Faster credit: 4 copies of each WU initially being sent out now

Message boards : Number crunching : Faster credit: 4 copies of each WU initially being sent out now
Message board moderation

To post messages, you must log in.

Previous · 1 · 2 · 3 · 4 · 5 · Next

AuthorMessage
N/A
Volunteer tester

Send message
Joined: 18 May 01
Posts: 3718
Credit: 93,649
RAC: 0
Message 57988 - Posted: 28 Dec 2004, 19:06:54 UTC
Last modified: 28 Dec 2004, 19:07:43 UTC

How about just saying that x amount of time was used, y units were crunched, z is the Dhrystone and z' is the Whitestone. Based on that, one can get a better sense of what the CPU is capable of, and it doesn't require validation - It's specific to the host, and the user can get feedback as soon as the WU completes. Once validated, IMNSHO, credits should be (time needed to crunch×average Dhrystone and Whitestone)÷number of processors.

.o0(This coming from a guy who's ppc7450 has topped out at 25cr/WU for the last three days non-stop and uninterrupted when it's been much higher in past...)

We really do need to have three units validate - Let's say that we hit the lotto and find ET's signal. If a host finishes crunching and says "Here's ET", it's chance; two hosts, it's coincidence; three, conclusive.
ID: 57988 · Report as offensive
Profile Misfit
Volunteer tester
Avatar

Send message
Joined: 21 Jun 01
Posts: 21803
Credit: 2,814,996
RAC: 43
United States
Message 57994 - Posted: 28 Dec 2004, 19:24:56 UTC - in response to Message 57877.  

> You complain that 3 validations takes to long, so the Seti people set it to 2.
> Now, you're mad that 2 shortchanges you on credits. Can't anything make you
> happy? :)
>
I don't think people will be happy until they finally get their own claimed credit.
ID: 57994 · Report as offensive
Profile Roks

Send message
Joined: 20 Dec 02
Posts: 55
Credit: 137,776
RAC: 0
Slovenia
Message 58019 - Posted: 28 Dec 2004, 21:59:01 UTC - in response to Message 57994.  

> > You complain that 3 validations takes to long, so the Seti people set it
> to 2.
> > Now, you're mad that 2 shortchanges you on credits. Can't anything make
> you
> > happy? :)
> >
> I don't think people will be happy until they finally get their own claimed
> credit.
>

I agree. I would be happy if I got the credit my computers claimed. At least classic did that. You crunched WU in x amount of time and you got your 1 credit. Most fair, I think. Or we really should get the amount of credit that we actually do (but seti can still compare WUs between them).
<img src="http://www.boincstats.com/stats/banner.php?cpid=d2319b8f0ad14565556d0ba45b64e779">
ID: 58019 · Report as offensive
N/A
Volunteer tester

Send message
Joined: 18 May 01
Posts: 3718
Credit: 93,649
RAC: 0
Message 58022 - Posted: 28 Dec 2004, 22:07:08 UTC
Last modified: 28 Dec 2004, 22:17:52 UTC

How about letting the user know in the BOINC output what the credit claim is? All I see is this:
2004-12-28 06:41:22 [---] May run out of work in 0.25 days; requesting more
2004-12-28 06:41:22 [SETI@home] Requesting 20980 seconds of work
2004-12-28 06:41:22 [SETI@home] Sending request to scheduler: http://setiboinc.ssl.berkeley.edu/sah_cgi/cgi
2004-12-28 06:41:23 [SETI@home] Scheduler RPC to http://setiboinc.ssl.berkeley.edu/sah_cgi/cgi succeeded
2004-12-28 06:41:23 [SETI@home] Started download of 24mr04ab.8365.18049.836068.225
2004-12-28 06:41:26 [SETI@home] Finished download of 24mr04ab.8365.18049.836068.225
2004-12-28 06:41:26 [SETI@home] Throughput 112257 bytes/sec
2004-12-28 10:55:42 [SETI@home] Computation for result 23ap04ab.24858.8433.623588.147 finished
2004-12-28 10:55:42 [SETI@home] Starting result 24mr04ab.8365.18049.836068.225_1 using setiathome version 4.02
2004-12-28 10:55:42 [SETI@home] Started upload of 23ap04ab.24858.8433.623588.147_3_0
2004-12-28 10:55:43 [SETI@home] Finished upload of 23ap04ab.24858.8433.623588.147_3_0
2004-12-28 10:55:43 [SETI@home] Throughput 102612 bytes/sec
2004-12-28 15:37:29 [---] May run out of work in 0.25 days; requesting more
2004-12-28 15:37:29 [SETI@home] Requesting 20988 seconds of work
2004-12-28 15:37:29 [SETI@home] Sending request to scheduler: http://setiboinc.ssl.berkeley.edu/sah_cgi/cgi
2004-12-28 15:37:30 [SETI@home] Scheduler RPC to http://setiboinc.ssl.berkeley.edu/sah_cgi/cgi succeeded
2004-12-28 15:37:30 [SETI@home] Started download of 23ap04ab.24858.23105.298554.160
2004-12-28 15:37:33 [SETI@home] Finished download of 23ap04ab.24858.23105.298554.160
2004-12-28 15:37:33 [SETI@home] Throughput 106043 bytes/sec
Nowhere does it say how long I crunched, how many flops/instructions were used, how many credits I should expect, or how many other people are crunching/pending the same WU!

It doesn't have to be posted here, but I'd like a little bit mroe feedback from an app - Especially programs that I can't tell if it's running. Maybe every ten minutes we could get something that says "Hi! I'm BOINC, and I'm 50% through the workunit you downloaded and I should be done by dinner."

[ADDENDUM] If you've got 2 WUs that agree exactly with each other, then yes I doubt that a 3rd WU is needed - But if the two conflict, you need a third WU.

OK. I'm done.
ID: 58022 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 10189
Credit: 136,380,569
RAC: 87,458
Australia
Message 58029 - Posted: 28 Dec 2004, 22:29:05 UTC


I've never bothered paying any real attention to the Credit arguments till now, so one quick question-

Are Whetstones actually used for calculating Credit? If so then they will never having anything to do with actual reality; not by a long, long, long way.
If you have a look at the results on Tom's Hardware Guide's lastest CPU comparision here you will see just how meaningless the Whetstone is.
The top 23 CPUs are Intel. The Dhrystone is closer to reality with 10 AMD CPUs in the top ten & most of them near the top. When you look at actual programme performance then even more AMD CPUs are actually at the top of the table & often by an even more considerable margin than the Dhrystone indicates.

If only the Dhrystone is used to calculate credits, then they will be fairly close to what is actually happening. If Whetstones are also used in any way to calculate credit then the claimed credits for Intel CPUs could be as much as 50% higher than they should be when compared to an AMD CPU.
Grant
Darwin NT
ID: 58029 · Report as offensive
Profile Martin P.

Send message
Joined: 19 May 99
Posts: 294
Credit: 25,717,684
RAC: 2,261
Austria
Message 58043 - Posted: 28 Dec 2004, 23:17:21 UTC - in response to Message 57761.  
Last modified: 28 Dec 2004, 23:24:26 UTC

> The same rules that's always been used. ;)
>
> If wu validated with only 2 results passing validation, the lowest claimed.
> If wu validated with 3 or more results passing validation, remove highest
> & lowest and average the rest.
>
> Any results validated after wu was validated is given the same credit as the
> others, no re-calculation is done.
>

Very bad idea!!!

I get the following error on a regular basis:
2004-12-28 15:38:13 [SETI@home] Starting result 24mr04ab.8365.14962.286082.240_2 using setiathome version 4.02
2004-12-28 15:38:13 [SETI@home] Started upload of 24mr04ab.8365.12753.965922.207_0_0
2004-12-28 15:38:16 [SETI@home] Finished upload of 24mr04ab.8365.12753.965922.207_0_0
2004-12-28 15:38:16 [SETI@home] Throughput 49206 bytes/sec
2004-12-28 17:27:44 [SETI@home] Computation for result 23ap04ab.24858.19520.78404.77 finished
2004-12-28 17:27:44 [SETI@home] Starting result 24mr04ab.8365.16210.1040906.145_1 using setiathome version 4.02
2004-12-28 17:27:44 [SETI@home] Started upload of 23ap04ab.24858.19520.78404.77_0_0
2004-12-28 17:27:47 [SETI@home] Finished upload of 23ap04ab.24858.19520.78404.77_0_0
2004-12-28 17:27:47 [SETI@home] Throughput 61475 bytes/sec
2004-12-28 17:28:16 [SETI@home] Result 24mr04ab.8365.14962.286082.240_2 exited with zero status but no 'finished' file
2004-12-28 17:28:16 [SETI@home] If this happens repeatedly you may need to reset the project.
2004-12-28 17:28:16 [SETI@home] Restarting result 24mr04ab.8365.14962.286082.240_2 using setiathome version 4.02

In this case I lost 2 hours of work due to this error.

This happens most of the time when processor 1 uploads a workunit. Whenever CPU 1 uploads, CPU 0 restarts. Percentage done stays as it was before, but CPU time starts from 0 therefore CPU 0 always claims too little credit (given that claimed credit is a function of time and benchmark results).
e.g.:
http://setiweb.ssl.berkeley.edu/workunit.php?wuid=6617637
http://setiweb.ssl.berkeley.edu/workunit.php?wuid=6521069
http://setiweb.ssl.berkeley.edu/workunit.php?wuid=6065972

Please also see this thread for more detailed argumentation:
http://setiweb.ssl.berkeley.edu/forum_thread.php?id=6634

P.S.: I did reset the project several times already - no change.
MacOS X 10.3.7 running on G4/Dual 1GHz and G5/Dual 2.5 GHz. Claimed credit is much lower than on Windows machines anyway.


ID: 58043 · Report as offensive
Arm

Send message
Joined: 12 Sep 03
Posts: 308
Credit: 15,584,777
RAC: 0
Message 58048 - Posted: 28 Dec 2004, 23:29:17 UTC - in response to Message 57758.  

> "December 27, 2004
> We are experimenting with redundancy parameters. We will continue to send 3
> copies of each workunit, but will validate results and grant credit when 2
> (rather than 3) of them are returned. This should result in faster
> validation."
>
> This brings up a number of new questions. If credit is granted on only 2
> validated results, what are the new rules for awarding credit? With only two
> results, the high / low elimination - give the middle value won't work
> anymore. Will it be an average of the first two validated results? Inquiring
> minds....
>
I think one of the most noticeable consequences would be that the number of the results returned with "Computation error" will rise dramatically :(

ID: 58048 · Report as offensive
Profile [HWU] GHz & CO. - BOINC.Italy
Volunteer tester
Avatar

Send message
Joined: 1 Jul 02
Posts: 139
Credit: 1,466,611
RAC: 0
Italy
Message 58051 - Posted: 28 Dec 2004, 23:35:35 UTC - in response to Message 58048.  

New news!



December 28, 2004
More on redundancy parameters. After some discussion we have decided to send out 4 copies of each workunit and continue with a quorum of 3 for validation. Validation should still proceed at a good pace because a single result in error will not mean distributing an additional workunit copy (and waiting for it's result) in order to reach a quorum.


Good news!! But the last wu returned (the 4th) take the same credit of first 3 wu used for validation?
GHz
BOINC.Italy
ID: 58051 · Report as offensive
Ingleside
Volunteer developer

Send message
Joined: 4 Feb 03
Posts: 1546
Credit: 11,780,490
RAC: 7,182
Norway
Message 58055 - Posted: 28 Dec 2004, 23:38:53 UTC - in response to Message 58051.  

>
> Good news!! But the last wu returned (the 4th) take the same credit of first
> 3 wu used for validation?
>

Yes, as already answered in this thread. ;)
ID: 58055 · Report as offensive
N/A
Volunteer tester

Send message
Joined: 18 May 01
Posts: 3718
Credit: 93,649
RAC: 0
Message 58060 - Posted: 28 Dec 2004, 23:55:28 UTC - in response to Message 58043.  
Last modified: 28 Dec 2004, 23:57:12 UTC

Very bad idea!!!
I have to agree: At the very least two WUs must conincide to within 4&#x03c3; or 99% or whatever it is that we use to compare results. If they're not close enough, send out a third WU. And when it comes down to claimed credit, a 4th would definitely help.

I get the following error on a regular basis:
I lost 24 hours that way. Next version of BOINC should be smart enough to realize that if it has an error, the host should send back an "I can't complete" message and move on to another WU. As is, does BOINC have the capability of telling the server that it's abandoning a WU?

MacOS X 10.3.7 running on G4/Dual 1GHz and G5/Dual 2.5 GHz. Claimed credit is much lower than on Windows machines anyway.
I was going to raise that point, but I thought that I might have come across as whining. I'd love to crunch at full capacity, but I don't know ppc7400-class assembler, and I'm willing to bet that the source code doesn't take advantage of the G4's AltiVec unit or the G5's SIMD units.
<rant>
It's a total waste of the CPU's capabilities!
</rant>
Somehere I recall some host (It was a Darwin 8.0.0b) claimed to be 1200 Dhry-/8000 White- stones. That's gotta be wrong. If it isn't, I'd like to know what the hell he/she/it is claiming for credit.

For the record, I've never claimed more than 50cr on any workunit.
ID: 58060 · Report as offensive
STE\/E
Volunteer tester

Send message
Joined: 29 Mar 03
Posts: 1137
Credit: 5,334,063
RAC: 0
United States
Message 58067 - Posted: 29 Dec 2004, 0:10:44 UTC - in response to Message 58019.  
Last modified: 29 Dec 2004, 0:12:58 UTC

>I don't think people will be happy until they finally get their own claimed credit.
>
>
> I agree. I would be happy if I got the credit my computers claimed. At least
> classic did that. You crunched WU in x amount of time and you got your 1
> credit. Most fair, I think. Or we really should get the amount of credit that
> we actually do (but seti can still compare WUs between them).
===========

The problem with giving out the credit that is claimed is it will lead to more cheating. I've been into several accounts where the person was claiming over 200 credits yet taking no more time to crunch the WU than I was & I'm only claiming 35 credits for the same WU.

If they gave the credit claimed then I think the driving force behind a lot of people will be to figure out a way to claim astronomical amounts of credit to pad their total credit stats ... IMO
ID: 58067 · Report as offensive
N/A
Volunteer tester

Send message
Joined: 18 May 01
Posts: 3718
Credit: 93,649
RAC: 0
Message 58077 - Posted: 29 Dec 2004, 0:44:40 UTC - in response to Message 58067.  
Last modified: 29 Dec 2004, 0:46:27 UTC

Then separate the two. Let's say that the Dhry-/White- are actually worth something. Then you can compare CPUs, OSs, RAM, etc., and tell whether or not someone's lying. But the user should be told what a WU appears to be worth in credits as well as what the WU is worth according to everyone else.

Giving people the credit they want is no way to go - Give them what they deserve after verifying the WU, the CPU, etc.

I'm sure that a small DB of CPUs and OSs wouldn't be too tough for Berkley to handle. Whenever someone comes in with a claim, find out how deviant the claim is to the stats of similar CPUs.
ID: 58077 · Report as offensive
SURVEYOR
Volunteer tester

Send message
Joined: 19 Oct 02
Posts: 375
Credit: 608,422
RAC: 0
United States
Message 58154 - Posted: 29 Dec 2004, 6:53:09 UTC

News - Technical News - Server Status

December 28, 2004
More on redundancy parameters. After some discussion we have decided to send out 4 copies of each workunit and continue with a quorum of 3 for validation. Validation should still proceed at a good pace because a single result in error will not mean distributing an additional workunit copy (and waiting for it's result) in order to reach a quorum.
Fred
BOINC Alpha, BOINC Beta, LHC Alpha, Einstein Alpha
ID: 58154 · Report as offensive
Stiller Cruncher
Volunteer tester

Send message
Joined: 18 Jul 99
Posts: 1
Credit: 4,024,549
RAC: 197
Germany
Message 58163 - Posted: 29 Dec 2004, 8:45:45 UTC

Sorry guy's,

i do not understand really right.
We hav etu crunch all WU's at 4 times ?
Can you please explain the reason.
My first opinion was to crunch for the wastebasket.
I will not crunch for any points or places, i crunch for the E.T. as
descriped in youir name S.E.T.I.

Robert
ID: 58163 · Report as offensive
Profile david9000
Avatar

Send message
Joined: 3 Apr 99
Posts: 32
Credit: 222,393
RAC: 0
United States
Message 58168 - Posted: 29 Dec 2004, 9:27:47 UTC

We hav etu crunch all WU's at 4 times ?
Can you please explain the reason.

4 copies of the same WU are handed out to different people to make sure all data is crunched complete and correct. You just crunch one copy and get credit for that when the required redundancy quorum has been attained from the other copies.
<I><B>Processing for the Planetary Society since July 5, 1999
ID: 58168 · Report as offensive
Profile Paul D. Buck
Volunteer tester

Send message
Joined: 19 Jul 00
Posts: 3898
Credit: 1,158,042
RAC: 0
United States
Message 58185 - Posted: 29 Dec 2004, 11:47:02 UTC

There will be additional changes to the system down the road. The plan has always been to get something working for the BOINC Software and the Science Application. We have the basics of the system done now. And we have been smoothing out the operation of the BOINC System as a whole for almost an entire year (9 months or so) as a "live" project open to all ...

The cross-platform GUI is cooking and I am expecting it out no earlier than 3 months and no later than 6 (of course I could be wrong about the schedule, and I am just guessing). Once the teething troubles with it have been worked out we should see the mass migration of the participants off of he SETI@Home Classic project and into the BOINC Software.

Once that has settled in, I would expect that the cosmetic issues will be addressed, and one of those is the credit system and all of its delights. As far as the granting, or not granting, credit for a work unit; the official statements about that are that you may be granted credit. On the other hand it is entirely possible that you will recieve no credit at all.

As far as optimization goes, there is a group that is working on the science code for SETI@Home and trying to get those optimized for the major families of processors. For the Macintosh the TeamNN guys have some posted and many iindicate that they have achieved higher processing speeds. I tried one of the versions but did not see any substantial increase in speed. Of course it is entierly possible that I did not install the application correctly too ...

My expectation is that we will see hosting of binaries of the optimized versions (I would even be willing to do that myself, but don't have the time to chase down the versions) for the processor families.
ID: 58185 · Report as offensive
Profile Roks

Send message
Joined: 20 Dec 02
Posts: 55
Credit: 137,776
RAC: 0
Slovenia
Message 58187 - Posted: 29 Dec 2004, 12:01:36 UTC

No credit? No way. I think most people wouldn't be willing to do work for berkeley if they got no credit at all...
<img src="http://www.boincstats.com/stats/banner.php?cpid=d2319b8f0ad14565556d0ba45b64e779">
ID: 58187 · Report as offensive
Profile Paul D. Buck
Volunteer tester

Send message
Joined: 19 Jul 00
Posts: 3898
Credit: 1,158,042
RAC: 0
United States
Message 58194 - Posted: 29 Dec 2004, 12:19:03 UTC - in response to Message 58187.  

> No credit? No way. I think most people wouldn't be willing to do work for
> berkeley if they got no credit at all...

Actually, I would guess that most people don't really care. It is just that the ones that seem to care are very noisy about their, ahem, dissatisfaction.
ID: 58194 · Report as offensive
Profile Robert Sullivan, MD
Avatar

Send message
Joined: 31 Oct 00
Posts: 221
Credit: 358,173
RAC: 0
United States
Message 58196 - Posted: 29 Dec 2004, 12:27:50 UTC

Cut the blue wire.


Sorry. For some reason I love that line. :->

ID: 58196 · Report as offensive
Profile PT

Send message
Joined: 19 May 99
Posts: 231
Credit: 902,910
RAC: 0
United Kingdom
Message 58197 - Posted: 29 Dec 2004, 12:29:02 UTC - in response to Message 58194.  
Last modified: 29 Dec 2004, 12:29:32 UTC

> Actually, I would guess that most people don't really care. It is just that
> the ones that seem to care are very noisy about their, ahem, dissatisfaction.

I do agree with you - I really don't care about the credits. For me it's more like a progress indicator, nothing else! They can remove the credits completely since I use "View Results" in "My Account".

Happy crunching
ID: 58197 · Report as offensive
Previous · 1 · 2 · 3 · 4 · 5 · Next

Message boards : Number crunching : Faster credit: 4 copies of each WU initially being sent out now


 
©2018 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.