Long WU's - Impending disaster?


log in

Advanced search

Message boards : Number crunching : Long WU's - Impending disaster?

1 · 2 · 3 · Next
Author Message
Angstrom
Send message
Joined: 20 Sep 99
Posts: 205
Credit: 10,131
RAC: 0
United Kingdom
Message 6246 - Posted: 10 Jul 2004, 21:47:07 UTC
Last modified: 10 Jul 2004, 21:58:34 UTC

I think it has now been established that the WU's indicating some 6x their normal time actually process normally. I would therefore guess that, in order to maintain the number of cached work units, many users may have increased their number of WU days by six times.

If this is the case then will this not cause a massive hit on work units when/if the Seti team revert to the normal WU times as cache settings will then be extremely high? This would be especially true for users that only connect occasionally.

Neil

STE\/E [BlackOps]
Volunteer tester
Send message
Joined: 29 Mar 03
Posts: 1137
Credit: 3,233,449
RAC: 160
Message 6247 - Posted: 10 Jul 2004, 21:49:02 UTC

More than likely Neil, it's like putting a donut in the hole of a leaking ship ... :/

Profile Keck_Komputers
Volunteer tester
Avatar
Send message
Joined: 4 Jul 99
Posts: 1575
Credit: 1,738,467
RAC: 610
United States
Message 6253 - Posted: 10 Jul 2004, 22:07:58 UTC

It is even worse than that Neil, when the proper estimates are put back in the people that didn't have to patience to wait out the problem will get tons of work they cannot finish in time. They will then likely be the same people whining about how long it is taking to grant credit.

John Keck
BOINCing since 2002/12/08

Angstrom
Send message
Joined: 20 Sep 99
Posts: 205
Credit: 10,131
RAC: 0
United Kingdom
Message 6255 - Posted: 10 Jul 2004, 22:12:48 UTC
Last modified: 10 Jul 2004, 22:13:55 UTC

I hadnt thought about WU's expiring I was just thinking about the initial load on the servers for downloads.

The only possibility I can see to returning to the normal times is to carry out a number of step reductions in WU times and hope that users who have increased their cache sizes also gradually reduce them. Whatever happens I bet is going to be ugly.

Neil

Heffed
Volunteer tester
Send message
Joined: 19 Mar 02
Posts: 1856
Credit: 40,736
RAC: 0
United States
Message 6260 - Posted: 10 Jul 2004, 22:24:40 UTC - in response to Message 6253.

> It is even worse than that Neil, when the proper estimates are put back in the
> people that didn't have to patience to wait out the problem will get tons of
> work they cannot finish in time. They will then likely be the same people
> whining about how long it is taking to grant credit.

I think that is a reasonable expectation.

<a> [/url]

Profile Petit Soleil
Avatar
Send message
Joined: 17 Feb 03
Posts: 1497
Credit: 46,577
RAC: 0
Canada
Message 6269 - Posted: 10 Jul 2004, 22:45:22 UTC
Last modified: 11 Jul 2004, 2:52:28 UTC

What a mess !!!

It is clear to me that if berkeley did that it's because they don't have WU at the moment and they don't want to receive millions of requests for nothing. What's the problem with that ? Everything was running smoothly since a few days
but now if users start to change their prefs in order to still be able to receive WU we will never get out of it.

I guess they know what they're doing at berkeley. We should just wait and see without starting messing arround with our preferences. That is the only way to assure a fairly and regular distribution of WU.

My 53 hours WU is the last one on the list and I won't receive any new one until
I have complited it. I am affraid that there won't be any more new WU left for me then if everybody changes their preferences.

The project is managed by Berkeley, not the volontary users.

Angstrom
Send message
Joined: 20 Sep 99
Posts: 205
Credit: 10,131
RAC: 0
United Kingdom
Message 6271 - Posted: 10 Jul 2004, 22:54:09 UTC

Precisely the point. I think the key is not to change preferences and hold on until they sort things out. Unfortunately I would guess that many people will already have made changes and more will follow as they watch their caches getting smaller and smaller.

What I dont understand is why we couldnt have been warned. Could they not have announced on the news feed that work unit times were to be artificially increased to temporarily ease server load and then appeal for people not to change their settings and indicate the consequences should they ignore that advice.

Sure some still would make changes but many, now knowing the background to the changes, would hold off.

Maybe they didnt think far enough ahead to realise the short term benefits may be just that - short term

Please SETI team in future let us know what is going on

Neil

Profile Purple Rabbit
Avatar
Send message
Joined: 31 Aug 99
Posts: 49
Credit: 3,492,591
RAC: 0
United States
Message 6277 - Posted: 10 Jul 2004, 23:11:15 UTC

It would have been nice to have been told the reason (if there is one) for the change in completetion times. Other threads suggest this is an "error" to be corrected later. I think it is a neat short term fix to make sure everyone gets at least one WU to chew on.

I'm satisified to get one WU when the old one is about to complete. If I miss a little time crunching because the new one is a bit late, so be it. Maybe I'm not "competitive" enough, but a big cache at this early stage is a short term gain at best. Dial up users may have a problem tho.

Purple Rabbit (Now at 2167 place and falling!)

Paul Opitz
Send message
Joined: 17 May 99
Posts: 1
Credit: 12,731,829
RAC: 21,598
United States
Message 6282 - Posted: 10 Jul 2004, 23:23:16 UTC

Might have been a smarter fix to leave time estimates alone and force everyone's cache settings back to the .1-1 default threshhold until the servers catch up. This is still doable (and would keep the system from getting sucked dry when the estimates go back to "normal").

Angstrom
Send message
Joined: 20 Sep 99
Posts: 205
Credit: 10,131
RAC: 0
United Kingdom
Message 6285 - Posted: 10 Jul 2004, 23:29:42 UTC - in response to Message 6282.
Last modified: 10 Jul 2004, 23:30:24 UTC

> Might have been a smarter fix to leave time estimates alone and force
> everyone's cache settings back to the .1-1 default threshhold until the
> servers catch up. This is still doable (and would keep the system from getting
> sucked dry when the estimates go back to "normal").
>
I think that would have been a good solution however changing user preferences would probably have been a step too far for many. I would guess that many users consider their settings as sacrosanct.

Maybe they believe that they can build up a large enough stockpile of WU's to be able to weather the storm.

Neil

Profile Keck_Komputers
Volunteer tester
Avatar
Send message
Joined: 4 Jul 99
Posts: 1575
Credit: 1,738,467
RAC: 610
United States
Message 6287 - Posted: 10 Jul 2004, 23:46:16 UTC

A not quite official explanation but closer than most of the guesses in this thread:
Someone saw that the alpha and beta estimates were 1/6th of what they should have been then changed the estimate on all three sites, instead of just those 2.

John Keck
BOINCing since 2002/12/08

Heffed
Volunteer tester
Send message
Joined: 19 Mar 02
Posts: 1856
Credit: 40,736
RAC: 0
United States
Message 6288 - Posted: 10 Jul 2004, 23:47:18 UTC - in response to Message 6285.
Last modified: 10 Jul 2004, 23:49:37 UTC

> I think that would have been a good solution however changing user preferences
> would probably have been a step too far for many. I would guess that many
> users consider their settings as sacrosanct.
>
> Maybe they believe that they can build up a large enough stockpile of WU's to
> be able to weather the storm.

Well, this is all still just conjecture, because they haven't said the reason behind it was to decrease WUs sent.

I personally don't think so, because they have absolute say over how many WUs a user can get daily. They already have a 50 WU per host limit in place. They could just as easily make it 4.

I simply think it was an error.

Edit: And John K. has just confirmed this. ;-)

<a> [/url]

Angstrom
Send message
Joined: 20 Sep 99
Posts: 205
Credit: 10,131
RAC: 0
United Kingdom
Message 6290 - Posted: 10 Jul 2004, 23:49:42 UTC - in response to Message 6287.

> A not quite official explanation but closer than most of the guesses in this
> thread:
> Someone saw that the alpha and beta estimates were 1/6th of what they should
> have been then changed the estimate on all three sites, instead of just those
> 2.


As you say, its an explanation. Wouldnt this have been picked up early on in the beta?

Ingleside
Volunteer developer
Send message
Joined: 4 Feb 03
Posts: 1546
Credit: 4,332,951
RAC: 1,204
Norway
Message 6291 - Posted: 10 Jul 2004, 23:51:12 UTC

Well, comparing wu, it looks like they've changed one of the analysis-limits. Maybe this is the reason they upped the expected crunch-times...


Anyway, as for cache-settings, remember BOINC has a deadline. So if you ask for 1000 days of work, you still will only get 14 days of work. :)
Therefore as long as the expected crunch-time is higher than the actual crunch-time, you very seldom will pass the deadline.
Ok, new installs will have "active-frac" at 1, so if you only crunches a couple of hours/day you can get too much work in the beginning, but this will automatically be decreasing.

Angstrom
Send message
Joined: 20 Sep 99
Posts: 205
Credit: 10,131
RAC: 0
United Kingdom
Message 6292 - Posted: 10 Jul 2004, 23:59:31 UTC - in response to Message 6291.

> Anyway, as for cache-settings, remember BOINC has a deadline. So if you ask
> for 1000 days of work, you still will only get 14 days of work. :)

Ok that doesnt look so bad then. I had assumed you would get whatever you asked for. Thinking about it now it makes sense to be only able to grab 14 days worth

Neil

Darren
Volunteer tester
Avatar
Send message
Joined: 2 Jul 99
Posts: 259
Credit: 277,051
RAC: 6
United States
Message 6298 - Posted: 11 Jul 2004, 0:22:14 UTC - in response to Message 6291.


> Anyway, as for cache-settings, remember BOINC has a deadline. So if you ask
> for 1000 days of work, you still will only get 14 days of work. :)

Remember, boinc is NOT seti. Boinc is simply the platform that seti runs in. Boinc does not have deadlines, and different participants (such as predictor@home and the weather prediction software) have different deadlines than seti does. Some programs have single work units that take months. If boinc imposed a 2 week deadline, these programs would never distribute a single work unit - as they can't be done in 2 weeks. It will ask for how ever many days of work you tell it to ask for. The only limitation at the moment for the seti program is the daily work unit max limit from seti - which is 50 wu per day. It is still possible to get 50 today, do 2, get 50 more tomorrow, do 2, get 50 the next day, etc...


Darren

SURVEYOR
Volunteer tester
Send message
Joined: 19 Oct 02
Posts: 375
Credit: 608,422
RAC: 0
United States
Message 6314 - Posted: 11 Jul 2004, 1:09:20 UTC

From Rom
"Okay, it seems one of the SAH dev's noticed a trend on a large number of
machines where they were off by a factor of 6. So he adjusted the splitter
estimates by increasing them by a factor of 6."

This change the estimated time for SetiAHome live wu's, but did not change the Alpha wu's.
The Beta site is down since Seti went live.
The Alpha is up and down.
P4 3.4HT two wu at the same time
Alpha estimates 00:25:14 complettion in 3:32+/-
SAH wu estimated 04:12:20 complettion in 3:30+/-
new SAH wu estimated 25:11:33
P4 2.53
Alpha estimates 00:28:35 complettion in 3:46+/-
SAH wu estimated 04:45:53 complettion in 3:45+/-
new SAH wu estimated 28:34:53

BOINC Alpha Tester
BOINC Beta Tester

STE\/E [BlackOps]
Volunteer tester
Send message
Joined: 29 Mar 03
Posts: 1137
Credit: 3,233,449
RAC: 160
Message 6316 - Posted: 11 Jul 2004, 1:10:04 UTC

Ok that doesn't look so bad then. I had assumed you would get whatever you asked for. Thinking about it now it makes sense to be only able to grab 14 days worth
==========

I wouldn't even want that many days cached because you still might not be able to get them all done by the deadline. When you set your Cache to 14 days BOINC can & will in all probability give you more than you can handle...

Profile Keck_Komputers
Volunteer tester
Avatar
Send message
Joined: 4 Jul 99
Posts: 1575
Credit: 1,738,467
RAC: 610
United States
Message 6322 - Posted: 11 Jul 2004, 1:20:48 UTC - in response to Message 6298.

>
> > Anyway, as for cache-settings, remember BOINC has a deadline. So if you
> ask
> > for 1000 days of work, you still will only get 14 days of work. :)
>
> Remember, boinc is NOT seti. Boinc is simply the platform that seti runs in.
> Boinc does not have deadlines, and different participants (such as
> predictor@home and the weather prediction software) have different deadlines
> than seti does. Some programs have single work units that take months. If
> boinc imposed a 2 week deadline, these programs would never distribute a
> single work unit - as they can't be done in 2 weeks. It will ask for how ever
> many days of work you tell it to ask for. The only limitation at the moment
> for the seti program is the daily work unit max limit from seti - which
> is 50 wu per day. It is still possible to get 50 today, do 2, get 50 more
> tomorrow, do 2, get 50 the next day, etc...
>
>
> Darren
>
The scheduler will not send out more work than can be completed before the deadline. Weather or not it includes work already in the queue properly I don't know.

John Keck
BOINCing since 2002/12/08

EclipseHA
Send message
Joined: 28 Jul 99
Posts: 1018
Credit: 242,251
RAC: 115
United States
Message 6330 - Posted: 11 Jul 2004, 2:10:27 UTC - in response to Message 6322.
Last modified: 11 Jul 2004, 2:11:23 UTC

One thing to note.

If you're not running the "gui" version of Boinc, you wouldn't even see that the completion time was increased 4 fold, unless you dug into the XML. So for folks running the Win cli, and everyone else, they won't react by changing their caches. They "might" notice that they have to request WU's more often, but this will be at the noise level based on the last two weeks of "no work from project". I'd bet that many folks with more than say 3 systems, only monitor enough to see if they're crunching - not look at each WU.

If intended or not, the 4 fold increase in time does have the effect of not distributing WU's as fast.

This is a case where (if intended) UCB, to me, made a good move. They reduced the draw on the supply of WU's, while letting all users have something to crunch. (maybe not as big a cache as they hoped, but they're still crunching!).

This is also a case where (if intended) UCB made a good move by NOT announcing it!, as there might be more users that tweek their profiles to get the backlog they had before!

Heck, if they can keep my systems busy where I have a small queue, that's cool!

Don't get me wrong. It would be nice if Boinc would keep up with real demands, or they'd limit the users/hosts crunching, but (if intended) this seems like a logical solution (why have thousands of WU's that won't be crunched for days, assigned out, when others can't get get a single WU!)

My 2c.

1 · 2 · 3 · Next

Message boards : Number crunching : Long WU's - Impending disaster?

Copyright © 2014 University of California