You are not logged in. Login Now
 0-24   25-44         
 
Author Message
rcurl
SETI@home on Grex? Mark Unseen   Jun 25 15:36 UTC 1999

SETI@home is being discussed in the science conf. It is a distributed
Search for Extraterrestrial Intelligence. I would like to propose that
Grex run it, though in a manner that would not significantly slow
general usage. There are now some 680,000 machines worldwide processing
radio signal data from the Arecibo telescope, contributing so far some
17,000 *years* of CPU time. SETI needs more. If this is feasible, it
would be a small contribution to SETI, and something interesting for
users to follow - sort of an ET lottery, where just maybe Grex would
be the server to detect a significant ET signal.

The technical details of doing so are beyond me, but SETI@home offers
some 50 different versions of the software for unix and other systems.
See the SETI item in science for more information, and connect to
http://setiathome.ssl.berkeley.edu/ for the real word.
44 responses total.
jshafer
response 1 of 44: Mark Unseen   Jun 25 19:48 UTC 1999

...crunching the same blocks over and over again...

Hm, I may give SETI@home a shot once they've got some of the bugs out...
kaplan
response 2 of 44: Mark Unseen   Jun 25 20:19 UTC 1999

My understanding is that SETI@home is like a screen saver in that it 
sleeps while you use the machine and only wakes up and starts processing 
when the machine has nothing else to do.  There is never an hour of the 
day or night when grex isn't busy doing something, so the SETI@home 
program would never get anything done here.
scg
response 3 of 44: Mark Unseen   Jun 25 20:58 UTC 1999

People have tried to run SETI @home here, thus being a pain for the staff to
deal with.

SETI@home claims that it doesn't slow down the computers that it's running
on, since it runs at a low priority, so other tasks take precidence over it.
However, it is a resource hog from the memory standpoint.  It's not a good
thing to have running on Grex.  I wouldn't run it on any computer I wanted
to be able to get work done on either.
drew
response 4 of 44: Mark Unseen   Jun 26 01:05 UTC 1999

They could just wait a few years, and then all the processing power they need
would be readily available in a handful of machines.
i
response 5 of 44: Mark Unseen   Jun 26 01:05 UTC 1999

With plenty of memory, an OS that handles low priority jobs well, and
typical single-user-system usage patterns (lots of CPU idle time),
SETI@home can do its number crunching without having a noticable
effect on system responsiveness.  I know people who run it in the
background on their systems and can't tell the difference in speed. 

However, grex certainly does not fit that description.  Our CPUs' time
is booked solid with "users are sitting there waiting for an immediate
response" tasks virtually all the time.  Running SETI@grex is just a
way to make this system slower for all the human user on it.
i
response 6 of 44: Mark Unseen   Jun 26 01:07 UTC 1999

s/human user/human users/
scott
response 7 of 44: Mark Unseen   Jun 26 03:24 UTC 1999

Besides which, Grex is really not any more powerful than a typical recent
Windows PC.  
mdw
response 8 of 44: Mark Unseen   Jun 26 05:03 UTC 1999

From a CPU crunching standpoint, grex's sun-4/670 is *less* powerful.
The Sun-4/670 is good for multi-tasking, because it has good I/O
hardware, MMU, etc.  And, at the time the sun-4/6xx series was
introduced by sun (ca. 1991) it was pretty much as fast or faster than
anything then available as a PC.  But times change.  The 68040's and
486's of that era have been replaced by the pentium, 601, pentium II,
and 604.  Any of these chips is significantly faster than grex.  A
decently equipped pentium II or 604e machine (even running macintosh os
or ms windows) will blow the doors off of grex.  I believe grex comes
out roughly equal to a 486 for a single-thread CPU task.  That means
something based on the pentium II or 604e is probably easily 10 times
faster than grex - for a single-thread CPU bound task.

The reason we use a sun-4/670 for grex is because they're Real Cheap,
and run a fairly secure and stable OS, oh and the hardware has a nice
design.  There are tons of pentiums and 601's out there, but they're not
being pushed off the loading dock out back yet to make room for newer
stuff.  And, before people *do* start giving those machines away,
they'll be getting rid of their SparcServer 20's and Ultra-5's.  Now,
the Ultra-5 is a *very* nice machine....

Think of grex as a place where you get together with friends.  Think of
your new desktop machine as a much better place to do CPU number
crunching.
janc
response 9 of 44: Mark Unseen   Jun 26 16:01 UTC 1999

When I last looked at SETI@home, there were two basic versions:

  - PC version - runs as a screen saver, only starting up after the
    computer has been idle for a few minutes.  Otherwise it doesn't
    run at all.

  - Unix version - runs as a process with a low CPU priority.  Only
    runs when no other process wants to use the CPU.

I installed the Unix version on my Linux box, and ran one block.  My
machine is a 133MHz Pentium with 32 Meg of memory, probably a bit faster
than Grex at a single task.  The block took just over a day to run. 
However during that day, my Linux system was noticably sluggish.  Let's
suppose I had SETI@home running while I was doing what I'm doing now -
typing text into a Netscape window.  Netscape isn't using much CPU - it
spends almost all of its time waiting for me to type the next character.
So since there is CPU available, SETI@home runs a lot.  But Netscape and
SETI@home are both huge memory hogs.  So whenever SETI@home starts
running, it starts grabbing memory, and Netscape starts getting swapped
out to disk.  If I then type again, then Netscape has to be loaded back
into memory before it can respond to my typed key.  Naturally things get
sluggish.  I've had some people tell me that their Unix boxes weren't
slowed much by SETI@home.  This is likely true, if they have lots of
more memory than I do.

For Grex, the situation would be similar.  Though our CPU is much busier
than most single-user systems, it is idle a significant amount of the
time.  This is a good thing.  It means that when you type a key in
something like "vi" or "gate", Grex can respond immediately.  The Sun
4/670 is faster at jumping between jobs than a Pentium is, but that
mainly helps with smaller processes.  Big memory hog processes like
SETI@home are more likely to be slow to switch.

So running SETI@home on Grex would significantly slow it down at what it
does best - juggling lots of tasks for lots of users.  At the same time,
since it doesn't run any single task very fast, it would contribute less
to the SETI@home project than if someone unleashed an idle 486 box on
the project.  It's just not a good deal for us.

The fact that Grex is a lousy place to run this thing shouldn't
necessarily discourage people from running it on their own machines.  If
you have a single user Unix box well-stocked with memory, it likely
won't be a problem.  If you have a Windows machine, the screen saver
version is almost certain not to slow anything down much.  If you've got
an old 486 that you aren't using for anything, bringing that back up to
run SETI@home would at least do more good than putting it on Grex.
cmcgee
response 10 of 44: Mark Unseen   Jun 27 16:27 UTC 1999

Is this something keesan could donate a Kiwanis computer to do?
scott
response 11 of 44: Mark Unseen   Jun 27 17:30 UTC 1999

It's the sort of thing that is most effective as a screen-saver on all of
those high-powered PCs out there.  A 486 from Kiwanis would be *much* better
used as the first computer of somebody with not a lot of money.

(and it's not like the world depends on Grex supplying a processor for running
this thing.  Sheesh.)
rcurl
response 12 of 44: Mark Unseen   Jun 28 05:25 UTC 1999

Most of the world, in fact, is totally uninterested in this.

This has been most informative, and I appreciate the explanations. My
son put seti@home on a server he manages for a company and he said
it ran several work units in a few hours, but he was anxious that it
might just cause a lag when his boss was looking...so he took it off.

One minor point re #9: you do not have to run it as a screensaver on
a PC. It works fine launching it as an application that runs along
with everything else. How well this works on the machines I have tried
it on is described in the science item.

Well...Grex may be a home to the world, but I guess not to ET.....
aruba
response 13 of 44: Mark Unseen   Jun 28 15:38 UTC 1999

Do you need to be connected to the net all the time to run seti@home?
rcurl
response 14 of 44: Mark Unseen   Jun 28 16:03 UTC 1999

No. However you can set it to either connect automatically when it
finishes a work unit (to deliver the results and download another
work unit), or to just alert you that it is done (when a dialog
allows you to choose to connect). 
aruba
response 15 of 44: Mark Unseen   Jun 28 16:26 UTC 1999

How long does it take to download a unit?
rcurl
response 16 of 44: Mark Unseen   Jun 28 16:34 UTC 1999

A work unit is about 300,000 bytes. 
steve
response 17 of 44: Mark Unseen   Jul 7 00:02 UTC 1999

   I think others have spoken quite well on the reasons why not to run
Seti here, as much as I like it myself.  Grex would be home to an ET
Rane, if we had the horsepower.  However I don't think a public access
system like Grex will ever have the ability to do so.
mwg
response 18 of 44: Mark Unseen   Jul 16 19:22 UTC 1999

I think the SETI@home people need to get distributed.net to tune up thier
clients. I have a Linux machine that manages to churn through several
hundred thousand RC5-64 keys per second without any noticable effect on
the rest of the system.
janc
response 19 of 44: Mark Unseen   Jul 17 13:17 UTC 1999

I think the seti@home program works on much bigger blocks of data than
the rc5 thing, and thus uses more memory, which I think is at the root
of the problem.
keesan
response 20 of 44: Mark Unseen   Jul 26 18:21 UTC 1999

Kiwanis is quite short of 486s, we are still mainly getting in older computers
such as an occasional 386, and PCJr, and TI, and TRS-80.  So far we have had
four working 486s donated in over a year.  
toking
response 21 of 44: Mark Unseen   Jul 27 16:49 UTC 1999

if I ever get back up to A2 I'll bring some 486 MB's, I'm not sure when
I"ll be back up there though
gull
response 22 of 44: Mark Unseen   Sep 29 04:42 UTC 1999

I'm running rc5des on my UNIX server (without a noticable slowdown.)
SETI@Home was way too memory-hungry when I tried it, and it mostly just
highlighted the slowness of my CPUs...on my K5-133 it took over a day to run
a work unit.  I think the culprit is the large work unit size, and the
relatively complex math needed -- RC5DES doesn't do any floating point math,
just integer.  RC5DES will probably take longer to complete, though; they've
been working over a year and have checked just 12% of the keyspace.
prp
response 23 of 44: Mark Unseen   Nov 28 05:07 UTC 1999

The SETI@home program for Grex (Sun OS) doesn't cause a noticable load.
It does crash though.  I sent them a message about it.  A reply may have
come while my id was reaped. 
other
response 24 of 44: Mark Unseen   Nov 28 06:03 UTC 1999

wrong.  it does cause a noticeable load.  SETI@home is made for use on
single-user, desktop machines, not 24hour, multi-user systems which don't have
any such thing as 'spare cpu cycles' to offer.  don't run it here.  period.
thank you.
 0-24   25-44         
Response Not Possible: You are Not Logged In
 

- Backtalk version 1.3.30 - Copyright 1996-2006, Jan Wolter and Steve Weiss