150 likes | 233 Views
ESD Support for UNIX Applications. Yet another common direction. Trigger for this discussion. LER not well served by online DIMAD “Opticians” using complex Unix-based codes People finding ways to “make do” for data acquisition, transfer, and implementation of results
E N D
ESD Support for UNIX Applications Yet another common direction
Trigger for this discussion • LER not well served by online DIMAD • “Opticians” using complex Unix-based codes • People finding ways to “make do” for data acquisition, transfer, and implementation of results • ESD has been moving to Unix too
Raises many questions • Which disk areas are to be shared? • How should they be managed? • How do offline and online model-based applications interfere with each other? • How can we best provide the necessary BPM data? • The idea of producing a “config” file is great, how can we better support that? • Where (physically and logically) do we put an Octo-CPU Barnburner-6 Linux box for analysis code? • How can OPS, ESD, and PEP-II best cooperate to get the required job done?
Two separate kinds of support • “Sandbox” support – allow users to test ideas and codes without worrying about infrastructure issues • Production support – how to promote worthy code to production and provide for ongoing maintenance
Unix MATLAB Progress • Note: VMS MATLAB is frozen by the vendor and is two major releases old • Fast Channel Access now available (SSRL and SNS) • MATLAB Compiler now being tested • No license required to run compiled version!
Data Access for Unix MATLAB • Channel Access extensions now provide access to “normal” SLC database items as well as, of course, EPICS variables. • Currently NOT provided:Direct BPM accessHistory or Archiver data
Idea for Production MATLAB Programs • Test under your user account on a machine which has read access to data of interest • Convince your colleagues that your program is robust and useful • Compile the file into an executable • Move executable to gateway production area • Provide command line or “SCP button” access
AIDA – the Data Conduit(Greg White, Bob Sass, Ron MacKenzie et al.) • Will soon provide transparent, fast data access to SLC database, Channel Access, SLC history, Channel Archiver, and Oracle databases. • Will extend to complex BPM acquisitions. • User need not know ultimate source of data. • Extensive design documents available. • MATLAB Java interface begs to be a client
Local large NFS server • Will provide local common data storage • Visible from gateways and other production machines • Reduces reliance on SCS • ESD would welcome PEP-II “buy-in”
ESD’s Reasons for supporting Unix • Physicists now running new code on Unix • We do not have the resources to support systems, robust data access, AND applications. • The long lead time for application extensions on VMS leads to frustration and backdoor solutions • Ray has asked us to focus forward, not backwards. VMS is NOT forward. • If you can’t fight ‘em, join ‘em!
Unix support from VMS/SCP • Provide robust data access • Provide appropriate control access AND/OR • Provide applications as servers
VMS-based application servers • Buffered BPM acquisition(AIDA access may prove sufficient) • Correlation plots(Also possible to move app entirely) • Gobs of model-based code • Wire scanners, feedback control, …… • Plenty of opportunity for everyone
A surfeit of questions remain • Deployment procedures(new/old versions, “blessed” production area) • Screen management for interactive code • Data access priorities • Unix data management procedures • Reliability of SCS-dependent machines • Linux? Solaris? Both? • Steering the chaos
How to proceed • Form a small working group with ESD, OPS, and PEP-II representation. • Work together on plans for Unix (Solaris or Linux) program support. • Work together to plan application moves and data access requirements.