C | 100 uF / 25 V | 2 | Polymer Capacitor | A755KS107M1EAAE025 |
| C | 470 uF / 16 V | 1 | Polymer Capacitor | 16SEPC470M |
| C | 680 uF / 10 V | 2 | Polymer Capacitor | 80-A758KR687M1AAAE08 |
| C | 2200 uF / 16 V | 2 | Electrolytic Capacitor | 661-EKY160ELL222MK2 |
| C | 3300 uF / 10 V | 2 | Electrolytic Capacitor | 661-EKY100ELL332MK2 |
| C | 2700 uF / 16 V | 2 | Electrolytic Capacitor | 661-EKY160ELL272MK3 |
| C | 150 uF / 450 V | 1 | Electrolytic Capacitor | 871-B43501B5157M |
In order to increase reliability, I went with a selection of film, polymer, and standard electrolytic capacitors. Practically, for the use cases that I plan for this server, I could've selected any type of electrolytic capacitor from one of the reputable brands.
Adding More Memory
The system planar PCB is equipped with four memory expansion slots. They accept 3.3 V PC133 registered ECC SDRAM modules. My machine was factory equipped with 512 Mb, but according to the documentation, it supports a total of 4 Gb if using 1 Gb modules.
Finding old memory modules, especially of type registered ECC, proved to be difficult. I am still looking for some compatible RAM sticks.
CompactFlash Card for the Operating System
Inspired by more modern servers that have a dedicated internal USB port for the hypervisor software, I decided to add some form of solid-state mass storage. And since there are two IDE interfaces, one being occupied by the CD-ROM drive, I went with a CompactFlash card in the primary IDE port.
I used a Transcend Industrial CF170 card with 32 Gb. It's funny how technology has evolved in all these years. When this server was new, CF cards were 16 Mb, 32 Mb, and maybe 64 Mb and went for outrageous amounts of money. These days, a 32 Gb garden-variety card costs less than a retro 36 Gb Ultra-160 SCSI drive. Industrial-grade CF cards are a bit more expensive, but nothing out of the ordinary.
Next, I installed a minimal variant of FreeBSD 4.11 and all the required packages. The pkg_add utility fails to find the default FTP address as it was moved years ago. But with a bit of fiddling around, I was able to find the FreeBSD 4.11 FTP repository online and manually typed the address. I was even able to install a variant of Midnight Commander.
Replacing the SCSI Drives
I opened the lid of the old SCSI disk and checked the internals. The head armature moves freely and, as expected, there is no stiction phenomenon. Since the drive was compromised anyway, I inserted it in the drive bay and powered the server with the drive lid open. When the controller wakes the disk, the spindle motor rotates correctly, but the head block remains parked. My guess is that either the electronics are fried or track #0 became bad. Or any other possible mechanical or electrical defect.
The Adaptec controller on this server can interface both Ultra-160 and Ultra-320 SCSI drives. These are electrically interchangeable. I finally found some IBM re-branded drives on eBay for next to nothing. These are noisy drives and are not really sought after by retro computing enthusiasts. Honestly, neither am I too happy with these 10 k rpm noisy devices. But I don't think I can find any slower 5,400 rpm Ultra-160 SCSI drives. Anyway, I bought four drives. Better to have spares as it's certain that magnetic drives will die sooner or later.
I'm glad I was able to find these drives, complete with compatible IBM trays and SCA backplane adapters. I initially wanted to test some HGST (Hitachi Global Storage Technologies) drives, model number IC35L036UCDY10-0, but I ended up receiving four identical Seagate drives of the same type and model as the original drive. Not bad at all; as long as they fit and are in working condition, it doesn't really matter who made them.
I inserted two trays with new drives and booted the machine. After formatting the drives, I ended up with a healthy storage system. And since I have two spares, I might also experiment with Windows 2000 Server.
My plan is to configure these two drives in RAID-1. Also, for the most basic networking experiments, where I don't depend on mass storage, I will unplug the trays so that I only use the CF card (in silence).
Adding a Better Network Card
The networking capabilities of this server are ensured via two 10/100 Ethernet interfaces, made by Intel. For such an old server hardware, dual 10/100 NICs should be enough for most tasks. However, I would like to experiment with routing tables and make this machine act as a router.
And I have the perfect network card for this purpose. It is made by Interface Masters and the model type is Niagara 2261 REV. D2. It connects to the server via a 64-bit PCI-X bus running at 133 MHz. The card has a built-in programmable bypass circuit to take the ports offline in case of power failure.
In the past, somebody must've been very proud of having a dual Gigabit network card, thus added a label.
Nice layout, and quite an unusual implementation with mechanical isolation relays.
There might be a catch, though. While this card will most probably work with the em(4) FreeBSD driver, it has to be programmed first. There was a software called niagara_util to program the relays. If I fail to find it, then things will get tricky. Or I could hardwire bypass the relays completely. But that would defy the intended purpose of the card implementation.
Curiosities
There is a factory repair (or improvement!?) consisting of a wire bodge going from one of the voltage regulators to a couple of logic ICs. The nice black and copper colored chip is the ServerSet III northbridge, part of the ServerWorks chipset. The R logo stands for Reliance Computer Corporation.
If the machine is booted into the diagnostics mode by F2 during the boot sequence, PC-DOS is loaded from ROM and a diagnostics program is executed. By examining the various options, one of them interrogates the SMBIOS memory. There is a lot of cryptic information, but if you scroll enough, the system planar switches and jumpers are described, together with their various configurations. I bet there is some PDF online showing the same thing, but having it actually embedded and accessible from within the server itself, is an unexpected nice feature.
Conclusions
This was quite an unexpected project. I always liked enterprise-grade hardware, and given their fairly low resale price, I wonder why I stayed away from them for such a long time. Maybe because of the memories of a jet engine running in my storage room when I hosted all my work locally. However, technologies evolved, and these old dinosaurs are now prone to extinction. I'm happy I could save one of them.
Now I can experiment with network booting of client PCs though PXE.
Your Help Matters
This content is provided as-is and is not for commercial purposes. It reflects my experiments and research and should be treated as such. I release my work to the public for educational purposes. I did all this at my own expense and in my free time. So if you like my work, or find it useful or inspiring for your projects, please consider making a donation.
Copyright © 2004- Alexandru Groza
All rights reserved.
VER. 1.0 | REV. A