| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Completed Research
Forum: Outsmart Ebola Together
Thread: OET performance on Android devices |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 48
|
|
| Author |
|
|
vlado101
Senior Cruncher Joined: Jul 23, 2013 Post Count: 226 Status: Offline Project Badges: 
|
Hello fellow crunchers,
----------------------------------------I currently have two android devices attached to WCG and they are running exclusively OET work units (since FightAids projects has stopped). I did not have any issues with the fight aid, however with OET I noticed some strange things with the duration of the work units and its check pointing. I sometimes noticed that sometimes when I would look on the BOINC app I would see a workunit that was 55% complete, but then it would go down to 45 or 40%. I also see a number of units that are staying at 100% for a long period of time (total time is close to 15 hours per work unit). Its not that I am concerned that the work units are getting errors (I see that they are eventually put under pending and then validated), but is OET optimized for the Android platform? Has sees this behavior? The android device in questions are: Product Name: HTC HTC One_M8 - SDK:21 ABI: armeabi-v7a Operating System: Android 3.4.0-gadd5817 (Android 5.0.1) BOINC version: 7.4.41 and Product Name: samsung GT-N5110 - SDK:19 ABI: armeabi-v7a Operating System: Android 3.0.31-1498778 (Android 4.4.2) BOINC version: 7.4.41 Thank you for your input.  [Edit 1 times, last edit by vlado101 at Jul 14, 2015 5:16:22 PM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
My Galaxy S5 Active has had some OET workunits that have sat at 100% for 12+hours then completed
|
||
|
|
vlado101
Senior Cruncher Joined: Jul 23, 2013 Post Count: 226 Status: Offline Project Badges:
|
My Galaxy S5 Active has had some OET workunits that have sat at 100% for 12+hours then completed I am finding the same issue. I am not sure if we should consider that as an issue. I mean I am fine if what its doing is crunching that work unit while it sits at 100%, but what I am concerned about is that its wasting that time when it could be working on other things. Also (might be a dumb question), since OET on my laptop take about 2 to 3 hours is the reason why it takes so long on an android is because it takes it a much longer time to process the same amount of data that takes 2 to 3 hours on a laptop? I understand the time difference but I was wondering if the android wu is pretty much the same amount of data as the one that my other machines get (ie desktop/laptop). |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Keep in mind that the progress percent and checkpointing is arbitrary. At start there's some algorithm that says "OK, this job is nn FPOPS which is so many cycles. When 1/8th of that estimate is done, write a checkpoint.". Then when the 8th checkpoint is written, and the job during that 100%, but still running is restarted, it reverts to 99.9%. Now why most all of the times the 0.01% takes a blink of an eye, and sometimes long hours... pass. Seen it on Windows, seen it on Android, particularly on Android.
So, practical application of this knowledge: BOINCTasks, who's not using it, shows number of checkpoints as [N] + time since last checkpoint. When seeing [8] were done, and the progress is 100% and the time since last checkpoint is getting a red background [you can set in BT when that is], don't interrupt or you will loose it all back to when 99.99% was reached. Regrettably, does not work with BOINC for Android, but did work for NativeBOINC too, the remote monitoring and logging. N.B.: Of course, as yet we've not heard from Tech [armstrdj] on what it's doing... [crunching or carrying coals to Newcastle], but long as the CPU time is counted, I'm tout OK. [And watching BOINC is so utterly boring, plus the program hates you for doing so... it's an AI without the 3 laws of robotics programmed into it ;o]. And vlado101, no no no, the 'original' _0 tasks are fully randomly assigned, no 'oh it's an android, let's give it a different sized on'. |
||
|
|
vlado101
Senior Cruncher Joined: Jul 23, 2013 Post Count: 226 Status: Offline Project Badges:
|
Thank you for that explanation SekeRob.
----------------------------------------Have you noticed any difference between Android versions? I have a tablet that is running Android 4.x version and it has a lot less issues that then the 5.x. Even with the OET it seems to work smoother and does not jump back and forth with the progress on some of the workunits. Too bad there is not an effort to make specifically android workunits. Do you think such an effort would help WCG return more research results? |
||
|
|
cristipurdel
Senior Cruncher Joined: Dec 13, 2008 Post Count: 158 Status: Offline Project Badges: 
|
I am getting the same behavior on 10 different android phones with OET.
----------------------------------------Sometimes it was stuck at one of the 8 check-point percentages, and just staying there for days until the deadline was approaching and had to manually abort the WUs For the moment I gave up hope to run OET tasks on phones which run it during the night. I will try to run on one phone continuously to see if it's better or not. My suggestion would be to have shorter WU's for Android OR Have 1 check-point every 30 minutes @1GHz ARM core FAAH did not have this behavior, and it was running much better. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
The difference between FAAH and other sciences versus OET is, multiple compounds could be packaged in one job and then checkpoint at each compound completion. OET is just 1 compound per job. It can run 2 minutes, it can run 20 hours, but the mean is under 2 hours, i.e. a checkpoint every 15 minutes. Each checkpoint is a write to memory... writing to memory is something you like to avoid.
Stuck at a checkpoint [Is the event log really showing as uninterrupted computing?], or is it a case of many restarts and thus not progressing from a certain percentage? The Result Status page Outcome link goes to the Result Log, so you could check there too. See it myself though, most of the times BOINC pauses [logged], but sometimes with longer hours interrupted [don't compute on battery] or manually suspended when on trip, the job goes back to last checkpoint or even 0.0%. I've stopped looking up as when my results go 90-96% valid for the tablet, depends on the 'mood' of what's computed it would appear. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
My Galaxy S4 has done merely any work on OET for a couple of weeks. I noticed a _2 replication wu at 85% progress with a deadline due weeks ago, always going a little up in progress, then falling back to 85%. The other 3 threads were doing the same, but at 0% instead of 85%. Sometimes, Boinc didn't do anything for hours, noticable from the cold touch of it. I aborted the old wu, seems to work now again. Unfortunately, the result's been purged from my stats already.
|
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
There's this preference along the line of "If screen is on", sort of like, if you are using the device, don't compute. Also, it the battery gets too hot, it will switch BOINC off, so if you set it to ambient temp [ours is now 31C indoors], it probably never computes. Don't know what default is...40C or so, but I've got it on 35C. The battery is non-replaceable, I want a long long use of my tablet.
 WCG is quick to remove anything 'Too Late' or validation state 3 [Too many errors]. Before it was during midnight , now also briefly after noon stats. Proximately 60 hours this past 7 days of perfectly good results, but 5 wingmen failing or not replying... no 90-96% score :/ |
||
|
|
cristipurdel
Senior Cruncher Joined: Dec 13, 2008 Post Count: 158 Status: Offline Project Badges:
|
Dumb question:
----------------------------------------Which is the time (CPU or Elapsed) considered for the badges? I am running 24/7 on an android device (4 cores) and on a laptop (1 core + HT) for OET and FAAH, and from what I see the ratio of CPU:Elapsed is 1:4 for android but only 1:1 for laptop. I was expecting to get 6 "days" of crunching daily, but from my current stats it's more like 2+1 days. Shouldn't it be 2+4? |
||
|
|
|