| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Active Research
Forum: Mapping Cancer Markers Forum
Thread: MCM asks not getting enough time and getting aborted |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 15
|
|
| Author |
|
|
Paul van Dijken
Cruncher Joined: Aug 15, 2011 Post Count: 5 Status: Offline Project Badges: 
|
I have reduced the CPUs used to 8 (50%) and the completion time has reduced from more than 6 hours to 3 hours. Still too late to prevent many aborted tasks, but hopefully good for the future.
Now it seems he WCG server is not running. |
||
|
|
pramo
Veteran Cruncher USA Joined: Dec 14, 2005 Post Count: 715 Status: Offline Project Badges: 
|
Garrulus glandarius, In reverse order (and drifting off the original topic, I fear)... Funny that my best CPU is estimated to run tasks 5x slower than it actually does. The initial estimates of time are based on the [absolutely useless] BOINC benchmarks and it can take quite a while for the system to work out what is right. If that's on a P-core, E-core system that could also confuse things!...a low-powered Ryzen 7 5825U throttled to 80C (resulting in ~3-3.1 GHz) but used as a dedicated BOINC rig performs much much better. Two of my (all-Linux) fleet members are mini-PCs; one is an Asus 5600H which I frequency cap at 3.6GHz and the other a MinisForum 7840HS which I cap at 4.1GHz. I also moderate the work mix (never allow more than 75% of CPU threads to be used by BOINC) and I divvy up the possible use of those threads by application; both systems have very respectable turnaround times, even when under a full workload -- on a good day the 5600H returns up to 120 MCM1 tasks a day (limited to 5 threads) and the 7840HS returns over 200 (limited to 8 threads) whilst only drawing about 25% more power. Nice little systems, and there's definitely a performance-per-watt improvement with the latest Zen architecture!I recently added a 8845HS to my BOINC farm, also throttled it to 80C (runs at around 4 GHz) and it has a 50% higher output than the 5825U. Unsurprisingly, it also uses almost 50% more power. Cheers - Al. So this just made me look- I did a quick search on "Asus 5600H" and "MinisForum 7840HS" and see prices for non bare bones systems from under $300 to $4digits. Some of the names and vendors I've never heard of but I haven't looked at hardware for 3-4 years anyway... about the cheapest I found- don't really need a 1TB SSD but thats what it comes with! If I get, it's be pretty much dedicated to Boinc, and maybe light web browsing https://www.alliwava.com/products/h56-mini-pc...untry=US¤cy=USD H56 Mini PC - Compact and High-Performance Computer AMD Ryzen 5 5600H / 32GB RAM + 1TB SSD. Base price $289.00 Was $469.00, free delivery $289.00 +tax Can you (and anyone else with similar mini's) share specs as far as make, mem, Storage, etc? Thanks all:)  |
||
|
|
Garrulus glandarius
Advanced Cruncher Romania Joined: Apr 10, 2025 Post Count: 88 Status: Offline Project Badges: 
|
I bought the 8845HS in a second-hand AceMagic AM18 with 32 GB DDR5 RAM + 1TB SSD (no dedicated GPU obviously) for about 400 euros.
----------------------------------------The 5825U is in a Beelink SER5 with 16 GB DDR4 RAM + 500GB SSD. Cost a bit over 200 euros as second hand. Spent another 100 to upgrade with 2x32 GB RAM to run CPDN and LHC smoothly.   |
||
|
|
TPCBF
Master Cruncher USA Joined: Jan 2, 2011 Post Count: 2173 Status: Offline Project Badges: 
|
I recently got a few hundred MCM tasks. They are estimated to run in 1.45 hours, but they need 7 hours to complete. With a short deadline (only a few days), many of them have not started before the deadline expires and they get aborted by BOINC. Why do you get "hundreds of MCM tasks" in the first place?I am not running anything else. Anyone else have similar problems? Sorry, but it seems that you messed the the default settings of the work queue and set it to a result loads it up beyond the reasonable limit your system can handle. The whole WCG system setup at this point still isn't stable, so anything can happen, including exceptionally long runtime estimates. When this happened immediately after the system as sending out new work a few days ago, on my systems the actual runtime was shorter than the estimate. The only other explaination is that yo have a system that is capping the performance for all or some of the CPU cores and hence an overloaded WU queue will also result in not being able to finish in time... |
||
|
|
Paul van Dijken
Cruncher Joined: Aug 15, 2011 Post Count: 5 Status: Offline Project Badges:
|
I did not make any changes to the default settings. This was working fine before the shutdown, but somehow the calculation of how much time is needed to complete is no longer realistic. Btw, the other projects (Einstein, primegrid, etc) work fine. Calculated time is roughly completion time.
I have set my laptop now to use only 50% of the CPUs. The completion time has reduced from hours back to 3.5 hours, but calculated time is still 1.5 hours. Which is just completing the same number of tasks (16 tasks in 7 hours = 2x8 tasks in hours) |
||
|
|
|