U.S. ATLAS S&C Planning Meeting - June 20151 ATLAS Software Infrastructure : Requirements and Goals...
-
Upload
edwin-burns -
Category
Documents
-
view
214 -
download
1
Transcript of U.S. ATLAS S&C Planning Meeting - June 20151 ATLAS Software Infrastructure : Requirements and Goals...
U.S. ATLAS S&C Planning Meeting - June 2015 1
ATLAS Software Infrastructure :Requirements and Goals at Run 2
Period
Alex Undrus
Alex Undrus – U.S. ATLAS S&C Planning Meeting – June 2015 2
Outline
Current status Plans for the next year Long term perspectives
This presentation reports for the following WBS items:
5.2.1 – Software Validation 5.2.2 – Librarian and Infrastructure
Services
Alex Undrus – U.S. ATLAS S&C Planning Meeting – June 2015 3
Current Status
Some statistical data Tendencies U.S. contribution to ATLAS
infrastructure
Alex Undrus – U.S. ATLAS S&C Planning Meeting – June 2015 4
Number of files in ATLAS offline release
C++
C/C++ h
eade
r
Pyth
onJa
vaXM
L
Fortr
an
Shel
l scr
ipt
0
5000
10000
15000
20000
25000
17678 20216 10531 835 1105 491 861
Calculated by cloc-1.62 for ATLAS nightly development release created on June 15, 2015 (cmt, InstalllArea, NICOS, platform
specific. genConf, dict areas excluded). External packages are not included.
Alex Undrus – U.S. ATLAS S&C Planning Meeting – June 2015 5
Number of lines in ATLAS offline release
0.00E+005.00E+051.00E+061.50E+062.00E+062.50E+063.00E+063.50E+06
3.06E+069.25E+051.18E+06
1.48E+051.53E+06
8.70E+04
5.30E+04
Calculated by cloc-1.62 for ATLAS nightly development release created on June 15, 2015 (cmt, InstalllArea, NICOS, platform
specific. genConf, dict areas excluded, comments and blank lines excluded). External packages are not included.
Alex Undrus – U.S. ATLAS S&C Planning Meeting – June 2015 6
Number of submitters and commits to offline SVN repository
05/16-06/15-2014
11/16-12/15-2014
05/16-06/15-2015
0 2000 4000 6000 8000 10000
4499
7227
7676
277
352
412
Number of submitters Number of commits
Alex Undrus – U.S. ATLAS S&C Planning Meeting – June 2015 7
New and persistent submitters to offline SVN repository (period of 05/16 – 06/15,
2015)
259
153 New authors (not active 1 year ago)Authors active 1 year ago
412 - number of unique submitters during 1 month (05/16/15 – 06/15/15)
809 - number of unique submitters during 1 year (06/16/14 – 06/15/15)
Alex Undrus – U.S. ATLAS S&C Planning Meeting – June 2015 8
ATLAS Nightly System at a glance
6
16
915
9
2 1 1
Nightly Branches in
December 2014
59 in total
8
15
8164
36
1123
Nightly Branches in June 2015 Major Branches
Migration
ASG
Patch Branches
Physics Anal-ysis
Athena Analysis
Exp. (e.g. CMAKE)
ROOT 6 integr. (NEW)
Athena Simula-tion (NEW)
MAC (RE-SUMED)Doxygen67 in total
NEW
NEW
Alex Undrus – U.S. ATLAS S&C Planning Meeting – June 2015 9
ATLAS Nightly System at a glance (2)
2013 Mar
2013 Jun
2013 Oct
2014 Feb
2014 Jun
2014 Sep
2014 Dec
2015 Feb
2015 Jun
0 20 40 60 80 100
SLC5SLC6
Number of ATLAS nightly jobs. Record high number of 100 daily jobs is registered on 08/01/2014. As of 06/22/2015 the Nightly System run
83 daily jobs.
Alex Undrus – U.S. ATLAS S&C Planning Meeting – June 2015 10
Statistics Interpretation
Athena development releases contain 7 million lines
The number of lines is not growing last 5 years, but they keep changing
Software development activity was always high and it is increasing since the Run 2 start 70% increase in SVN commits 50% increase of developers community Hundreds of new developers joined Many new nightly branches are opened (total number
reached the record high 67) Unprecedented rate of new stable releases: 1.4 per day
(~ 250 stable release Jan. 1, 2015 – June 24, 2015) New kinds of release for targeted use
Alex Undrus – U.S. ATLAS S&C Planning Meeting – June 2015 11
Intensity of Run 2 Software Development
. . . is high It is considerable higher than at LS1 period:
Some important LS1 software projects continue: CMAKE, ROOT 6 integration
New kind of releases for targeted use Athena Simulation Athena Analysis AtlasP1Mon (for Tier 1 online monitoring)
RootCore based releases are actively developed
New compilers are probed (gcc 4.9, clang) Release configuration management is under
major change (cmake)
Alex Undrus – U.S. ATLAS S&C Planning Meeting – June 2015 12
Infrastructure Challenges and U.S. Contribution
All areas (code configuration, documentation, externals, platforms&compilers, QA/QC, release building and distribution, code
repositories, information protection, etc.) must cope with increased workloads when the work force remains the same
Infrastructure needs both support and development (tools must be updated in sync with software they support)
ATLAS Software Infrastructure Team includes 19 persons contributing 7.5 FTE (with 50 % shortage: additional 3.7 FTE are needed for an adequate support) as of 04/02/2015
U.S. ATLAS contribution to SIT is currently 1 FTE as it appears in OTP (Alex Undrus and Shuwei Ye)
U.S. contributes to critical areas: the Nightly System (Alex Undrus), environment setup (Shuwei Ye), LXR service
Expertise Gain Influence on ATLAS-wide policies and
decisions Parallel and effective user support for
U.S. based physicists Librarian and User services at U.S. Analysis
Center, BNL PROOF farm and Tier I Center Capture innovations and new ideas
Alex Undrus – U.S. ATLAS S&C Planning Meeting – June 2015 13
Rationale of U.S. Participation in Infrastructure Projects
14
Plans for Next Year
Permanent goal: create supportive environment for code development, data processing and analysis jobs across all ATLAS sites and file systems (local, afs, cvmfs)
Key areas: Nightly builds (details in the next slides) Build and run-time (details in Shuwei's
presentation) Criteria of success: users satisfaction and
absence of complaints Explore new innovative technologies and
tools Alex Undrus – U.S. ATLAS S&C Planning Meeting – June 2015
Alex Undrus – U.S. ATLAS S&C Planning Meeting – June 2015 15
Plans for the Nightly System
Further improve ATLAS Nightly web and database services brought by successfully completed Nightly System LS1 upgrade
Add new on-demand functionality to the System Current system is for daily builds at fixed times Software coordinators increasingly request urgent nightly
builds Up to 100 nightly jobs are manually restarted monthly Some branches need irregular builds separated by few days New on-demand functionality will be demonstrated on July
9 at the Annual Nightlies Workshop (https://indico.cern.ch/event/387000)
Alex Undrus – U.S. ATLAS S&C Planning Meeting – June 2015
NEW ADMIN PANEL
Administrative functions for privileged
users, authentication via
CERN SSO
Main task button (restart, cancel, etc.)
Current and detailed progress
information
Alex Undrus – U.S. ATLAS S&C Planning Meeting – June 2015 17
Plans for the Nightly System (2)
Complete projects according to the schedule New Nightlies CVMFS server Hot spare for the Nightlies CVMFS server (greatly
improves the system reliability) ATLAS Nightly Mail Facility (personalized emails about
nightlies results) Adapt the System for cmake builds
Experimental cmake nightlies support compilations, no tests yet
Assess recent requests Expand doxygen documentation builds git repository support
Continue to encourage users to use ATLAS Nightlies DB for customized views (successful experience with Trigger developers) example on the next slide
Alex Undrus – U.S. ATLAS S&C Planning Meeting – June 2015
New panel of Yasu Okumura with summary
of Trigger related problems across major
nightly branches
Alex Undrus – U.S. ATLAS S&C Planning Meeting – June 2015 19
Plans for the Nightly System (technical work)
Update NICOS for new Tag Collector 3 Builds on CC7 (CERN CentOS 7) Optimization and testing new machines on
the nightly farm CERN IT plans to replace all real hardware
machines with VMs VMs performs differently, need a lot of testing
and optimization Key problem: I/O bottleneck prevents full CPU
usage New VM machines with SSD disks allow to
reach 60% CPU utilization (a success after some VM types could not be loaded above 20%)
20
Long Term Perspectives
Rising concerns: Unclear relationship between releases Confusion where essential parts of software
are located (simulation, digitization, reconstruction, derivations)
Dissatisfaction with standard release coordination bureaucracy, ways for development of new algorithms and techniques
Single platform support, essentially no software portability
All-inclusive offline releases ( installation size 12 GB, with few tens of externals )
Alex Undrus – U.S. ATLAS S&C Planning Meeting – June 2015
21
Long Term Perspectives (2)
Increasing requests for smaller releases with targeted purpose (Simulation, Athena Analysis)
Ideas about software restructuring
Alex Undrus – U.S. ATLAS S&C Planning Meeting – June 2015
Core
Event
Reco
Analysis
Core
Simulation
HLT
Analysis2
Analysis 1
Reco
???
22
Long Term Targets
Improved software structure Compact releases with targeted
purpose Expanded multiplatform support Decrease of centralized tag
validation bureaucracy and effort Discussion on July 2 at 13:30 -
15:00 – "Release build: technical session"
Alex Undrus – U.S. ATLAS S&C Planning Meeting – June 2015
Alex Undrus – U.S. ATLAS S&C Planning Meeting – June 2015 23
Perspectives for the Nightly System
Expansion to new platforms (e.g. PowerPC) 90% of the Nightly System is already
portable Moving some operations to the GRID
Making and testing releases on sites where jobs run would bring better results and save human and machine efforts
Nightly testing is already available on the GRID but attracted few tests so far
Desirable: developing communality with CMS (CMSSW), LCG (Jenkins) nightly builds
Keep users and management happy with the System is always a priority
24
Summary
U.S. contributes to key areas of ATLAS software infrastructure Infrastructure
Despite thin manpower the ATLAS Nightly System and Environment Setup procedures sustain increasing load and demand successfully
Keeping abreast of new technologies trends is at the top of U.S. contributors priorities The Nightly System LS1 upgrade brought new
database and web technologies and greatly improved user experience
The Nightly System is proactively prepared for meeting demands of new platforms (PowerPC), builds tools (cmake) and software restructuring
Alex Undrus – U.S. ATLAS S&C Planning Meeting – June 2015