Tasks duration
22 Feb 2016 17:24:12 UTC
Topic 198456
(moderation:
Hello.
I observed that, in the last month, the tasks from E@H became longer to achieve. End of January, they where measured in hours and now they need days of calculations.
Does it come from my computer (which is the same, with the same CPU and GPU's) or does it come from the tasks themselves ?
For info, my computer is a laptop, 4 monthes old and the fans seem to be OK (ventilation holes clean, etc.)
Tasks running at present time are gamma-ray pulsar binary search #1 1.00 (3 of them).
Cheers and thanks
JF
Tasks duration
)
Hello again.
I forgot to give an example :
for task labelled LATeahOOO1L_528.0_0_0.0_2206050_0 (gamma-ray pulsar binary search)
The spent time is 1d 11:13:50 and the remaining time is 1d 4:05:30 but the amount achieved is 96.660 %
Strange for me
In the same vein I have had a
)
In the same vein I have had a number of runs which never apparently end, being at 100% with no remaining time needed for days - sometimes over a week. There is no pending upload, the task just sits there blocking one of my two (normally) enabled task uses of my MAC (iMAC Mid 2011 2.7GHz Intel Core i5 12GB). So I only get 1 BOINC task running for a lot of the time. Eventually I abort the task, which frees up the blocked BOINC path for use by any available BOINC task. I have to assume that at the end of the run it appears to have gone into some sort of loop or wait state, though the OS/X task manager does not show any activity for the Einstein task. Most Einstein tasks do, however, appear to terminate without a problem.
RE: Hello. I observed
)
I don´t have a clue to help you, but I can say that the actual FGRPB1 WUs complete quite a bit faster than the previous FGRP4 (about 15% less on average, on my old iMac). So if you see the opposite, odds are that something is interfering your side, provided of course that you let it run full time to completion (at least once, to be able to compare).
To Nigel : I had once the
)
To Nigel : I had once the problem you point out : a task that never ends and I had to abort it, as you did.
To Jasper : I use to run my computer continuously 3 days and 3 nights, then I allow it to "rest" one night.
My computer is a laptop ASUS with NVIDIA and Intel GPU's and did not change since I bought it 4 monthes ago. It is running Win 10.
I think you are right and I should look at the other tasks running in the background (each upgrade of Win10 seems to add something !)
Thanks a lot to both of you.
Cheers
JF
Something else I was thinking
)
Something else I was thinking of meanwhile, maybe interaction with some antivirus software? You probably have that running as well; if so, it is usually safe to exclude the BOINC data directories from the scans, can make a difference also.
RE: Hello again. I forgot
)
Don't know if this is related but I got this this morning:
"In addition, we were able to utilize the superior "Pass 8" data from the Fermi-LAT team, and search in longer data sets than ever before."
The units are LONGER now!!
Also your L2 cache only being 256kb could be holding you back quite a bit, the unit is probably swapping stuff on and off the hard drive like crazy as the unit progresses along, since the L2 cache is small by todays new standards.
Thanks Jasper. I will try
)
Thanks Jasper.
I will try this too. My AV is McAfee which is allways running but it is new for me and I have to find out how to exclude a particular file from scanning.
Thanks for comments. I see
)
Thanks for comments.
I see my MAC has 256KB L2 per core and 6MB L3(1 processor, 4 cores). As I don't understand what the Einstein code actually does, I cannot comment about any swapping, though if I am not doing anything else there does not seem to be any disk activity in Activity Monitor. Being OS/X I don't have the same antivirus as Windows, but I do have the AVG Link Scanner and Rapport running. I don't turn my m/c off unless it is demanded by OS/X or other significant applications, as I like to run Climate Prediction as one of the other BOINC applications; I just let it use what it wants. At the moment Climate Prediction has few tasks for my m/c, and those that do come can be quite long. None of the other BOINC applications (ENIGMA and Rosetta) give me any trouble, so for the time being I am carrying on as before. If it appears to stop again (there are no E@H tasks on my m/c at present) I will copy and report the actual task that appears to be stuck before I decide to abort it.
RE: Don't know if this is
)
Those comments you quote were in relation to the FGRP4 run which made the discoveries and has now essentially finished. That's not a comment about how long the tasks will take for FGRPB1.
Actually they're not. There are already a few observations posted that suggest the new tasks have been sized very well and they are actually a bit faster than for FGRP4. I believe there is less 'content' per individual task but that content is analysed more deeply so that the outcome is designed to be a fairly constant crunch time when compared to the previous run.
I would regard my knowledge of memory caches (L1, L2, L3) as being pretty sub-standard, to the point that normally, I would not comment. In this case I feel I should. I could be quite wrong, but isn't there a big flaw in what you are saying? Isn't the basic benefit of these caches their speed? They are simply *faster* than *main memory* - ie nothing to do with disk swapping. If a commonly executed set of instructions can fit in cache it can be executed much faster there, than if it has to reside in main memory. If there is sufficient main memory there shouldn't be any disk swapping, irrespective of whether or not there is sufficient cache.
If I'm wrong, I'll quite happily stand corrected.
Cheers,
Gary.
To make it easier to answer
)
To make it easier to answer your queries, I've combined both your messages into this single response. In future, if you think of extra information to add or if you want to make any correction, please edit the original message rather than create a new one. You have an hour in which to do this (edit button is visible for that period).
Neither really :-).
The tasks are quite uniform in content so (on its own) a single task for a given science run is likely to take pretty much the same time as any other single task of the same type. Differences will probably be caused by how you mix different science runs and how this might exhaust the resources of your machine. Your machine currently has this list of tasks showing. Currently there are 3xO1AST, 6xBRP6, 2xBRP4G, 4xFGRPB1 and 5xFGRP4.
Please realise that a relatively low end laptop like this will always struggle with the compute content of these tasks. Your processor is seen by BOINC as having 4 'cores' but it's really a dual core with hyperthreading (HT) - so 4 virtual cores and only 2 'real' ones. You are also using your GPU (930M). You seem to be getting tasks from every single science run offered by the project. You are able to control this with preferences. I suspect that certain task mixes may slow down your crunch times whereas a different mix may work a bit better. This may account for the variations you see.
It would be worth your while to experiment with science runs that perhaps work better on your machine. It would also be worthwhile to test what would happen if you turned off HT. You should be able to do that in hardware settings (UEFI) or you could try simulating it by restricting BOINC to use only 50% of CPUs. FGRPB1 tasks have a higher RAM requirement than O1AST. However O1AST will use less RAM but may take longer to run - perhaps a lot longer. We wont know until one is completed. The current estimate will not be a reliable guide.
The best GPU app may well be BRP6 if you wish to crunch GPU tasks. You could turn off all runs except O1AST and BRP6 and see how that combination worked. You could then try just FGRPB1 and BRP6 for a period and observe any differences in performance. You could also try both of those with HT on or HT off to see what that did to performance. This is not something that can be done quickly - it will take quite a bit of time. With HT on you will have 4 concurrent CPU tasks. With HT off you will only have two, but they will run a lot quicker (and likely cooler). This could well be better for the longer life of your machine. If you run a laptop at 100% load it is likely to fail a lot sooner than otherwise.
Please realise that both original estimates and the estimated time left can be wildly inaccurate. This is particularly true when you run the whole range of different science runs, both CPU and GPU. If the elapsed time shows 1d 4:05:30 at 96.660% the task is likely to finish in a few more hours - certainly not in a further 1d 4hr. From the task list I linked above, the actual final time was 131,209.12 secs, which is 1d 12:26:49.
Also, if you compare (in the linked list) elapsed time and CPU time for any given CPU task, you can see quite a big difference. For the task that took 131,209.12 the CPU time was only 111,337.20. Big differences like this are indicative of the CPU struggling to meet all the demands being placed on it. It could be that you have other compute intensive jobs running or (if essentially only crunching) that the crunching mix is too much for the CPU to handle. In that case, the only thing you can do is perhaps experiment with different/lighter loads until you discover what's best for your machine.
Cheers,
Gary.