CMS numbers in Table1 (Updates marked in Red)
(1) Sample Statistics: 9.4->12.3B (we missed to count some samples)
(2) Reminder: Time/Evt numbers from Computing Experts
G+S = 50 sec /ev/core - D: 10 sec/ev/ecore - R: 25 sec/ev/core : - Tot: 85sec/ev/core
(3) GEN fraction: updates from https://indico.cern.ch/event/804266/
Old GEN faction: (1.41*0.05+1.96*0.17+0.43*0.05+1.94*0.04+0.712*0.05+0.154*0.05+0.876*0.05+1.95*0.05)/9.4 ~7.3%
New GEN faction: (1.62*0.05+4.58*0.17+0.43*0.05+1.94*0.04+0.712*0.05+0.154*0.05+0.876*0.05+1.95*0.05)/12.2 ~9.3%
where 0.05, 0.17 and 0.04 ....are from numbers achieved from local tests.
Thus for GEN we have 85*9.3%~7.9 sec/ev/core.
And the total GEN CPUs is 7.9*12.26B~96.9B
(4) In Table 1, we also provided GEN CPUs for each sample category:
For example, DY MG MLM~1.9s/event for GEN step based on local tests, thus the total CPUs is 1.9*1618M=3.07B.
Summing over the numbers we get in Table1 for each sample category, we get 95.5B, which is quite close to 96.9B
ATLAS numbers (Josh)
(1) Changes to numbers since last time came from two places:
- Double counting of a small number of samples
- Accidental removal of samples that should be included
--> Relatively small effect
(2) HS06s vs s
All the ATLAS numbers are raw seconds. Would be preferable to have in HS06seconds instead. Technically this is possible but will require relatively major update to the code, in the end it's just a matter of finding the time to do it.