BT 1.28

Started by Pepo, November 22, 2011, 09:03:33 AM

Previous topic - Next topic

0 Members and 4 Guests are viewing this topic.

Pepo

Quote from: Purple Rabbit on December 01, 2011, 02:54:12 PM
If you can't repeat the problem then you can't fix it  :'( What data can I collect to help?
The version(s) of connected BOINC client(s) might often give a clue...
Peter

Purple Rabbit

#46
All my computers are running BOINC 6.12.34 except one old Linux computer that won't upgrade beyond BOINC 6.10.58.

Purple Rabbit

#47
For fun I played around this afternoon. I opened Windows Task Manager and checked the BT CPU usage on the Q6600 as I viewed different tabs and computers. I found some interesting behavior. Note that 25 is one CPU on a quad.

Q6600/Win Vista (BT host): BT minimized 0-2, viewing Q6600 BOINC tasks 25, viewing Q6600 transfers 4-8. Other tabs show 4-8 as well. Well, this is interesting  ;)

Looking at 7 remote hosts (all tabs): 4-8 (on Q6600 CPU) except viewing AMD Phenom II X4 955/Win 7. This AMD host shows the same behavior as the Q6600; 25 viewing tasks, 4-8 otherwise.  The Q6600 and the AMD Phenom are my 2 fastest computers. They also have the largest number of tasks; 137 and 210 respectively.

I offer this as an observation hoping someone will have an "ah ha!" moment and will know exactly what's happening. I just find it strange   :o

Purple Rabbit

The problem seems to be with the number of tasks displayed. CPU usage went to normal for both computers with excessive usage when I displayed only active tasks. If I display all tasks again the CPU usage goes back to 25.

Beyond

Quote from: Corsair on November 27, 2011, 10:26:03 PM
during this evening I've seen BT using nearly the whole CPU and nothing could be done except re-start, and twice there has been to reboot the computer as immediately went to the top.
While I haven't seen this kind of usage, 1.28 uses many times more CPU than 1.21.  Lately as an experiment I've been running 1.21 on one of my desktop machines and 1.28 on the other.  1.21 has proven to be FAR more stable and uses very little CPU.  All settings are the same for both versions.  LT history is disabled in 1.28.  Both machines are running the same programs/projects and both have AMD CPUs with 2 GPUs (although the 1.21 box has 1 ATI and 1 NVidia while the 1.28 box has 2 ATIs).  If there is even a remotely flaky connection with one of the computers 1.28 becomes unstable and often has to be restarted.  1.21 will either pick the connection back up or restore it when double-clicked.  The situation gets worse when running many WUs.  I stopped running MW for Collatz a few days ago but when running MW had an AVERAGE of over 5 WUs/minute finishing.  1.21 also seemed to handle this situation better than 1.28.

Purple Rabbit

I need to amend a statement I made earlier. Given the "interesting" behavior on my Q6600 I tried BT versions 1.27, 1.28, and 1.29 a bit more thoroughly. They all exhibit "CPU Hogging". Falling back to BT 1.25 eliminates the problem on the Q6600. I deleted all history and disabled LT history which did not seem to have an effect.

My laptop (T8300, BT 1.28) has not exhibited this behavior over the past few days since I installed it. I'm guessing a configuration issue, but I usually (but not always) accept the default.

Saenger

Quote from: fred on November 22, 2011, 03:12:29 PM
Quote from: Pepo on November 22, 2011, 02:57:09 PM
Another machine, 4 cores, each of 3 CPU tasks is consuming approx. 85% of a core - should be ~64% of CPU. The graph again displays its line with spikes to 0.25-0.5% - the values are inconclusive.
I'm not quite sure what the question is, if there is any.
The graph display BoincTasks use not of any other program and it's not TThrottle related.
As I'm helping PinQuin to translate BoincTasks to German, I'm quite confused about the "Threads" and how to translate this.
Do they really just show the CPU-usage of the overhead program "BoincTasks"?
What's the use for this futile graph? Is there really anyone out there, who is interested in this? The use of the different real tasks, especially that of the GPU-ones, may be of some real concern, but who gives a flying f*** about the wee percentages of BoincTasks????

fred

Quote from: Saenger on December 08, 2011, 07:16:10 PM
As I'm helping PinQuin to translate BoincTasks to German, I'm quite confused about the "Threads" and how to translate this.
Do they really just show the CPU-usage of the overhead program "BoincTasks"?
What's the use for this futile graph? Is there really anyone out there, who is interested in this? The use of the different real tasks, especially that of the GPU-ones, may be of some real concern, but who gives a flying f*** about the wee percentages of BoincTasks????
That's why it's in the Expert tab, not used by almost anyone. Maybe only by myself. :o
This is added for trouble shooting and for those few who want to know everything.
It shows the CPU times the main program uses and parts of the program that access BOINC.
On most computers this isn't an issue at all. But some users have 10000++ Tasks on their machines and that may cause some overhead in BoincTasks.
This graph lets you see effects of changes. E.g. keeping a shorter history or disabling it.

Pepo

Quote from: Saenger on December 08, 2011, 07:16:10 PM
I'm quite confused about the "Threads" and how to translate this.
Do they really just show the CPU-usage of the overhead program "BoincTasks"?
What's the use for this futile graph? Is there really anyone out there, who is interested in this? The use of the different real tasks, especially that of the GPU-ones, may be of some real concern, but who gives a flying f*** about the wee percentages of BoincTasks????
Saenger, I hope you do not belong to the sort of people, who just on principle do kick aside anything they do not understand :( You could at least think of that maybe someone might be interested in (I am personally) or even might need such a tool (see all these request to solve BT's 100% CPU usage)...

I'm sure there are lot of hot tasks around, but everyone can have his/her own different opinion on which of them are "real", "the ones" to be solved with priority.
Peter