Stack of 486's

[PhotoLenny]

One of my friends, a computer trainer, has 8 old 486 cpu's (whole box)with the old coaxial ethernet installed are these good for any of the projects, especially SETI? We are also toying with the idea of trying to use linux in a Beowulf type of arrangement - does this have any merit? Will SETI on Linux distribute over multiple cpu's?

 

[LeBlatt]

You bet they will be of ANY use for the Team !
For the OS, I guess Linux is a candidate : low cpu needs, and DOS client is not network enabled.
I think Seti specialists will confirm that the S@H client is not SMP capable.

The only bad side of 486/lowend cpus is they low KW/h vs crackpower ratio and storage space they need.

 

[PeterN]

I maybe wrong (don't know much about Seti) but I think putting 486's on Seti is not much good, since it would take to long to verify a WU. Also there's the issue of memory. Seti does need a fair amount of that to run smoothly.

RC5, OGR and Gamma are less memory-intensive.

Hopefully some Seti-guy can shed a light on this.

Regards, Peter

 

[blade47]

Yeah, I'm sorry to say they wouldn't be much use to us on seti. It just takes to much power and memory bandwith. They would do RC5 or OGR, sure they'll still be slow but much better than they'd do for seti. If it were me I'd put them on RC5.

 

[Michael]

SETi does do SMp ok as long as you run two instances. The client itself is not SMP aware (RC5 is, it automatically starts up two crunchers).

Michael

 

[LeBlatt]

Yeah, forgot to state that. Blade_47 and PeterN are right on seti too hooked on memory xfers.
I read the new client will be more memory independent.
But rc5 is still the best candidate as each machine will be able tu turn in a few WUs a day.

 

[Ken g6]

Actually, from my Opportunity Cost table, you can see that 486s are relatively better at SETI than at RC5.

Okay, maybe you can't, since the tables aren't labeled well. The second table is the appropriate one. Lower processors are better at SETI; higher ones are best on RC5. Since the 486 runs close to the same clock speed as its memory, it does fairly well. But that will probably change with V3.0...

 

[blade47]

Ken_g6 I don't doubt your tables but I have no idea what they're about. Just a little over my head there. I didn't see anything on seti in the charts. Could you explain all of that to me please. I'm always up for a little tutoring session if you don't mind.

Also seti resends the wu's after a week if they're not returned. Could a 486 return one in less than a week? It really wouldn't be benefiting the project if the results are already resent.

BTW I'd love it if the 486's could really benefit us some.

 

[PhotoLenny]

Since I'm much more an alienhead over a crackhead (not that crack'sall bad), if the 486's could do a seti wu in less than a week, it would be ok to run seti on them, 8 more units a week would be a good improvement and an interesting start to my herd (OT - we need a new name for a group of seti-crunching machines, herd is too d-net). I have also found out thru other means that the "beowulf attacks seti" won't work. Would a 486 plow thru a wu in less than a week?

 

[blade47]

You'd get credit for them even if it took longer than a week. It's just that the guys at berkley will have resent it to do over. Which is no big deal really. I saw a time on a 486 but don't remember where it was at. I'll look around for it.

BTW PhotoLenny & Michael two dedicated seti guys like yourselves should have a link in your sig to our seti team. That way everytime you post we'll get a little plug and if someone does a search for seti in the forum your sigs with links to the team will pop up. Just a little extra exposure for the team.

Edit: What speed are the 486's and how much ram?

 

[blade47]

Here's a link to some 486 times Seti Times. Some of their numbers look way off to me though and they don't list which client it was with. I'll look some more when I get time because I know I saw something somewhere else. From there numbers though it appears the 486's would add about 8 wu's a week. Not to shabby.

 

[sciencewhiz]

A bump for ken to explain the opportunity costs, I think I understand but want to hear from the master

 

[Ken g6]

"Opportunity Cost is how much you have to give up of something to get something else." That was lesson #1 (or so) in my economics class. In this case, it's how many FLOPs you give up to get a KKey in one second (I think.)

Unfortunately, I do doubt my tables. :Q SETI is a lot more complex than any other project, because it depends more on memory speed than CPU speed. I got these numbers from SetiSpy, but actually, they don't take into account CPU speed.

Mika has asked me to update the OC pages, so I think I'll substitute OGR for SETI as a compared project. At least until SETI V3.0 comes out.

 

[blade47]

Ken_g6 I thought that was what you were talking about. But didn't see how you could calculate the efficiency for seti so easily, since it's so memory and chipset dependant. Good to see I wasn't the only one wondering that. Now I don't feel quit as stupid.

It does appears that v3.0 of seti will totally change what makes a good seti system. I've been pondering starting my own little seti crack rack but am waiting on the new client and some more funds. If I do build one it'll be at a very slow pace.