[halimedia]

Originally Posted by Todd Madson

I wonder how you would set it up so that 2 or 4 processors would divide the workflow. Would it create a measureable speedup?

The worker would have to be multi-threaded insead of single-threaded. Individual threads of the computation would then automatically be distributed to all processing cores by the kernel. There are a number of reasons why the SETI-workers have never been multi-threaded:

- Multi-threaded coding is far more labor intensive
- The end result (i.e. RAC) of one machine wouldn't be any better (rather a touch worse, most likely)
- You couldn't allocate just a fraction of the cores to BOINC, which is an important feature of BOINC in high-availability server environments.

HTH,

Ron

[Knightrider]

I came across this web site, which deals with Parallel computing and clustering.

It occurred to me that this type of system (Clustering) may allow BOINC to be able to process very large work units by using two or more remote computers to process different parts of the same wu and assemble the results either on one machine acting as a surrogate server and it reports, or onto a server proper collecting and unifying the separate results.

It may also be a way of improving the work rate for multiple cpu machines by treating each wu in a cluster style and portioning it out to the available cpu's.

it was just a thought.

K.

[halimedia]

Since the BOINC infrastructure is a distributed computing platform by itself, the whole SETI-crunching community already represents an enormous distributed cluster (what an oxymoron!). Additional clustering solutions would therefore be of very limited advantage...

[BTBlomberg]

Originally Posted by alexkan

v5 prints its identifying information to stderr every time it starts. If you pause computation (and BOINC is set up to not keep suspended applications in memory), the application will print its information again once computation resumes. Crunch3r's app used to do this as well, I think, but the problem is exacerbated by the fact that your G4 500 takes a long time to crunch in the first place and winds up being interrupted quite a few times.

I got you. Since this is a PowerBook and I move it around during the day (unplug, move, plugin) that makes total sense. It was not problem, just an oddity I was pointing out. All other machines are desktops so unless a restart is needed mid WU I guess most people would not see this. Also, as you say a faster Laptop with shorter times would likely have fewer interruptions per WU. Always learn something new. Adds to teh point that I have known for a while that I need a new laptop.

Thanks!

[beadman]

Originally Posted by BTBlomberg

I got you. Since this is a PowerBook and I move it around during the day (unplug, move, plugin) that makes total sense. It was not problem, just an oddity I was pointing out. All other machines are desktops so unless a restart is needed mid WU I guess most people would not see this. Also, as you say a faster Laptop with shorter times would likely have fewer interruptions per WU. Always learn something new. Adds to teh point that I have known for a while that I need a new laptop.

Thanks!

My iBook is only a year and a half old, but I'm already trying to figure out how to justify to my wife my need for a new MacBook Core Duo. I'll probably have to wait until next spring when I get my tax refund. Maybe I can offer to send the iBook to our son and daughter-in-law, since they're stuck with an iMac G3 333Mhz and OS 9.

beadman

[Gecko_r7]

Originally Posted by beadman

My iBook is only a year and a half old, but I'm already trying to figure out how to justify to my wife my need for a new MacBook Core Duo. I'll probably have to wait until next spring when I get my tax refund. Maybe I can offer to send the iBook to our son and daughter-in-law, since they're stuck with an iMac G3 333Mhz and OS 9.

beadman

Doesn't she realize that the laptop is really for her?
Ilife 06' and other goodies that make pictures and vid clips so fun.

My wife can't wait for hers...now.
And Seti will be happily crunching away in the background as a service.........

Worked for me.
Good luck!

[BTBlomberg]

Latest v5 result on the old PB G4 500. It is in the 30 point range and I have some others to compare it to from v4.

THis finished in 32,009.92 sec. (8:53 hours) and claimed 32.34 but got 21.25 because somehow some PC users chrunched it on "core_client_version-4.13" for 9.90 and core_client_version-4.45 for 21.25. the other user was on 5.13 so they got 32.33 which is what ti should have been. Looks to me that this was a Old WU for these old clients that are horribly calibrated (4.xx) and becasue of it they bring down the whole quarum. Sux. I thought they no longer were getting WUs. Maybe they were reissues of failed WUs.

Anyway, I have this to compare to 3 v4 WUs:
33,763.41 sec. (9:23 hours) claimed 29.10 Received credits 29.09
35,207.26 sec. (9:47 hours) claimed 29.95 Received credits 29.95
33,553.58 sec. (9:19 hours) claimed 29.14 Received credits 29.13

Average:
34,174.75 sec. (9:30 hours) Received credits 29.39

So there was some improvement there, about 40 mins. Not sure if they old Seti WUs, but the size and credits are right. I really thought old cleints could not used anymore though.

[beadman]

BTBlomberg:
Here's a comparison for you. My iBook 1.33 GHz with v.5, last six valid results

Secs Claimed Granted
8429 12.94 12.94
17831 32.4 32.4
17911 30.68 30.68
17984 30.64 30.64
20242 34.53 34.53
22396 48.36 61.53

The average of this is 554 seconds/credit. It looks to me like the principal difference between yours and mine is clock speed, which implies to me that we're both about where we should be. As an aside, I checked this afternoon on my einstein output for this same computer and using Bernd's 4.56 client and got an average of 396 seconds/credit. Inverting these numbers, that's 40% greater credit per cpu second crunching einstein.... I switched my iBook from 75% SETI/25% Einstein to 80% Einstein/20%SETI.

beadman

[fran]

I had the same problem with this one, and this was with the standard (5.4.9) client and worker on my windows machine.

http://setiathome.berkeley.edu/worku...?wuid=80775341
My machine 57561 seconds, claimed 61.42, granted, 47.80
One intel machine with "core_client_version>4.45<" claimed 47.80
One intel machine with "core_client_version>5.2.2<" claimed 39.25

Compare it to this one:
http://setiathome.berkeley.edu/worku...?wuid=80870239
58010 seconds, claimed 60.89, granted 60.89 (which is normal for this machine)
The other two machines (windows and power mac) both claimed and granted the same amount, using 5.2.13

Erik

[mac1896]

Originally Posted by Todd Madson

But it typically hasn't worked like that - in the present case of my machine, both
processors work on a separate work unit at the same time.

On a quad, it works on four at the same time.

I wonder how you would set it up so that 2 or 4 processors would divide the
workflow. Would it create a measureable speedup?

I don't know how it would work either, but I DO notice on my DA 1.8GHz duallie, that I CAN make BOTH PROCESSORS work on only ONE WU.

The above statement depends on the accuracy of Activity Monitor.

I have never run the WU all the way through to see if it will validate or not.

Strange thing is, that the processors only run at about 50% capacity each, so that there is no gain in time on the given WU.

I'm wondering if this is something that can be further exploited (not like exploited in a cheating way).

[BTBlomberg]

What's the Difference. One WU done in a little over half the time (due to managment overhead) or two WUs done at regular time. Basically in the same time you get two WUs finished. You may report a WU back faster, but it still takes time on two CPUs so it's the same or greater time. I say greater becasue when you do something like that one prosessor still needs to manage the other one so that should slow down the overall process. One at a time on two is better. 1 CPU will allways complete a little faster in real time because one processor will get tied up with OS activities more so they will diverge in processing position, but each will still produce about the same # of WUs with the first getting hit with any other activity on the machine.

Overall two CPUs for Two WUs is really better.

[liebsmaschine]

Originally Posted by Gecko_r7

And Seti will be happily crunching away in the background as a service.........!

"Oh, honey, you don't need to take the iBook with you today...no, don't put it to sleep...just leave it on the desk open...in fact, just leave it open on the desk every day..."

When I first got back into running SETI/BOINC three or so months ago, I vowed that I would NOT LET it determine how I used my computer and whether I took my computer (and let it sleep all day) in my bag.

In the last month, it's been running 24/7 on my nightstand, and I'm hesitant to run iTunes, leave apps open, or even to activate Dashboard.

Oh, well, it's all about the science, isn't it?

[Gecko_r7]

Originally Posted by jackal

In the last month, it's been running 24/7 on my nightstand, and I'm hesitant to run iTunes, leave apps open, or even to activate Dashboard.

Jackal, I almost fell out of my chair laughing so hard!!!
The G4's been 24/7 on Seti since 03. I never leave aps open.
Since installing on my laptop, poor IBM has been running 24/7 for 2 years (with shutdown/reboots every 3-5 days since Windows gets more unstable the longer it's up).
I don't take it with me on bus. trips unless I KNOW I'll have to use it, and it irks me when I have to actually miss 4-6 hours on a plane flight. (battery is only good for 2 at full speed).
Pretty darn ridiculous. We may have to start a support group and meet at the YMCA every Wed. night.

On topic, v5 has been flawless w/ 30 WUs crunched. No invalid WUs, no errors specific to the G4. Smooth sailing!

[alexkan]

As the author of the optimized client, let me definitively put a rest to the questions about processing work units in parallel--the current client doesn't do it, and it's more efficient at this point for each worker to be independent. Think of it this way: the cores on your machines are still dividing up the SETI work among themselves, just with much larger granularity, and without any need for the cores to synchronize with (read: wait for) each other.

If it looks like your cores are working on the same work unit, they're probably duplicating a lot of work. You don't want this to happen.

Speaking of specific to the G4, I've posted the G4-specific compile athttp://tbp.berkeley.edu/~alexkan/set...-ppc-v5-g4.zip. The difference in speed only amounts to about 5%. It's less than I was hoping, but G4s need all the speed that they can get. Don't forget to generate your wisdom if you want an extra edge!

[wrhowell]

I see the v5 worker has 3 threads. Only 1 for the computational work. I assume the other 2 are for graphics. I've seen that there is a -nographics option for the worker, but how do you get the CLI boinc client (5.4.9) to start each worker with the -nographics option, and does it have any impact on run time compared to not specifying any options?

[fran]

Originally Posted by alexkan

Speaking of specific to the G4, I've posted the G4-specific compile athttp://tbp.berkeley.edu/~alexkan/set...-ppc-v5-g4.zip. The difference in speed only amounts to about 5%. It's less than I was hoping, but G4s need all the speed that they can get. Don't forget to generate your wisdom if you want an extra edge!

THANKS!! Already installed on both my G4s. What's this about "generate your wisdom"? I've seen this file referenced in several threads, but have no idea what it all means. What's a wisdom file, and how do I generate it?

Thanks again,
Erik

[liebsmaschine]

Originally Posted by alexkan

Speaking of specific to the G4, I've posted the G4-specific compile athttp://tbp.berkeley.edu/~alexkan/set...-ppc-v5-g4.zip. The difference in speed only amounts to about 5%. It's less than I was hoping, but G4s need all the speed that they can get. Don't forget to generate your wisdom if you want an extra edge!

Is that 5% better than the G5 v5 compile or 5% better than the G4 v4 compile? (Say that 5 times fast...)

[alexkan]

Originally Posted by jackal

Is that 5% better than the G5 v5 compile or 5% better than the G4 v4 compile? (Say that 5 times fast...)

It's 5% faster than the G5 v5 compile. v5 creams v4 on G4 regardless of which compile you're using.

[liebsmaschine]

Originally Posted by zombie67

THANKS!! Already installed on both my G4s. What's this about "generate your wisdom"? I've seen this file referenced in several threads, but have no idea what it all means. What's a wisdom file, and how do I generate it?

Thanks again,
Erik

Try <a href="http://forums.macnn.com/showthread.php?p=2994256&highlight=wisdom#post2994 256">this post</a> for alex's original instructions and the 300,000 posts that follow for help with issues when running it.

Basically, you'll download the zip file linked in that post (oh, all right, <a href="http://tbp.berkeley.edu/~alexkan/seti/fft_test3.zip">here's the link</a>), unzip it, run it with ./fft_test3 (or whatever the unzipped filename is--I forget) in Terminal, and when it's done, look for the wisdom.sah file (it should be in whatever directory you ran fft_test3 in) and copy that to /Library/Application Support/BOINC Data/projects/setiathome.berkeley.edu/. You may end up replacing a wisdom.sah file that's already there, but that's OK (and better).

HTH.