1. Perspectives on LHC Computing José M. Hernández (CIEMAT, Madrid) On behalf of the Spanish LHC Computing community Jornadas CPAN 2013, Santiago de Compostela

2. José Hernández The LHC Computing Challenge  The Large Hadron Collider (LHC) delivered in Run 1 (2010-2012) billions of recorded collisions to the experiments  ~ 100 PB of data stored at CERN on tape  The Worldwide LHC Computing Grid (WLCG) provides compute and storage resources for data processing, simulation and analysis  ~ 300k cores, ~200 PB disk, ~200 PB tape  The computing challenge resulted in a great success  Unprecedented data volume analyzed in record time delivering great scientific results (e.g. Higgs boson discovery) LHC Computing Perspectives 2

3. José Hernández Global effort, global success LHC Computing Perspectives 3

4. José Hernández Computing is part of the global effort 28 October 2013, Seoul, Korea CMS Computing Upgrade and Evolution 4 Computing

5. José Hernández WLCG (initial) computing model  Distributed computing resources managed using Grid technologies that needed to be developed  Centers interconnected via private and national high-capacity Ethernet networks  Centers provide mass storage (disk/tape servers) and CPU resources (x86 CPUs)  Hierarchical tiered structure  Detector data prompt reconstruction and calibration at the Tier-0 at CERN  Data intensive processing at Tier-1’s  User analysis and simulation production at Tier-2’s (LHCb only simulation)  Data tape archival at Tier-0 and Tier-1’s  Data caches at Tier-2s (except LHCb) LHC Computing Perspectives 5 All available WLCG resources have been intensively used during LHC Run 1

6. José Hernández ATLAS Computing scale in LHC Run 1 LHC Computing Perspectives 6  150k slots continuously utilized  ~1.4M jobs/day completed 10GB/s  More than 5 GB/s transfer rate worldwide

7. José Hernández CMS Computing scale in LHC Run 1 LHC Computing Perspectives 7  ~100 PB transferred between sites  ~2/3 for data analysis at T2s  Resource usage saturation. In 2012:  70k slots continuously utilized  ~500k jobs/day completed

8. José Hernández Computing challenges for Run2  Computing in LHC Run1 was very successful but Run 2 from 2015 poses new challenges  Increased energy and luminosity delivered by LHC in Run 2  More complex events to process  Event reconstruction time (CMS ~2x)  Higher output rate to record  Maintain similar trigger thresholds and sensitivity to Higgs physics and to potential new physics  ATLAS, CMS event rate to storage 2.5x  Need a substantial increase of computing resources that we probably cannot afford LHC Computing Perspectives 8

9. José Hernández Upgrading LHC Computing in LS1  The shutdown period is a valuable opportunity to asses  Lessons and operational experiences of Run 1  Computing demands of Run 2  The technical and cost evolution of computing  Undertake intensive planning and development to prepare LHC Computing for 2015 and beyond  While sustaining steady state full scale operations  With an assumption of constrained funding  This has been happening internally to the experiments and collaboratively with CERN IT, WLCG, common software and computing projects  Upgrade in parallel to accelerator and detector upgrades to push the frontiers of HEP LHC Computing Perspectives 9

10. José Hernández Computing strategy for Run2  Increase resources in WLCG as much as possible  Try to conform to constrained budget situation  Make a more efficient and flexible use of the available resources  Reduce CPU and storage needs  Less reprocessing passes, less simulated events, more compact data format, reduce data replication factor  Intelligent dynamic data placement  Automatic replication of hot data and deletion of cold data  Break down the boundaries between the computing tiers  Run reconstruction, simulation and analysis at Tier-1/Tier-2 indistinctly  Tier-1s extension of the Tier-0  Keep higher service level and custodial tape storage at Tier-1  Centralized production of group analysis datasets  Shrink ‘chaotic analysis’ to only what really is user specific  Remove redundancies in processing and storage, reducing operational workloads while improving turnaround for users LHC Computing Perspectives 10

11. José Hernández Access to new resources for Run 2  Access to opportunistic resources  HPC clusters, academic or commercial clouds, volunteer computing  Significant increase in capacity with low cost (satisfy capacity peaks)  Use HLT farm for offline data processing  A significant resource (>10k slots)  During extended periods with no data taking and even inter-fill periods  Adopt advanced architectures  Processing in Run1 done under Enterprise Linux on x86 CPUs  Many-core processors, low-power CPUs, GPU environments  Challenging heterogeneous environment  Parallelization of processing application will be key LHC Computing Perspectives 11

12. José Hernández Computing resources increase LHC Computing Perspectives 12  ~25% yearly growth preliminary requests for Run 2  Benefit from technology evolution to buy more capacity with same money HS06 PB

13. José Hernández Processing evolution  Sustaining throughput growth by replacing ever faster processors with a higher number of cores, co-processors, concurrency features  New environment: high concurrency, modest memory/core, GPUs  Multi-core now  many-core soon  finer grained parallelism needed  Many or most of our codes require extensive overhauls  Being adapted: geant4, root, reconstruction code, exp. frameworks LHC Computing Perspectives 13 Transistor count growth is holding up… …but clock speed growth suffered a heat death…

14. José Hernández Data Management  Where is LHC in Big Data Terms? LHC Computing Upgrade and Evolution 14 Business emails sent 3000PB/year (Doesn’t count; not managed as a coherent data set) Google search 100PB Facebook uploads 180PB/year Digital health 30PB LHC data 15PB/yr YouTube 15PB/yr US Census Lib of Congress Climate DB Nasdaq Wired Magazine 4/2013 Big Data in 2012 We are big… Current LHC data set, all data products: ~250 PB Reputed capacity of NSA’s new Utah data center: 5000 PB (50-100 MW, $2 billion)

15. José Hernández Data Management evolution  Data access model during LHC Run1  Pre-locate and replicate data at sites, send jobs to the data  We need more efficient distributed data handling, lower disk storage demands and better use of available CPU resources  The network has been very reliable and has experimented a large increase in bandwidth  (Aspire to) send only the data you need, only where you need it (and cache it when it arrives)  Towards transparent distributed data access enabled by the network  Industry has been at this approach for years, in content delivery networks  Already successful approaches during Run 1… LHC Computing Perspectives 15

16. José Hernández Data Management evolution in Run 1  Scalable access to conditions data  Frontier for Scalable Distributed DB Access  Caching web proxies provide hierarchical, highly scalable cache based data access  Experiment software provisioning to the worker nodes  CERNVM File System (CVMFS)  Evolve towards a distributed data federation… LHC Computing Perspectives 16

17. José Hernández Data Management evolution  Distributed data federation  A collection of disparate storage resources transparently accessible across a wide area via a common namespace (CMS AAA, ATLAS FAX)  Needs efficient remote I/O  CMS has invested heavily in I/O optimizations within the application to allow efficient reading of the data over the (long latency) network using the xrootd technology while maintaining a high CPU efficiency  Extending initial use cases: fallback on local access failure, overflow busy sites, allow interactive access to data, use diskless sites  Interesting approach: ATLAS event service  Ask for exactly what you need, have it delivered by a service that knows how to get it to you efficiently  Return the outputs in a ~steady stream, such that a WN can be lost with little lost processing  Well suited to transient opportunistic resources, volunteer computing where preemption cannot be avoided  Well suited for high-CPU low I/O workflows LHC Computing Perspectives 17

18. José Hernández From Grid to Clouds  Turning computing into a utility providing infrastructure as a service  Clouds evolve, complement and extend the Grid  Decrease heterogeneity seen by the user (hardware virtualization)  VMs provide a uniform user interface to resources  Integrate diverse resources manageably  Isolate software from physical hardware  Dynamic provision of resources  New resources (commercial, research clouds)  Huge community behind Cloud software  Grid of clouds already used by LHC exps  Several sites provide Cloud interface  ATLAS ~450k production jobs from Google over a few weeks  Tests on amazon EC spot pricing ~economically viable LHC Computing Perspectives 18

19. José Hernández Conclusions  LHC computing performed extremely well at all levels in Run 1  We know how to deliver, adapting where necessary  Excellent networks, flexible and adaptable computing models and software systems paid off in exploiting resources  LHC computing needs to face new challenges for LHC Run 2  Large increase of computing resources required from 2015  Live within constrained budgets  Use resources we own as fully and efficiently as possible  Support major development program required  Access to opportunistic and cloud resources, explore new computer and processing architectures  Evolve towards dynamic data access & distributed parallel computing  Explosive growth in data and (highly granular) processors in the wider world gives us a powerful ground for success in our evolution path  Evolve towards a more dynamic, efficient and flexible system LHC Computing Perspectives 19