Hi

I changed the number of days that I want Seti Work units in my preferences to 5 - 7 days. Predictor is set to .5 - 1 days

Seti is set to be 99.8 % and Predictor is set to .20%.

Now when Seti tries to communicate and is unsuccessful( all the time lately), Predictor seems to be taking the setting from Seti when the Update button is used.

Seti tries to contact the scheduler,and fails to contact then immediately Predictor contacts it's server and downloads 26 work units with an average tome of 05:01:40.

an example is here

2004-08-05 20:06:27 [---] Starting BOINC client version 3.20 for windows_intelx86
2004-08-05 20:06:27 [Predictor@home] Project prefs: using separate prefs for home
2004-08-05 20:06:27 [SETI@home] Project prefs: using separate prefs for home
2004-08-05 20:06:27 [Predictor@home] Host ID is 5279
2004-08-05 20:06:27 [SETI@home] Host ID is 12208
2004-08-05 20:06:27 [---] General prefs: from SETI@home (last modified 2004-07-27 23:05:57)
2004-08-05 20:06:27 [---] General prefs: using separate prefs for home
2004-08-05 20:06:27 [Predictor@home] Resuming computation for result t0247C_1_439_2 using mfoldB110 version 3.12
2004-08-05 20:06:27 [Predictor@home] Resuming computation for result t0247C_1_6760_2 using mfoldB110 version 3.12
2004-08-05 20:06:29 [Predictor@home] Started download of t0218C_p3_9744
2004-08-05 20:06:29 [Predictor@home] Started download of t0218C_q3_9744
2004-08-05 20:06:32 [Predictor@home] Finished download of t0218C_p3_9744
2004-08-05 20:06:32 [Predictor@home] Approximate throughput 4022.857143 bytes/sec
2004-08-05 20:06:32 [Predictor@home] Finished download of t0218C_q3_9744
2004-08-05 20:06:32 [Predictor@home] Approximate throughput 4276.855721 bytes/sec
2004-08-05 20:06:32 [Predictor@home] Started download of t0218C_s4_9744
2004-08-05 20:06:32 [Predictor@home] Started download of t0218C_s1_9744
2004-08-05 20:06:36 [Predictor@home] Finished download of t0218C_s1_9744
2004-08-05 20:06:36 [Predictor@home] Approximate throughput 27556.148148 bytes/sec
2004-08-05 20:06:36 [Predictor@home] Started download of t0218C_s2_9744
2004-08-05 20:06:36 [Predictor@home] Finished download of t0218C_s4_9744

Lots of WU's downloaded I did not show them all.

Yes, you're right: the WU cache settings are global, which means they are shared between ALL BOINC projects you support. They are transferred each time you choose to Update a project (newest preferences overwrite the older ones).
Pitfully there's no way to stop e.g. Predictor from filling up the whole cache with WU when SETI isn't answering. So in most cases you'll be downloading lots of Predictor WUs, process them, when cache low water mark is hit, try to get some SETI WUs, but SETI won't respond, so you again download lots of Predictor WUs.... and so on.
Hopefully BOINC staff will soon implement a feature which adjusts the cache limit dynamically (i.e. if one project, which is set on high priority, doesn't answer, the BOINC client downloads only a few WUs from the secondary project).
But this problem will vanish when SETI is running stable. Until then, I recommend to set the WU cache limit very low, so you'll not get stuck with Predictor WUs. The disadvantage though is obvious: if SETI (once in our life) has enough WUs ready someone else will catch them...

Raphael Schweizer
Schweizer Informatik

EDIT: typo

Thanks for the quick responce and I will change the preference now before all the PC's are full of Predictor. Better Predictor than being Idle I suppose.

Why do both Projects have their own Preferences if BOINC takes the first one it gets and applied to both projects. Seems very strange to me that their is two sets of preferences and BOINC only uses the first one. Is it redundant information or is BOINC not operating correctly ???



> Yes, you're right: the WU cache settings are global, which means they are
> shared between ALL BOINC projects you support. They are transferred each time
> you choose to Update a project (newest preferences overwrite the older ones).
> Pitfully there's no way to stop e.g. Predictor from filling up the whole cache
> with WU when SETI isn't answering. So in most cases you'll be downloading lots
> of Predictor WUs, process them, when cache low water mark is hit, try to get
> some SETI WUs, but SETI won't respond, so you again download lots of Predictor
> WUs.... and so on.
> Hopefully BOINC staff will soon implement a feature which adjusts the cache
> limit dynamically (i.e. if one project, which is set on high priority, doesn't
> answer, the BOINC client downloads only a few WUs from the secondary
> project).
> But this problem will vanish when SETI is running stable. Until then, I
> recommend to set the WU cache limit very low, so you'll not get stuck with
> Predictor WUs. The disadvantage though is obvious: if SETI (once in our life)
> has enough WUs ready someone else will catch them...
>
> Raphael Schweizer
> Schweizer Informatik
>
> EDIT: typo
>

> Thanks for the quick responce and I will change the preference now before all
> the PC's are full of Predictor. Better Predictor than being Idle I suppose.
>
> Why do both Projects have their own Preferences if BOINC takes the first one
> it gets and applied to both projects. Seems very strange to me that their is
> two sets of preferences and BOINC only uses the first one. Is it redundant
> information or is BOINC not operating correctly ???
>
There are people running BOINC that are not running S@H. There are others that are not running Predictor. The same is and will be true for any particular project. Since there is no central authority, there is no central location for changing any settings. Therefore they have to be available everywhere, and migrate from project to machine to project until all are the same for that user.

Another point is that if you put the high and low water marks closer together you will get less work per connection. This should reduce the times that the queue is full of work from the less desired project. Making them the same or very close is not advised (it will work though) because it will generate more server traffic and can make the servers less stable.

My personal advise is 2 to 3 days this gets you enough work to last a weekend outage, you only fill a third of you queue at a time and you only generate server traffic once a day.

John Keck
BOINCing since 2002/12/08