With the imminent demise of SETI Classic and the concern about what old machines can do I'd like to report that my 486 runs BOINC just fine. My 486 (Dinosaur) is able to run a SETI Classic work unit to a successful completion in 11 to 13 days. Running a SETI BOINC WU is very similar.

Now this isn't just any 486. It was originally a 486-DX66, but it's been upgraded to an AMD 586-133 (super 486 running faster with a Pentium rating of 58). Dinosaur has 64MB of RAM, but it reports 96MB. I think its reporting of 96 MB is important to get around the 64MB restriction for SETI. This is some weird motherboard thing, but I'm not complaining.

I have found that I can run SETI and Predictor on Dinosaur. Predictor WUs take about 4 days (one hour on my 3.4 GHz P4). Of course I do each project one at a time.

Before someone asks why, it's because it's there! I bought the darn thing in 1993 and I wanted it to do something useful. It may be trash, but it can really do some useful work.

A question: Dinosaur is currently running Win98 SE. What's the latest LINUX distribution that supports such an ancient machine? I'm running SuSE 10.0 on my "real" machines and there is no way this will run on Dinosaur. I've got SuSE 5.3 (kernel 2.0.35) in a box, but this seems really old.

Hi!
I was very happy to see your posting, and I'm not laughing at all!
Some time ago I posted a challenge to see how small hardware you could use for SETI crunching.
I got so excited over the feedback that I tried to revive my old 486, 80MHz, 64MB RAM, by putting a Linux distro on it.

Until now without any luck, the trouble is that the standard BOINC client times out while doing CPU benchmarks. So I've tried to compile a 486-optimized BOINC client(!), also with no luck: If I compile it on my Celeron system, my 486 can't execute it, it complains about a missing libstd6 or something, and if I compile it on my 486 system, I get an internal compiler error...

The distros I've tried have been
DSL 1.4, which is Debian based, and
The latest Debian Stable

I'm considering trying the Slackware distro, since it's also announced to support 486 systems.

I think we should keep each other informed about our progress - I'll post my experiences here in this forum.

---

"Are you suggesting coconuts migrate?"

11-13 days for ONE workunit? I real waste of electricity!!

---

11-13 days for ONE workunit? I real waste of electricity!!

No.

---

"Are you suggesting coconuts migrate?"

11-13 days for ONE workunit? I real waste of electricity!!

what would you rather have it do?

The latest [url=http://www.debian.org/]Debian[url] Stable

Martin, might wanna edit your post while you still can to stick in that missing /
;)

Hi!
I was very happy to see your posting, and I'm not laughing at all!

<snip>

I think we should keep each other informed about our progress - I'll post my experiences here in this forum.

@Martin I certainly will keep you advised. I'm a real newbie to Linux so I haven't tried compiling anything yet. I'm a retired engineer, but I'm a hardware guy! This Linux stuff is new to me. I'll take the plunge soon. That's the whole point, I want to learn.

11-13 days for ONE work unit? I real waste of electricity!!

Electricity is relative. I recently (last 9 months) got two more P4 computers. My electric bill has doubled since then. Dinosaur is a pimple on the elephant. It's a harmless endevour :-)

Until now without any luck, the trouble is that the standard BOINC client times out while doing CPU benchmarks. So I've tried to compile a 486-optimized BOINC client(!), also with no luck: If I compile it on my Celeron system, my 486 can't execute it, it complains about a missing libstd6 or something, and if I compile it on my 486 system, I get an internal compiler error...

Martin,

I'd be quite interested if my generic BOINC client (you can get it here) runs on your 486. I didn't use any speciffic compiler switch to enable only Pentium or whatever instructions, so in principle it should run even on 386. But one never knows. And it is statically linked, so it shouldn't rely on shared libraries of any particular version.

---

Metod ...

Martin,

I'd be quite interested if my generic BOINC client (you can get it here) runs on your 486. I didn't use any speciffic compiler switch to enable only Pentium or whatever instructions, so in principle it should run even on 386. But one never knows. And it is statically linked, so it shouldn't rely on shared libraries of any particular version.

Great! Thanks!
I'll try it out as soon as possible.

If you're interested, I'll post some benchmark results like the ones you provide on your site. Provided I have any success, of course...

---

"Are you suggesting coconuts migrate?"

If you're interested, I'll post some benchmark results like the ones you provide on your site. Provided I have any success, of course...

That would be great. Especially if you would manage to run official BOINC CC (in benchmark-only mode would do) to have a couple of numbers to compare ...

---

Metod ...

That would be great. Especially if you would manage to run official BOINC CC (in benchmark-only mode would do) to have a couple of numbers to compare ...

Well, like I mentioned earlier, the official BOINC client timed out when I tried to run the CPU benchmarks.

So, at least for now, the benchmarks are "timed out" or whatever you could call it.

That being said, I think my system was running some unnecessary daemons while doing this benchmark, and I also installed it with the ext3 filesystem type, which I've read somewhere, is a bad idea on older/slower systems. But I don't know if it's only disk I/O, CPU usage, or both that are affected by that choice...

So I hope to install a minimal crunching-only Linux system in the near future, running with ext2 filesystem and no unnecessary daemons. Maybe then I will be able to complete a CPU benchmark. After all, it must be within the reach of my 486, since Purple Rabbits 586 system obviously must have pulled a benchmark through(?)

---

"Are you suggesting coconuts migrate?"

The latest [url=http://www.debian.org/]Debian[url] Stable

Martin, might wanna edit your post while you still can to stick in that missing /
;)

Thanks and done!

---

"Are you suggesting coconuts migrate?"

11-13 days for ONE workunit? I real waste of electricity!!

No.

Let's say a new Pentium 4 consumes 200 watts of power (without monitor) and finishes 2 workunits in 2 hours (with HT on). It cost 200 watthours per workunit.

And a 486 running at 20 watts (rough estimate). It takes 11 days to complete a workunit. That's 11x24x20=5280 watthours per workunit.

So why is it not a waste of power to use a 486 to crunch?

---

---

11-13 days for ONE workunit? I real waste of electricity!!

No.

Let's say a new Pentium 4 consumes 200 watts of power (without monitor) and finishes 2 workunits in 2 hours (with HT on). It cost 200 watthours per workunit.

And a 486 running at 20 watts (rough estimate). It takes 11 days to complete a workunit. That's 11x24x20=5280 watthours per workunit.

So why is it not a waste of power to use a 486 to crunch?

Because the 486 might be on anyway doing other things - like being a firewall or router than doesn't need a whole lot of processing power. So if it's running 24/7 anyway, why not crunch on it too?

Ned

---

*** My Guide to Compiling Optimised BOINC and SETI Clients ***
*** Download Optimised BOINC and SETI Clients for Linux Here ***

So why is it not a waste of power to use a 486 to crunch?

Because we do it for fun, not for speed, efficiency or glory.

You could also say that lighting a cristmas tree is waste of electricity. And that's perfectly OK, but if I do it, and I'm willing to pay for that electricity, and it bears a value for me to look at that lighted christmas tree, then I will not feel it's a waste of electricity - and then it isn't, at least for me.

Same argument goes with
- running a marathon (take a bus!)
- drinking beer (snaps has a higher alcohol-percentage!)
- playing World of Warcraft (waste of time and electricity, you should sweep your floor and have your computer do something useful, like crunching SETI WU's!)

Luckily there is lots of room to disagree here - I would never try to convince you to crunch on a 486 if you don't see the fun in trying to do it, because then it actually would be a waste. Therefore I hope you will leave room for us to crunch on our 486's as long as we pay our own electricity bill (which I actually do).

---

"Are you suggesting coconuts migrate?"

Ok, then let's assume he already owns his 486, but not a P4 w/ht. he must then purchase a P4. We must figure the cost of that into the $/result calculation.

Let's assume a 5 year lifespan for a computer and set the price at $1000 USD for this fictional machine/example. Let's also assume a one hour/result completion average. OK there's 24 hours in a day, 365 days in a year (excluding leap year), so 24 X 365=8760 results/year X 5 year lifespan=43800 Results/puter lifespan. divide $1000 by 43800 results = 2.28311 cents/result.

I know my P3-500 drew 104 watts (100%seti load, 80 watts no load)w/o monitor, so my guess for the 486 would be closer to 50/60 watts. let's use 60. 60 watts x (12 days x 24 hours)288 hours=17280 watts or 17.28 KWH X 10 cent/KWH = $1.728 dollars/result.

With the 200 watt system we get. 200 watts is .2 KWH times $.10/KWH or 2 cents/wu.

Summary:

486, no additional equipment outlay, $1.73/result
P4, 2.283 cents/result (equipment cost) + 2 cents/result electricity = 4.283 cents/result.

Ok, then let's assume he already owns his 486, but not a P4 w/ht. he must then purchase a P4. We must figure the cost of that into the $/result calculation.

Let's assume a 5 year lifespan for a computer and set the price at $1000 USD for this fictional machine/example. Let's also assume a one hour/result completion average. OK there's 24 hours in a day, 365 days in a year (excluding leap year), so 24 X 365=8760 results/year X 5 year lifespan=43800 Results/puter lifespan. divide $1000 by 43800 results = 2.28311 cents/result.

I know my P3-500 drew 104 watts (100%seti load, 80 watts no load)w/o monitor, so my guess for the 486 would be closer to 50/60 watts. let's use 60. 60 watts x (12 days x 24 hours)288 hours=17280 watts or 17.28 KWH X 10 cent/KWH = $1.728 dollars/result.

With the 200 watt system we get. 200 watts is .2 KWH times $.10/KWH or 2 cents/wu.

Summary:

486, no additional equipment outlay, $1.73/result
P4, 2.283 cents/result (equipment cost) + 2 cents/result electricity = 4.283 cents/result.

But Tony, the currency here is not "results" but "credits". So you'll have to make a credits based analysis. And in order to do that, you'll have to have a large population of results crunched by 486 based systems to derive accurate mean and standard deviation. So in order to provide you with that, we now must crunch a lot of WU's on 486's in order to provide you with a reliable base of data for your calculations.

So now we're not even doing it just for fun anymore, now it's actual economic science.

Thanks, Tony ;-)

---

"Are you suggesting coconuts migrate?"

So in order to provide you with that, we now must crunch a lot of WU's on 486's in order to provide you with a reliable base of data for your calculations.

Seems right to me. I'll bookmark this thread and we'll start it back up in a year or two once you get enough results.

LOL