From sorensen@uther1.phy.ornl.gov Tue May 13 13:51:14 1997 From: "Soren P. Sorensen" < sorensen@mail.phy.ornl.gov > Date: Tue, 13 May 1997 17:00:28 -0400 Subject: Early prototype To: challenge@sseos.lbl.gov Mime-Version: 1.0 Dear Challengers, Doug asked me to make a write-up on my thoughts for an early prototype for our Grand Challenge system, so here we go: The purpose of creating an early prototype is to test, if all components in our proposed system actually can work together and works by themselves. It will also test our infrastructure with questions like: How do we set up our code management system, who has access to the code, which platforms can the complete system run on etc. And we will get tested if we actually a thinking about the same system, because sometimes I have a feeling, that each of us within this collaborations have their own agenda, which might not be identical to the overall agenda. The early prototype should use data generated by Monte Carlo Event generators, like RQMD, VENUS, FRITIOF etc. generated at the T3E (and maybe other places) and written in a standard format and with a standard content (I will address that later in this message). All user interface should go through STAF and Craig should develop 1 or 2 ASPs interfacing STAF to Dave's stuff via the appropriate ODMG subset. Dave's stuff (I don't know what name to associate with Ari's and Dave's software!) should communicate with Ari's cluster software in order to know in what sequence or in which files to store the data (This is murky to me how this is done and I hope a prototype can force us to address how this interaction takes place). The system should basicly be able to run in three modes: 1) Restructure mode: Apply Ari's algoritmes to already existing data in order to restructure them in an optimal way 2) Storage mode: Storing new data in an object store, but in such a way that Ari's cluster indexes determine where and how the data is stored 3) Query mode: Data are retrieved from the object store. The general architecture I envision is something like: --------------------------------- - STAF - --------------------------- - - 3 - Mallon ODMG 2 - 1 - - ------------------------- - - Ari Cluster - --------------------------------- - Object store - --------------------------------- - HSS (Storage system) - --------------------------------- As you can see I don't really understand how exactly Ari and Dave code is interacting in the various scenarios. I assume once the data has been stored in the proper way in the object store, then we actually do not need Ari's software for doing the query? What is going to be our storage format? Does Dave's lightweight persistent object manager specify an object format? The input events for the prototype should be based on event generator output and should be in STAF's internal format. It seems to me, that we should aim for getting such a prototype up and running within the next 4-6 months and then use it to identify problem areas. ************************** Some comments on an Object Model for event generators: I have had some communication with Ron Longacre from STAR in order to try to agree on a common event format for Heavy Ion event generators at RHIC. We might as well use such a scheme within the grand challenge. There will basicly be 4 objects: (I will send out a message on objects attributes tomorrow) | / \ \ < --- one to many relationship Run (One instance per run or per file) | / \ Event (One instance for each produced event) | | | / \ | Particle (Many instances for each event) | / \ vertex (Many instances for each event) The relationship between particles and vertices is fairly complicated since each particle might be associated with up to two vertices: a creation vertex and a destruction vertex). I will try to expand on this tomorrow. Maybe I will learn one day to put it into OPM. Cheers, Soren -- ======================================================================== Soren P. Sorensen Dept. Physics, U. of Tennessee, Knoxville, TN 37996-1200 (423-974-7808) Oak Ridge National Lab., MS 6374,Oak Ridge, TN 37831-6374 (423-576-8764) 2408 Staplehurst Lane, Knoxville, TN 37932-1321 (423-694-9213) email: Soren-Sorensen@utk.edu fax: (423-576-5780) ========================================================================