The Information Technology Solution Denis Foueillassar TELEDOS project coordinator
TELEDOS objectives (TELEservice DOSimetrie) Objectives The steps to reach the objectives 2 Provide dose calculation in less than 10 minutes. Dose distribution inaccuracy < 2% (statistical uncertainty) Assure total confidentiality & integrity of patient files IT ease of use for management and production associated to high availability and limited cost. To succeed in code parallelization and optimization MC code integration in a TPS (Treatment Planning System) Thorough metrological and clinic validation (photons and electrons beams) Integration success in hospital environment
Sharing responsibilities over project partners Selection, parallelization and optimization of the MC code IT (Information Technology) specification (hard, soft) and remote service solution design Dosisoft Integration and clinical validation 3 CEALIST, IGR Prototype development Bull Dosimeter, metrological validation CEALIST Dosisoft, IGR, IC, CAV et CAL
Presentation content 4 (IT) Information Technology expression of needs Architecture of the solution Hardware infrastructure presentation Software design Software architecture presentation Security aspects Performance measurement results
Information Technology: Expression of needs 5 Constraints Solution High Performance Computing required to support MonteCarlo code Cluster architecture and parallelization of the code Performance adaptability Scalable solution Precise calculation at low cost Ease of use and security Centralized solution and remote service offering Open Source software usage Submission through Internet or from ISOGray station No human intervention Confidentiality, Integrity, availability and traceability
Software architecture (first approach) Users Operator Administrator Software Oriented Architecture High Performance Computing 6 Finance
Hardware infrastructure Browser LAN H or Internet Service node Interconnect Isogray Compute nodes 7
Software design Web application JEE bus and SOA (Software Oriented Architecture) Jonas Dose calculation process adaptability JEE bus Application server Batch scheduling to execute a job queue associated to resource Mgr. JOB Queue Message Passing Interface (MPI) for the code Isogray MPI cpus Cluster management 8 Portability, scalability and simplicity to program Software deployment, monitoring, Cluster management
Software components HPC SOA WEB application Application server JONAS JEE bus OSGi EJB JCA Servlet JSP Tomcat Registry JNDI Transaction Mail Workflow BONITA BAS (Bull Advanced Server) 9 Linux kernel & distribution debug & dump analysis tools FileSystem (NFS) Interconnect (Eth) Resource management (SLURM) Cluster administration (NSMaster) Batch scheduler Compilers (GNU and commercial) Parallel libraries (MPI) Scientific libraries Profiling & performance measurement tools High Availability, RAID Storage & storage admin
Software architecture Management node Isogray or Browser Web application LDAP Workflow Bonita Jonas Tomcat connector JCA MySQL BatchMgt SLURM HPC Libraries Services offered to Users Dose calculation submission and recovery Job queue consultation and cancellation mailer 10 SOA Accounting MC code MPI Nœuds de calcul
User profiles Results recovery User : Hospital Status display Jo b ow ne r Display jobs waiting queue Input files teleload Operator : Job cancellation in the waiting queue Job submission All job s Send job request Send files Developer J2EE component addon 11 Change priority in the waiting queue Programming env. (IHM, IDE, ) Log & file cleanup
Scheduling and scenarios User Job submission Web appli) Workflow (Bonita) send(jobreques t) send(filerequest) send(request_granted) send(filesack) submitjob send(jobid) send(viewqueue) send(queue) send(queue) Send(getResults) send(getoutputfile) Send(JobRunning ) Send(JobRunning ) Alt : running job send(outputfile) send(outputfilename ) Alt : job ending 12 Slurm send(jobrequest) send(inputfiles) Results recovery JCA Connector send(queue) srun( ) send(jobid) send(viewqueue) send(queue)
Job processing Job properties : ID 1 : input file teledos_root (default) checkup Backup (default) Repository (default Type Accounting data save User Input_path Checksum save Mail to user 3 : Waiting (User resquest) File checksum File save Submit to Job scheduler Poll 2 : wait for result Cancel command User getting results 4 : End of job and data saving 13 Admin End User
Security aspects Access control (firewall, ldap) Input file coherency control GEthernet SSL NS422 NS460 VPN 32 Xeon cores Authentication (ldap) Encryption (SSL) 3x150Gb RAID SAS Confidentiality (VPN, no name) Availability (RAID) Integrity (file checksum) Traceability (Log, storage) 14
Performance measurement results Performance Dose calculation which require 5mn on 1 cpu makes 20s on 20 cpus 1000 REAL Time (s) Performances PenFast MPI v2.31 100 10 1 0 10 20 30 40 50 60 70 # procs utiles Code scalability Penelope/Penfast scalability is excellent (up to 80 cpus overhead 10% / 63 cpus. Performances PenFast MPI v2.31 SpeedUp (CPU Time) 80,00 70,00 60,00 50,00 40,00 30,00 20,00 10,00 0,00 0 10 20 30 40 # procs utiles 15 50 60 70 80
TELEDOS project roadmap 16
Worldwide references in a variety of sectors Educ/Research Manufacturing Aerospace and many others 17 Other
Merci de votre attention Denis.Foueillassar@bull.net