FSR – frustrating summary records

13
Rob Lambert, NIKHEF Core Soft, 10th October 2012 1 FSR – frustrating summary records Update R. Lambert file ^

description

FSR – frustrating summary records. file. ^. Update R . Lambert. Possible Solution. Fixing any of these will fix FSRs for good Get sparse trees in Root to work properly Have basket size and optimization tuneable tree-by-tree Markus realized this is possible in LHCb code - PowerPoint PPT Presentation

Transcript of FSR – frustrating summary records

Page 1: FSR –  frustrating summary records

Rob Lambert, NIKHEF Core Soft, 10th October 2012 1

FSR – frustrating summary records

Update

R. Lambert

file

^

Page 2: FSR –  frustrating summary records

Possible Solution

Rob Lambert, NIKHEF Core Soft, 10th October 2012 2

10,000 FSRs Sparse Trees in TES

ROOT can’t handle sparse trees

Sparse Trees Persisted

ROOT Tree used

Fixing any of these will fix FSRs for good

8. Get sparse trees in Root to work properly• Have basket size and optimization tuneable tree-by-tree• Markus realized this is possible in LHCb code

• Available in head or Online/RootCnv … let’s try it …

Page 3: FSR –  frustrating summary records

Two Parameters Buffer:

Reserved memory for each leaf in the Tree

Basket: Supposed sizes of

all the leaves

Rob Lambert, NIKHEF Core Soft, 10th October 2012 3

Page 4: FSR –  frustrating summary records

Test Cases

Rob Lambert, NIKHEF Core Soft, 10th October 2012 4

A B

Lxplus 135Only me using itMy code on local diskLHCb code on cvmfsWrite to local diskRead from Castor

Verified stability of CPU/Vmem

CPU : 10%Vmem: 0.5%File size: 0%

MC Filtered

40 FSRs, 7 levels

Real Data

13475 FSRs, 4 levels

Page 5: FSR –  frustrating summary records

Buffer

Rob Lambert, NIKHEF Core Soft, 10th October 2012 5

A B

Page 6: FSR –  frustrating summary records

Buffer

Rob Lambert, NIKHEF Core Soft, 10th October 2012 6

Original RootCnv

A B

Page 7: FSR –  frustrating summary records

Buffer

Rob Lambert, NIKHEF Core Soft, 10th October 2012 7

Markus/Ivan optimized

A B

Page 8: FSR –  frustrating summary records

Buffer

Rob Lambert, NIKHEF Core Soft, 10th October 2012 8

Suggested optimum

A B

Page 9: FSR –  frustrating summary records

Basket

Rob Lambert, NIKHEF Core Soft, 10th October 2012 9

Completelynot relevantA B

Page 10: FSR –  frustrating summary records

Only minor improvements possible!

Results

Rob Lambert, NIKHEF Core Soft, 10th October 2012 10

SetupProject DaVinci v32r2p1time gaudirun.py ~rlambert/public/forMarkus/fsrMemLeak/options.py#100 files with one event each, but 13475 FSRs in total

Version Time (usr+sys) / s VMEM / MiB File Size / MiBOld RootCnv 2834 2209 18.7

Markus/Ivan 2620 1303 18.7

Rob 1942 1276 18.7

Version Time (usr+sys) / s VMEM / MiB File Size / MiBNo FSRs 55 527 7.3

Merged FSRs 88 722 9.4

All FSRs 3067 1276 18.7

Page 11: FSR –  frustrating summary records

Summary Re-optimizing for FSRs not as good as we hoped

Writing *any* FSRs causes a leap in Vmem and CPU time

Need to keep merging, such that we never have 10k FSRs

Even 40 FSRs are a problem if buried under many levels

Again we can turn to these other options, but they are all very difficult … Automatic merging Event-like FSRs Flattening (Three possible directions) Abandon trees altogether

Rob Lambert, NIKHEF Core Soft, 10th October 2012 11

Page 12: FSR –  frustrating summary records

End Backups are often required

Rob Lambert, NIKHEF Core Soft, 10th October 2012 12

Page 13: FSR –  frustrating summary records

Discussion

Rob Lambert, NIKHEF Core Soft, 10th October 2012 13