Dear Mr Anonymous


log in

Advanced search

Message boards : Number crunching : Dear Mr Anonymous

Author Message
rob smithProject donor
Volunteer tester
Send message
Joined: 7 Mar 03
Posts: 8535
Credit: 59,446,922
RAC: 86,023
United Kingdom
Message 1385386 - Posted: 28 Jun 2013, 14:03:56 UTC

[rant alert]
Your computer, known to SETI@Home as id 5049618 http://setiathome.berkeley.edu/show_host_detail.php?hostid=5049618 is producing so many "invalid" results as to be worthless. Do something about it,a and stop hiding behind your anonymity you coward.


Over 2000 invalid results, best part of 4000 results "invalid" and nearly 5000 awaiting validation. This person has to be ignorant of the situation he is in, and just doesn't care enough about it to be contactable.

[/rant alert]
____________
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?

Sten-Arne
Volunteer tester
Send message
Joined: 1 Nov 08
Posts: 3557
Credit: 20,756,219
RAC: 23,460
Sweden
Message 1385391 - Posted: 28 Jun 2013, 14:13:59 UTC - in response to Message 1385386.

[rant alert]
Your computer, known to SETI@Home as id 5049618 http://setiathome.berkeley.edu/show_host_detail.php?hostid=5049618 is producing so many "invalid" results as to be worthless. Do something about it,a and stop hiding behind your anonymity you coward.


Over 2000 invalid results, best part of 4000 results "invalid" and nearly 5000 awaiting validation. This person has to be ignorant of the situation he is in, and just doesn't care enough about it to be contactable.

[/rant alert]


After only (mostly) producing invalids, he still can get thousands of WUs/day.
(cuda32) 1106 today.
(cuda42) 555 today
(cuda50) 1098 today

There's got to be something seriously wrong with how the quota works (or doesn't work).
____________

Profile Tim
Volunteer tester
Avatar
Send message
Joined: 19 May 99
Posts: 205
Credit: 248,098,118
RAC: 162,725
Greece
Message 1385392 - Posted: 28 Jun 2013, 14:16:56 UTC

One thing is for sure…

All those ‘’pendings’’ will go to the ‘’invalids’’. :-0

Tim
____________

Josef W. SegurProject donor
Volunteer developer
Volunteer tester
Send message
Joined: 30 Oct 99
Posts: 4305
Credit: 1,074,011
RAC: 1,230
United States
Message 1385406 - Posted: 28 Jun 2013, 15:28:49 UTC - in response to Message 1385392.

One thing is for sure…

All those ‘’pendings’’ will go to the ‘’invalids’’. :-0

Tim

Only one of the 4 GPUs (device 3) produced the invalid results I looked at, the other 3 are apparently OK. So not ALL pendings will become invalid. Unfortunately the problem on device 3 makes it overflow very quickly so more than 85% of the tasks end up invalid.
Joe

Profile Link
Avatar
Send message
Joined: 18 Sep 03
Posts: 828
Credit: 1,571,842
RAC: 247
Germany
Message 1385426 - Posted: 28 Jun 2013, 16:18:16 UTC - in response to Message 1385391.
Last modified: 28 Jun 2013, 16:20:08 UTC

After only (mostly) producing invalids, he still can get thousands of WUs/day.
(cuda32) 1106 today.
(cuda42) 555 today
(cuda50) 1098 today

There's got to be something seriously wrong with how the quota works (or doesn't work).

Yes, the quota system in it's current implementation is almost completely useless. We've seen it here and also on many other projects many times.
____________
.

Profile MikeProject donor
Volunteer tester
Avatar
Send message
Joined: 17 Feb 01
Posts: 24551
Credit: 33,885,836
RAC: 24,338
Germany
Message 1385429 - Posted: 28 Jun 2013, 16:24:43 UTC

The quota is only rising so long you return valid work.

See Joe`s post.

____________

Sten-Arne
Volunteer tester
Send message
Joined: 1 Nov 08
Posts: 3557
Credit: 20,756,219
RAC: 23,460
Sweden
Message 1385433 - Posted: 28 Jun 2013, 16:35:59 UTC - in response to Message 1385429.
Last modified: 28 Jun 2013, 16:36:30 UTC

The quota is only rising so long you return valid work.

See Joe`s post.


Sure, but what good does the quota system do, if you return 355 valid work, and have Invalid (2096), Error (99), Validation inconclusive (3930) (which probably all will go to invalid, and Validation pending (4611) (of which probably 99% also will go to inconclusive first, and then to invalid.

Useless quota system IMO.
____________

Profile MikeProject donor
Volunteer tester
Avatar
Send message
Joined: 17 Feb 01
Posts: 24551
Credit: 33,885,836
RAC: 24,338
Germany
Message 1385434 - Posted: 28 Jun 2013, 16:39:33 UTC - in response to Message 1385433.

The quota is only rising so long you return valid work.

See Joe`s post.


Sure, but what good does the quota system do, if you return 355 valid work, and have Invalid (2096), Error (99), Validation inconclusive (3930) (which probably all will go to invalid, and Validation pending (4611) (of which probably 99% also will go to inconclusive first, and then to invalid.

Useless quota system IMO.


I partly agree but not everyone can babysit his PC 24/7.
He could be on vacation or just in hospital.
Who knows.

There`s nothing like perfect.

____________

Oddbjornik
Volunteer tester
Avatar
Send message
Joined: 15 May 99
Posts: 74
Credit: 87,369,543
RAC: 58,508
Norway
Message 1385462 - Posted: 28 Jun 2013, 17:31:18 UTC - in response to Message 1385439.

When something gets broken, there is nobody at the helm to fix it.
It is just a background program to them. Long ago forgotten.


That's why the quota system needs to be fixed. The users, obviously, can't be fixed.
____________

Profile Link
Avatar
Send message
Joined: 18 Sep 03
Posts: 828
Credit: 1,571,842
RAC: 247
Germany
Message 1385476 - Posted: 28 Jun 2013, 18:01:05 UTC - in response to Message 1385434.

I partly agree but not everyone can babysit his PC 24/7.
He could be on vacation or just in hospital.
Who knows.

There`s nothing like perfect.

And for exactly such cases there should be a working quota system. The current system is not working mainly for two reasons:
1. Invalids and validate errors are not counted as errors so they don't lower the quota.
2. The quota is doubled for just 1 valid result, so just 1 out of 50 results need to be valid (and that only if the other results were errors, invalids don't matter). I think 2% good results is a bit low quality requirement.
____________
.

Cosmic_Ocean
Avatar
Send message
Joined: 23 Dec 00
Posts: 2290
Credit: 8,813,816
RAC: 4,080
United States
Message 1385524 - Posted: 28 Jun 2013, 19:54:16 UTC

It also doesn't help that the "floor" value for the quota was set to 33, either.

Personally.. I think good setup (and I think this maybe got used YEARS ago..) is:

You start with a floor of 5. You return a good result, you get +2. You get an invalid/error, you get /2. That way if your hardware is behaving and you get yourself built up to 1000+, one bad result doesn't drop you back to 100 or 33 or near-zero, but down to 500, and then you can build it back up by 2's again.
____________

Linux laptop uptime: 1484d 22h 42m
Ended due to UPS failure, found 14 hours after the fact

rob smithProject donor
Volunteer tester
Send message
Joined: 7 Mar 03
Posts: 8535
Credit: 59,446,922
RAC: 86,023
United Kingdom
Message 1385531 - Posted: 28 Jun 2013, 20:13:53 UTC

Mr. Anonymous presents a number of issues.
He may be a "set and forget", but we have no way of informing him that his computer has a problem, so it will just continue generating garbage until it finally breaks. A little bit of preventative maintenance (such as a dust bunny chase) might save him a big bill when his GPUs expire due to frying their silicone brains.

His computer is generating garbage, and is continuing to get a substantial number of new tasks every day, because the results are "invalid", not "error".
Now we can all return the odd computational error, or invalid, but hundreds of errors and thousands on invalids is a bad situation which should be being trapped by the servers. The quota system is not working as well as it should, he should not be being sent as many tasks as he is.

____________
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?

Profile Wiggo
Avatar
Send message
Joined: 24 Jan 00
Posts: 7358
Credit: 96,905,274
RAC: 66,628
Australia
Message 1385534 - Posted: 28 Jun 2013, 20:28:51 UTC - in response to Message 1385531.

Yes, somehow the odd bad rig is slipping through the gaps in the floorboards but on the whole they are very much reduced from the SETI V6 days.

It's just a shame that they couldn't have implemented the throttling mechanism a year or so ago.

Cheers.

Profile Link
Avatar
Send message
Joined: 18 Sep 03
Posts: 828
Credit: 1,571,842
RAC: 247
Germany
Message 1385552 - Posted: 28 Jun 2013, 21:15:10 UTC - in response to Message 1385524.

You start with a floor of 5. You return a good result, you get +2. You get an invalid/error, you get /2. That way if your hardware is behaving and you get yourself built up to 1000+, one bad result doesn't drop you back to 100 or 33 or near-zero, but down to 500, and then you can build it back up by 2's again.

That might be not enough for faster hosts. I think 33 per day and core + 264 (8x33) per GPU to start with is OK, but the rest of the system has to become less "friendly". My idea would be something like this:
- return one valid result: +1
- return error or invalid result: -1

Now we all know, that GPUs generate an error every now and than, so to prevent reset to 33 (or actually 264) each time that happens we should have something like "consecutive error tasks" and for example allow up to 5 of them before the quota is reset to 33. So if a GPU is at 1000 tasks per day and returns 1 bad result, the quota will go down to 999, on next bad result 998 and so on. After 5 consecutive bad results it should however go down to 33-5=28, like it would do with the current system in case of errors.
____________
.

rob smithProject donor
Volunteer tester
Send message
Joined: 7 Mar 03
Posts: 8535
Credit: 59,446,922
RAC: 86,023
United Kingdom
Message 1385710 - Posted: 29 Jun 2013, 6:49:49 UTC

Link - that would be a fair way of doing it.
____________
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?

Message boards : Number crunching : Dear Mr Anonymous

Copyright © 2014 University of California