SEE-GRID-2 The SEE-GRID-2 initiative is co-funded by the European Commission under the FP6 Research Infrastructures contract no Bringing Interactivity to 3D Visualisation on the Grid Branko Marović Branislav Čukanović, Dragan Okiljević, Ivica Barišić, Jovana Vuleta, Danijel Bašić, Vladimir Čakarević et al. University of Belgrade Belgrade University Computing Center - RCUB

CE EGEE and SEE-GRID-2 Summer School on Grid Application Support – Budapest 6th July Introduction Objective Complete interactive analysis of 3D datasets from medical imaging devices (CT, MR, PET, SPECT…) Easy remote access over the Internet Generalized 3D tools for diagnosis, surgical planning, and therapy evaluation Virtual distant examination based on 3D datasets Basic idea Split the system into a client and server part Dedicate the server part to access and processing of volumetric data Produce a lightweight client for user interaction and 3D rendering Exploit grid resources and paradigms to overcome implementation issues

CE EGEE and SEE-GRID-2 Summer School on Grid Application Support – Budapest 6th July 20063

CE EGEE and SEE-GRID-2 Summer School on Grid Application Support – Budapest 6th July Screnshots - From Atoms to Galaxies

CE EGEE and SEE-GRID-2 Summer School on Grid Application Support – Budapest 6th July Open Issues Resolved by VIVE Architecture and Use of Grid 3D datasets require memory, network and computationally intensive processing How to get 3D data?  Reader of DICOM files from medical devices  Upload from PACS or HIS Computation resources?  Intensive computation is done by server part of the application  Web front end performs Grid job submission for the field server  Job start delays - Jobs are started in advance and pooled in binder part of the front-end Accessing and managing storage?  Portal upload, reformatting, copying to Grid storage and replication  Additional interfaces to data sources?  SEs, LFC replica catalog,  GridFTP, dCache, RFIO, and gsidcap protocols  GFAL, LFC, and LCG-UTILS interfaces Network - communication bottleneck?  Instances of server application have high bandwidth to data storage  Server can start with partially loaded segments of data  Optimized traffic between the server and client 3D in a Web browser?  VRML for 3D rendering, spatial interaction, and navigation  Java for GUI (applet) and communication  Client is dedicated to visualization and interaction only  Soon standalone Java client based on Java Web Start, Java 3D, Netbeans

CE EGEE and SEE-GRID-2 Summer School on Grid Application Support – Budapest 6th July What Can We Gain From the Grid? Distribution of processing and data over several nodes  Grid submission used to allocate server jobs  Allocating several machines for texture and surface mesh creation  Field stripping is not necessary for the regular medical images  Grid reduces overall demand for CPUs even with one server job per session  Could build distributed virtual data centers, with data abstracted within electronic health record Data transfer and replication  Replication of data to locations close to computation elements  Remote file access Security – high sensitivity of medical data  Data encryption in transfer (GridFTP, dCache, RFIO and gsidcap)  Anonymisation - removal of personal details to prevent intrusion at machine level, solved by gLite Hydra Application-specific monitoring  Could select a site from performance indicators (responsiveness, bandwidth, round trip time, processing speed) provided by the grid  Currently we just collect and display monitoring data using in our implementation guidelines and ideas from various grid/network monitoring projects  Fault tolerance

CE EGEE and SEE-GRID-2 Summer School on Grid Application Support – Budapest 6th July VIVE Gridification General benefits  Distribution of processing and data over several nodes  Data transfer and replication  Security – high sensitivity of medical data  Application-specific monitoring New grid-specific components  TCP binder (Client-server  Three-tier)  Mediating between clients and server jobs  Pool of server jobs  Persistence  Collecting monitoring and performance data  3D image management (LFC web front end, spatially segmented file format)  Measurement archive and web view of collected data Adaptations  Server (lcg-ManageSoftware, file I/O to both nearby classic and remote DPM/SRM SEs)  Glue web user interface  Client – longer delays and protocol changes

CE EGEE and SEE-GRID-2 Summer School on Grid Application Support – Budapest 6th July VIVE Gridification Client-server  Three-tier – TCP binder  Mediating between clients and server jobs  Pool of server jobs  Persistence – can respond to some client requests (mesh caching, scene data)  Collecting monitoring and performance data (network delays and request processing times) for measurement archive service 3D image management  LFC Web front end – image transformation, upload to the grid, replication  Spatially segmented format – work with partially loaded datasets, efficient RFIO Measurement archive and web view of collected data Glue web user interface & 2D viewer Adaptations  Server  Per-site installation with lcg-ManageSoftware facility  Efficient remote file access to both nearby (classic) and remote (DPM/SRM) SEs  Stripping of complex requests  Glue web user interface  Client – longer delays and protocol changes

CE EGEE and SEE-GRID-2 Summer School on Grid Application Support – Budapest 6th July Performance Patterns Start-up delay must be avoided at any cost! Network delay  Must be made acceptable for small transactions  OK for long-running requests

CE EGEE and SEE-GRID-2 Summer School on Grid Application Support – Budapest 6th July VIVE Grid Architecture

CE EGEE and SEE-GRID-2 Summer School on Grid Application Support – Budapest 6th July

CE EGEE and SEE-GRID-2 Summer School on Grid Application Support – Budapest 6th July

CE EGEE and SEE-GRID-2 Summer School on Grid Application Support – Budapest 6th July

CE EGEE and SEE-GRID-2 Summer School on Grid Application Support – Budapest 6th July TCP Binder Pool of server jobs  Pre-emptive submission prevents delays  Adapt to site responsiveness  Prevent flooding of queues  Job lifetime management  Maintaining the satisfactory number of jobs in the pool  Per-site policy

CE EGEE and SEE-GRID-2 Summer School on Grid Application Support – Budapest 6th July Performance measurement

CE EGEE and SEE-GRID-2 Summer School on Grid Application Support – Budapest 6th July Stripping: Partitioned Volume Processing Split a complex request into subvolumes Allocate satellite jobs from TCP binder Issue several parallel requests as client Retrieve partial results Integrate and return the final result Site A Site B Site C

CE EGEE and SEE-GRID-2 Summer School on Grid Application Support – Budapest 6th July Developer Resources Grid environment is constantly evolving, but  Useful features persist  New are constantly being added  Bugs are being fixed  Gained knowledge remains relevant, must be updated  Applications can be easily migrated to new/updated APIs LCG-2 User Guide  SEEGRID Wiki  Waiting for gLite User Guide

CE EGEE and SEE-GRID-2 Summer School on Grid Application Support – Budapest 6th July Software/Job Management Software installation management using ESM tools Experiment Software Installation in LCG-2  SEEGRID Wiki “SEE-GRID ESM Software Installation Guide”   Site configuration  Using ESM tools  Example JDL and shell scripts for installation, validation, and removal Job management EDG WMS API (LCG-2) – C++ & Java  index.html Exec/spawn of shell scripts that create and submit JDL – simple, but no direct feedback on jobs gLite 

CE EGEE and SEE-GRID-2 Summer School on Grid Application Support – Budapest 6th July Data Management LCG-2 (LCG-2 User Guide, “man” pages)  LCG-UTILS API – C/C++  LFC API – C/C++, Python  GFAL API – C/C++, Python  SEEGRID Wiki “SG Using file replicas and RFIO: UI configuration, rfiod, usage in apps, limitations and workarounds”  see.org/index.php/SG_Using_file_replicas_and_RFIO:_UI_configuration%2C_rfiod% 2C_usage_in_apps%2C_limitations_and_workarounds  Configuring UI, SE, RB  Site testing of RFIO/GFAL  Typical problems and solutions  Java access to LFC and LCG-UTILS Java LFC/GFAL wrapper  Customizable LFC web front end (upload, list, replicate, delete)  gLite  deployment/documentation/DataManagement/R3.0/

CE EGEE and SEE-GRID-2 Summer School on Grid Application Support – Budapest 6th July What VIVE Can Offer to Developers? Ready Experience in ESM software installation management Java LFC/GFAL wrapper Customizable LFC web front end In development Performance measurement facility  Service implementing measurement archive  Event recording API used by applications, sensors, and monitoring points or agents to put events into measurement archive  Basic archive querying API suitable for creation of reports  HTML-based web application for querying of the archive and display of reports  Advanced features, like generation and display of aggregations, a Web service, and MRTG graphs support could also be added. 3D Java environment framework  NetBeans/Swing based GUI: view and tool management  Integration of tool GUI and 3D appearance  Scene, lighting, navigation  Java 3D sensors, 3D and GUI events VIVE  Visualisation of other volumetric data sources : Earth sciences (atmospheric science, geology, oceanography), aerodynamics, astrophysics, thermodynamics, and dynamics of fluids. App-ware for interactive applications? Based on TCP binder experience

CE EGEE and SEE-GRID-2 Summer School on Grid Application Support – Budapest 6th July Java Web Start NetBeans Java 3D EJB3/Hibernate MySQL Servlets/JSP Java JNI Unix shell scripts Grid APIs (GRAM, MDS, WMS, GFAL, LCG-UTILS, LFC,…) DICOM JPEG Applets Sockets Java CGI-Bin VRML C++ Implementation Technology Evolution t

CE EGEE and SEE-GRID-2 Summer School on Grid Application Support – Budapest 6th July Thank you.

CE EGEE and SEE-GRID-2 Summer School on Grid Application Support – Budapest 6th July Field Server: Geometry Engine Two uses of isosurfaces – orientation in 3D and precise objects boundaries Creates and modifies geometry models Isosurface generation using Marching Cubes algorithm Compression through error-bound mesh simplification Grid can be used to parallelize mesh generation and simplification

CE EGEE and SEE-GRID-2 Summer School on Grid Application Support – Budapest 6th July VIVE Portal 2D preview Persistence for user sessions Web user interface to the grid UI Upload and replication of 3D images Binder  Connects clients and servers  Starting of jobs in advance  Pooling of “ready” jobs  Caching of surface meshes  Monitoring and logging of network delays and request processing times  Could select a Grid site by analyzing the recent performance and indicators provided by the Grid

CE EGEE and SEE-GRID-2 Summer School on Grid Application Support – Budapest 6th July Field Server: Field Engine Interpret data from  HIS (Hospital Information System)  PACS (Picture Archiving and Communication Systems)  Numerical simulations (e.g. blood flow simulation) Provide vector or scalar value for arbitrary location within image volume  V=F(x, y, z, t) Manage caching of volume parts Compute cross-section or volume textures for selected sub volumes Grid allows for distribution a large dataset over a number of nodes  Only if and when it becomes necessary

CE EGEE and SEE-GRID-2 Summer School on Grid Application Support – Budapest 6th July VIVE Client Requirements Current software requirements  Internet Explorer  Cortona VRML client  Microsoft Java VM Pure Java multi-view client in development