Register a SA Forums Account here!
JOINING THE SA FORUMS WILL REMOVE THIS BIG AD, THE ANNOYING UNDERLINED ADS, AND STUPID INTERSTITIAL ADS!!!

You can: log in, read the tech support FAQ, or request your lost password. This dumb message (and those ads) will appear on every screen until you register! Get rid of this crap by registering your own SA Forums Account and joining roughly 150,000 Goons, for the one-time price of $9.95! We charge money because it costs us money per month for bills, and since we don't believe in showing ads to our users, we try to make the money back through forum registrations.
 
  • Post
  • Reply
You Am I
May 20, 2001

Me @ your poasting



Sweevo posted:

There's literally no reason to buy a 500 Plus. They never work due to battery damage, and the few that do work sell for more than the minor upgrade over the 500 is worth. Get a 500 or 600, or if you're feeling rich a 1200.

Too late, just did the purchase

Battery has been removed from the system from what the eBay store has told me. Already got a 1200. Probably slap in one of those PiStorms into it for acceleration/RAM/storage

Adbot
ADBOT LOVES YOU

Gromit
Aug 15, 2000

I am an oppressed White Male, Asian women wont serve me! Save me Campbell Newman!!!!!!!


I've had an Amiga 2000 since I bought it in 1988 or so, and it's been in a quantum state of "battery leaked or not?" since then. The longer I don't open the case, the longer I can imagine that all is fine.

SubG
Aug 19, 2004

It's a hard world for little things.


longview posted:

Everyone was either on an actual bus (i.e. same physical coax) or connected using hubs (which basically emulate a bus network even though it's wired as a star network).
There was no way to avoid collisions where two or more computers wanted to transmit at the same time.
So each computer was required to try to detect if it had tried to transmit on top of another computer, if this was detected then the computers involved would delay a random amount and try again.
If the network isn't super busy this works (the random delay is to prevent them from immediately colliding again).

Pictured above is an actual packet colliding on a 10Base-2 network with only two computers on it, from earlier this year. The normal signal amplitude is 4 divisions, when the collision occurs the voltage doubles.
This really just begins to scratch the surface of how much thinnet sucked. Like collisions are a way that thinnet could suck while still theoretically functioning, but it also had dozens of ways of sucking that would bring down literally the entire network.

Improper termination, improper physical spacing of devices, segment too long, improper grounding, all poo poo that could take the entire segment down. A single bad BNC connector (or fried AUI) on one device could gently caress the entire segment, and the only way of troubleshooting it was usually to go from host to host swapping poo poo out until it started working. Also a lightning strike nearby could fry every AUI on the segment.

longview
Dec 25, 2006

heh.


SubG posted:

This really just begins to scratch the surface of how much thinnet sucked. Like collisions are a way that thinnet could suck while still theoretically functioning, but it also had dozens of ways of sucking that would bring down literally the entire network.

Improper termination, improper physical spacing of devices, segment too long, improper grounding, all poo poo that could take the entire segment down. A single bad BNC connector (or fried AUI) on one device could gently caress the entire segment, and the only way of troubleshooting it was usually to go from host to host swapping poo poo out until it started working. Also a lightning strike nearby could fry every AUI on the segment.

Oh for sure, I can attest to the termination being tricky.
When I set up that network I figured "Well this is just point to point so termination isn't needed", long story short turns out the interface doesn't work at all unless you put 25 Ohms loading on the line.

If you really wanted to using a TDR could have helped a lot back then, but they were probably ludicrously expensive when they would have been needed.

Lightning is interesting, I noticed both the Thinnet and AUI ports on the Apple Ethertalk NB were fully isolated. My guess is touching the coax and the computer case at the same might have tingled a bit back then (just like touching cable-TV coax and anything grounded back at my parents house).
If the coax had proper ground bonding, it shouldn't have been particularly sensitive to lightning. (It wasn't properly grounded, it was installed by whoever did the networking who just wanted it to work.)

-
I have now gone down the rabbit hole of building an actual server using my old Q9300 based motherboard, stuffing it into a 2U random enclosure I had laying around (not a computer case).
This will be my old-world server, either W2k or W2k3 server (both should have Mac compatibility, but 2003 might be required to get some of the drivers installed, we'll see).
The Q9300 is ridiculously powerful for this purpose, really should have kept the Athlon 64 3000+ for this purpose, but I figure I'll underclock it as far as it'll go to save power anyway.

Also got the Apple Ethernet NB Card working yesterday, after spending 1 hour+ fixing the pinout of my weird adapter cable. Note: completely unrelated to the EtherTalk NB [Ethernet] Card.
So now the AAUI port goes into a 9-pin D-Sub I added to the computer (pretty neat job), then to an internal AUI transceiver modified for 5 V power, then that plugs into an internal switch, which then connects to the duplex LC fiber optic port next to the D-Sub plug. There were a lot of places to get the wiring wrong.

The actual Ethernet NB Card is cool since it has a dedicated 68000 CPU running A/ROSE which can offload what appears to be a fairly large amount of work from the CPU.
I added the optional extra 4 Mbit of RAM as well before testing but not sure it makes a difference.
It's basically as fast as the old Novell Etherport card at actually transferring data over FTP etc. (~200-250 kb/s).
The difference is that with the unaccelerated card the CPU is always busy managing the ethernet/IP stack stuff, while the new card definitely reduces the interrupt rate to the CPU.
End result is the computer is way more responsive when doing file transfers (and if you background Fetch, it will actually still go fast in the background). Web browsing became noticeably faster as well, especially loading images.

E: also I found this in the ROM:

It's been tainted with libertarianism, might have to perform an exorcism at some point...

longview has a new favorite as of 07:06 on May 5, 2021

Computer viking
May 30, 2011
Now with less breakage.

longview posted:

My guess is touching the coax and the computer case at the same might have tingled a bit back then (just like touching cable-TV coax and anything grounded back at my parents house).

More than a little bit, depending on the electrical state of the connected computers. I've gotten a few unpleasant jolts.

You Am I
May 20, 2001

Me @ your poasting



Gromit posted:

I've had an Amiga 2000 since I bought it in 1988 or so, and it's been in a quantum state of "battery leaked or not?" since then. The longer I don't open the case, the longer I can imagine that all is fine.

Nooooo

Open it up and get rid of that lovely Varta barrel battery

stevewm
May 10, 2005


Computer viking posted:

More than a little bit, depending on the electrical state of the connected computers. I've gotten a few unpleasant jolts.

I had this happen not too long ago with a DVI cable. Computer kept locking up, I noticed my hand buzzed when I put it on the case. Got a slight shock unplugging the DVI cable. Measured ~40VAC between the case and just about everything.

This led me to discover a few failures. None of the outlets in that room had a ground; it had become disconnected in an upstream outlet. Also my monitor was leaking current to ground due to a failure in its power supply board. Without a ground, it ended up on the chassis instead, and thus on the shield of any connected video cable. Plugging the monitor into a GFCI outlet would make it pop immediately.

DreadUnknown
Nov 4, 2020


longview posted:

Oh for sure, I can attest to the termination being tricky.
When I set up that network I figured "Well this is just point to point so termination isn't needed", long story short turns out the interface doesn't work at all unless you put 25 Ohms loading on the line.

If you really wanted to using a TDR could have helped a lot back then, but they were probably ludicrously expensive when they would have been needed.

Lightning is interesting, I noticed both the Thinnet and AUI ports on the Apple Ethertalk NB were fully isolated. My guess is touching the coax and the computer case at the same might have tingled a bit back then (just like touching cable-TV coax and anything grounded back at my parents house).
If the coax had proper ground bonding, it shouldn't have been particularly sensitive to lightning. (It wasn't properly grounded, it was installed by whoever did the networking who just wanted it to work.)

-
I have now gone down the rabbit hole of building an actual server using my old Q9300 based motherboard, stuffing it into a 2U random enclosure I had laying around (not a computer case).
This will be my old-world server, either W2k or W2k3 server (both should have Mac compatibility, but 2003 might be required to get some of the drivers installed, we'll see).
The Q9300 is ridiculously powerful for this purpose, really should have kept the Athlon 64 3000+ for this purpose, but I figure I'll underclock it as far as it'll go to save power anyway.

Also got the Apple Ethernet NB Card working yesterday, after spending 1 hour+ fixing the pinout of my weird adapter cable. Note: completely unrelated to the EtherTalk NB [Ethernet] Card.
So now the AAUI port goes into a 9-pin D-Sub I added to the computer (pretty neat job), then to an internal AUI transceiver modified for 5 V power, then that plugs into an internal switch, which then connects to the duplex LC fiber optic port next to the D-Sub plug. There were a lot of places to get the wiring wrong.

The actual Ethernet NB Card is cool since it has a dedicated 68000 CPU running A/ROSE which can offload what appears to be a fairly large amount of work from the CPU.
I added the optional extra 4 Mbit of RAM as well before testing but not sure it makes a difference.
It's basically as fast as the old Novell Etherport card at actually transferring data over FTP etc. (~200-250 kb/s).
The difference is that with the unaccelerated card the CPU is always busy managing the ethernet/IP stack stuff, while the new card definitely reduces the interrupt rate to the CPU.
End result is the computer is way more responsive when doing file transfers (and if you background Fetch, it will actually still go fast in the background). Web browsing became noticeably faster as well, especially loading images.

E: also I found this in the ROM:

It's been tainted with libertarianism, might have to perform an exorcism at some point...

Yessss this is the good poo poo, this is why I read this thread.

EVIL Gibson
Mar 23, 2001

Internet of Things is just someone else's computer that people can't help attaching cameras and door locks to!


Switchblade Switcharoo

When we upgraded our shared house with the fastest of 100 Mbps Ethernet so we all could split the internet freshest and hot DSL (this was 2001 and we lucked out because we lived right next door to the switching house) I used Windows internet connection sharing and a hub with the LONGEST cables we saw at the time. 50' cable was very expensive.

Used a god drat hub and that collision light was on 24/7 if someone was using Napster and the rest of us were playing Tribes.

I got ShowEQ running I think only because I was using a hub. I kept loving up the proxy rules and I think the only reason I was able to see anything was it was broadcasting the same EQ1 traffic to both my main EQ box and the ShowEQ box.

Guy Axlerod
Dec 29, 2008


I had a share house where the wifi router was on the opposite end of the house from my room, and it was pretty crap. I managed to dumpster dive a hub and a poo poo router we put into switch only mode, and used those to extend ethernet to my room. The poo poo router was in one of the other guy's rooms, so we all ended up on ethernet. Even with the collisions, it was still better than the wifi.

My most important use case was to SSH into our shift trading system. A reliable connection was more important than speed for that. Hearing the beep of a new shift available before someone else picked it up was key.

I also didn't have a desk at the time. I sat the monitor on top of the tower, and the keyboard and mouse were balanced on my lap.

longview
Dec 25, 2006

heh.


So after spending a few hours with a dremel cut-off wheel, the 2008 era ATX motherboard is installed in this tiny 2U rack box. It was a perfect fit for low profile PCI cards, fit to the mm.
Now I just need to figure out if Windows 2000 is happy running on a quad core from 2008, looks like there are drivers available for everything I need so I'll give it a go.
First attempt at USB booting the installer wasn't super successful, errored out at reading the EULA.

Currently waiting on:
AMD HD 7300 poo poo tier graphics card, using a PCI card for now since I don't have any low profile pci-e cards but I don't want to burn a PCI slot for graphics since the whole point is having a PCI capable server
Low profile heat sink (the one I have now fits, but there will be literally 0 airflow if I use it since the fan sits against the top cover)
64 GB IDE SSD (the little HyperDisk DOM type that plugs right into the motherboard plug) - using a 16 GB one I bought for the 486 for now
A few 60 mm 10 mm thick 3-wire fans to get some airflow into this terrible case

The Q9300 was a stately processor for its time, ran Crysis fairly well when paired with a Radeon HD 4870, but it's way too power hungry for this little airflow. Terrible for overclocking if I remember correctly.
Seems to be stable running at 1 V VCore, 1.5 GHz core with 600 MHz RAM.
Below that it wouldn't POST, but it appears to be running extremely cool like this so hope it's stable.

E: and I can barely believe it, but I had actually ran out of VGA cables, something that has literally never happened to me before. I guess making all those Mac II adapter cables ate up my entire supply.

longview has a new favorite as of 16:29 on May 6, 2021

Laslow
Jul 18, 2007



longview posted:

So after spending a few hours with a dremel cut-off wheel, the 2008 era ATX motherboard is installed in this tiny 2U rack box. It was a perfect fit for low profile PCI cards, fit to the mm.
Now I just need to figure out if Windows 2000 is happy running on a quad core from 2008, looks like there are drivers available for everything I need so I'll give it a go.
First attempt at USB booting the installer wasn't super successful, errored out at reading the EULA.
You can use a USB boot disk utility to make a MS-DOS boot disk and copy the files from the Windows ISO to the FAT32 portion of the drive. Then you can boot it up and run WINNT.EXE from the i386 folder in DOS the old fashioned way. I think the USB utility I used back in the day was from HP. Whatever you use needs to support the floppy images you can get your hands on, usually .FLP or .IMG or maybe .IMA. Itís been well over a decade since Iíve had to do that, sorry Iím not more specific.

longview
Dec 25, 2006

heh.


I got it working with WinSetupFromUSB 1.9, which adds a giant GRUB structure on top of the installer, remounts the USB disk so it doesn't take priority of the primary etc etc.
It even has an automatic driver loader feature that can run before the 1st stage install that magically added appropriate drivers for the storage.

Had some issues I thought were caused by 2000, so I just got Windows 2003 running, but I think the real issues were hardware.

The issues were seemingly caused by 1) bad network card, 2) SSD overheating

I remember Realtek PCIe network cards were always troublesome in a really terrible way on this hardware, randomly corrupting data is very bad. The card I've tried now is not the one I had issues with before but seems to have the same problem.
Going to have to find an Intel card (that worked well previously with this exact hardware), but looks like the on-board nVidia ethernet is fine for now.

SSD overheating is easily fixed with a fan, it will be sitting in the air stream anyway.

So gonna try with 2000 server again, it's much easier to deal with (activating 2003 Server is not trivial these days even if the key is legitimate).
If the storage issues were in fact a driver issue with W2k then I guess I'll go for 2003.

Sweevo
Nov 8, 2007




I ran 2000 on a Q9300 until 2012. The only issues I remember were that 2000 doesn't support .net 3.x or above, which started to become an issue around 2010 when most things switched over from 1.x/2.x. Other than that it ran pretty much everything XP did.

longview
Dec 25, 2006

heh.


Literally re-installing 2000 (1st stage install) right now, I have high hopes!

It might have also been the southbridge overheating (it was definitely very warm), it just got stuck with the HDD LED hard on for several seconds at a time. HDTune gave a thermal warning of over 70 degrees when I started it so that's probably related.

I noticed the SSD reports UDMA 6 capabilities, but I only got 30 MB/s. Seems it doesn't ground out pin 34, so the BIOS complained about using a 40 pin cable and limits the speed to level 2 or something.
There's literally no cable here, so I just soldered a strap to ground out that pin, we'll see if there's more performance to get here when I get the OS installed.

E: oh yeah, faking an 80 pin cable got the speed up from 30 MB/s to 70 MB/s, it's very noticeably faster, and that's basically within the spec too (they claim up to 80, but I'll call that close enough).

longview has a new favorite as of 21:40 on May 6, 2021

Computer viking
May 30, 2011
Now with less breakage.

Disabling the onboard realtek and switching to a 10+ year old intel card just solved a weird hang problem in modern Linux on a fairly new Ryzen system here, so rest assured their long and proud tradition of making horrible "technically this provides networking on Windows most of the time" hardware is still going.

longview
Dec 25, 2006

heh.


Computer viking posted:

Disabling the onboard realtek and switching to a 10+ year old intel card just solved a weird hang problem in modern Linux on a fairly new Ryzen system here, so rest assured their long and proud tradition of making horrible "technically this provides networking on Windows most of the time" hardware is still going.
I'm just hoping it's not a PCIe related issue - I used to run a Intel gigabit PCI card back in the day after kicking out the Realtek crab.
I ordered a low profile "Intelģ PRO/1000 PT Dual Port" which is a bit overkill but I figure a decent PCIe dual gigabit card will come in handy in the future as well.


Strange situation now, the northbridge is the single warmest component in the computer.
Since the CPU is so underclocked & undervolted it only runs around 10 degrees above ambient with the fan at idle, while the northbridge is mostly running full speed. Frankly it would make more sense to put the CPU heat sink on there instead and use passive cooling for the CPU.

Added a 40 mm maglev fan to the heat sink, which brought it down to around 25 degrees above ambient (~50 degrees).


That dip near the start of the plot is very consistent, not sure what's going on there.


This is as mentioned with the 80 wire "hack" and a DMV340 SSD, I expect the 64 GB variant will perform about as well.

Humphreys
Jan 26, 2013

We conceived a way to use my mother as a porn mule



longview posted:

I'm just hoping it's not a PCIe related issue - I used to run a Intel gigabit PCI card back in the day after kicking out the Realtek crab.
I ordered a low profile "Intelģ PRO/1000 PT Dual Port" which is a bit overkill but I figure a decent PCIe dual gigabit card will come in handy in the future as well.


Strange situation now, the northbridge is the single warmest component in the computer.
Since the CPU is so underclocked & undervolted it only runs around 10 degrees above ambient with the fan at idle, while the northbridge is mostly running full speed. Frankly it would make more sense to put the CPU heat sink on there instead and use passive cooling for the CPU.

Added a 40 mm maglev fan to the heat sink, which brought it down to around 25 degrees above ambient (~50 degrees).


That dip near the start of the plot is very consistent, not sure what's going on there.


This is as mentioned with the 80 wire "hack" and a DMV340 SSD, I expect the 64 GB variant will perform about as well.

What is with people going to 80 wire IDE? I hear it on the OG XBox groups but never really tried it.

Computer viking
May 30, 2011
Now with less breakage.

After a quick wikipedia refresher on the details: the 80 pin cables pairs each wire in the 40-pin standard with a ground wire, to reduce crosstalk. Otherwise they work the same, except that the motherboard end of an 80-pin connector grounds pin 34 instead of connecting it to the drives.

As for why, the signal rate in the 66 and 133 MB/s modes would struggle over a 40-pin cable, so the controllers don't even try. When you plug a modern IDE SSD directly into the motherboard there is no cable to have crosstalk problems in at all, so the high speeds should be fine - but you still need to trick the controller into trying.

Buttcoin purse
Apr 24, 2014



I just run Windows NT 4 and 2000 Server under VirtualBox, I wonder how much the performance of my old DOS/Windows 3.1/MacOS 7/etc. clients is affected?

Computer viking
May 30, 2011
Now with less breakage.

It's probably a lot faster than running it on supported hardware - even if it's emulating period correct hardware, the emulation can pretend to be as fast as theoretically possible. Depending on the host machine, of course.

I don't expect any real overhead from the networking side, either; routing a gigabit stream in software and then dumping it into the real network card should barely register on a modern CPU?

Computer viking has a new favorite as of 12:36 on May 10, 2021

longview
Dec 25, 2006

heh.


Emulation overhead is quite good on modern hardware, but depends on the exact implementation and what your CPU can support.
Really it's irrelevant for legacy clients - the clients are so slow that the server is never going to be the limiting factor.

W2k on Hyper-V is annoying since it's a bit unstable, there's no integration services so have to run NTP to update the time when it's been suspended, and the network is limited to 100 Mbit/s (vs. the emulated 10 Gbit/s interface modern VMs use - makes a difference when copying data to/from it from a modern computer).
Emulating drive interface speeds is more accurate but may or may not be emulated, depends on what the intended use for the emulator is. My guess is Hyper-V for example doesn't care at all about that stuff and will run as fast as it can.

The reason I'm doing real hardware is 1) for fun and 2) to run legacy hardware like Token Ring networks.

Speaking of, doing some more digging it looks like it's not possible to just hard-wire up a ring with IBM token ring - though I think the IEEE spec would allow it.
There's a link-join procedure I found here https://www.opennet.ru/docs/FAQ/network/LANs/token-ring-faq.html which is also noted in some other sources. Can't seem to find the actual IEEE spec for free anywhere though.
Basically a MAU is required to be there for this sequence to work:
1) Station (computer) does a data loopback test via a relay in the MAU to check the wiring quality and self-test
2) Station applies a 5 V current limited bias signal through each wire in the TX pair, return through the RX pair (appears to be a common return)
3) MAU detects this and allows the station to join the ring by actuating some more relays
There is a range of allowed resistance that the station can measure in the 5 V test, and it tests the positive and negative wire separately (by separately measuring the current).

Though it looks like you could make a single token ring adapter "link up" with itself by wiring TX1 to RX1 and 2 to 2 with 2.2 kOhm resistors and maybe some AC bypass caps across them. This would basically look like it's seeing the MAU (resistors) and will be able to loop back. When joined it will still be in loopback, which looks identical to joining an empty ring.

So it does look like I need a MAU, but I think I can build one relatively easily with a pile of relays and optocouplers, it will be a powered one but that should be fine for my use case.

If my prototype works I'll spin up a 4-port+RI/RO PCB design, already got a schematic down for that based on how I think it should work.
Not sure about how the RI/RO automatic failover functionality was implemented in commercial MAUs, so that might end up incompatible with old hardware but shouldn't be a real issue.

Adbot
ADBOT LOVES YOU

Computer viking
May 30, 2011
Now with less breakage.

I have to say, a token ring MAU was not really one I expected to see in the list of "I'll just whip up a modern replacement for this ancient component" projects.Sounds entirely doable, just ... even more niche than most of these things.

  • 1
  • 2
  • 3
  • 4
  • 5
  • Post
  • Reply