Veracity of current CPU time values

[kenbloom]

I got the numbers for the current-day GEN/SIM/DIGI/RECO out of the spreadsheet that was used for our most recent C-RSG request:

https://cms-docdb.cern.ch/cgi-bin/DocDB/ShowDocument?docid=13271

I used PU=35 and assumed that was a suitable luminosity scenario through both Run 2 and Run 3, given that the instantaneous luminosity isn't supposed to change much from here through Run 3. But I might have totally misinterpreted the numbers, so can someone please check? Maybe I will try assigning the issues to Chris and Liz. I've never assigned an issue before.

Also: there wasn't a separate value for DIGI in the spreadsheet. But since re-DR = 350 HS06 and RECO alone is listed as 250 HS06, I assumed DIGI = 100 HS06.

[Dr15Jones]

@davidlange6 @slava77 I was hoping either of you might be able to provide meaningful comments on the numbers provided about processing requirements.

[slava77]

It's a question to Tom (I think he entered the numbers based on my inputs). @tommasoboccali

[tommasoboccali]

ehm, where is the question?

tom

On Fri, May 26, 2017 at 7:29 PM, Slava Krutelyov notifications@github.com wrote:

It's a question to Tom (I think he entered the numbers based on my inputs). @tommasoboccali https://github.com/tommasoboccali

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/kenbloom/resource-modeling/issues/2#issuecomment-304342241, or mute the thread https://github.com/notifications/unsubscribe-auth/AD52YRe6t7mrL8AMlL16bVo64cH9a7asks5r9wvsgaJpZM4NnxI8 .

-- Tommaso Boccali INFN Pisa

[tommasoboccali]

ok, now I see it - btw interesting, can we discuss at some point what is this python ;)

We used these numbers from

• RECO only: ttbar reco from Slava. Assuming (which we believe is ~ true) his machine has 10 HS06 per core, we ended up in 20 sec per ev + some 10% (going to 25 sec/ev = 250 HS06s/ev)
• DR: the D part (which gives a +10 sec) can even be too low ... I think I saw a table by Slava at some point

AH, ok: here https://indico.cern.ch/event/624140/contributions/2533493/attachments/1438070/2212438/reco_2017_OnC_040317.pdf

it says D is 16, so more than 10. The numbers in the presentation (page 10), would give on

• reco (which hs to include miniaod time) = 18.5 sec (I think for real reco the +10% missing is about alca but not sure i remember)
• D+R = 34.5 sec

so in this last release (4 months after the estimates which entered the doc), reco is probably a bit too high, DR seems ~ ok

tom

[kenbloom]

Hey, it's becoming a party around here! What is this python? I keep telling people it's an expression of my midlife crisis (but I'm kidding).

So the bit of code (currently in cpu.py, but we'll get it into a JSON file I'm sure) that currently is

data_reco_time_lhc = 250 #this is per HS
mc_gensim_time_lhc = 500 #this is per HS
mc_digi_time_lhc = 100 #this is per HS

should really be more like

data_reco_time_lhc = 185 #this is per HS
mc_gensim_time_lhc = 500 #this is per HS
mc_digi_time_lhc = 160 #this is per HS

My main concern is that I have got the current reconstruction and the HL-LHC expectations in the same units...there is that factor of 10 for Slava's computer that I keep worrying that I'm not counting or double counting or whatever.

[slava77]

On 5/26/17 10:57 AM, Ken Bloom wrote:

Hey, it's becoming a party around here! What is this python? I keep telling people it's an expression of my midlife crisis (but I'm kidding).

So the bit of code (currently in cpu.py, but we'll get it into a JSON file I'm sure) that currently is

|data_reco_time_lhc = 250 #this is per HS mc_gensim_time_lhc = 500 #this is per HS mc_digi_time_lhc = 100 #this is per HS |

should really be more like

|data_reco_time_lhc = 185 #this is per HS mc_gensim_time_lhc = 500 #this is per HS mc_digi_time_lhc = 160 #this is per HS |

the numbers above are OK in HS06-s for Run2 and Run3 (phase-1)

My main concern is that I have got the current reconstruction and the HL-LHC expectations in the same units...there is that factor of 10 for Slava's computer that I keep worrying that I'm not counting or double counting or whatever.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/kenbloom/resource-modeling/issues/2#issuecomment-304349091, or mute the thread https://github.com/notifications/unsubscribe-auth/AEdcbnXIcnLkLXmIQBx6ey6hRTF2bvMPks5r9xKcgaJpZM4NnxI8.