Project

General

Profile

Bug #8193

Updated by Evgeny Novikov over 7 years ago

Here is example of consumed resources (http://ldvstore:8998/jobs/200): 

 AVTG:  		 82.02 h 		 89.05 h  	 454.4 MB 
 CPAchecker:       45.14 h 		 72.98 h  	 12.24 GB 
 Core:  		 *27.61 h* 		 2.392 min 	 18.29 MB 
 EMG:  		 21.24 h 		 18.08 h  	 454.2 MB 
 LKBCE:  		 5.299 h 		 17.78 h  	 162.9 MB 
 LKVOG:  		 69.28 h 		 3.55 min  	 36.12 MB 
 RSB:  		 122.1 h 		 9.185 h  	 295.3 MB 
 RSG:  		 5.679 min 	 4.725 min  	 21.06 MB 
 SA:  		 20.29 h 		 29.15 h  	 177.0 MB 
 Sub-job:  	 0 ms  		 0 ms  		 0 B 
 VTG:  		 82.05 h 		 9.5 h  		 295.3 MB 
 Weaver:  	 33.96 h 		 41.54 h  	 287.3 MB 
 Total:  		 509.1 h 		 *287.4 h*  	 12.24 GB 

 Klever version: master (d70964d). 

 Total CPU time is about 287 hours, whereas total wall time is about 28 hours, then 287/28 = 10.25, which is impossible, since 8 CPU cores were used. 
 Therefore, CPU time for some components is computed incorrectly. 
 

 Most likely, this bug is related to AVTG and VTG components, since they require only a few minutes before. 
 If we subtract their CPU time (~98 hours), then we get 189/28 = 6.75, which is expected CPU load. 

 Current maximum RAM of components doesn't actually show real maximum RAM. For instance, LKBCE needs 162.9 MB while it's just a Core component, so Core needs at least 162.9 MB but its maximum RAM is just 18.29 MB. Total maximum RAM hasn't sense since it's actually unknown because of CPAchecker isn't a Core component and their RAM consumptions are evaluated independently. 

 The worst situation is with wall time because of it isn't additive by nature. I suppose to get rid of total wall time and of wall time of components that have more than one instance (#6676).  

 Some trivial fixes should be done in Bridge also.

Back