From: Graham E Fagg Newsgroups: comp.parallel Subject: netsolve announcement Date: 22 Feb 1999 22:01:42 GMT Organization: University of Tennessee Approved: bigrigg@cs.cmu.edu Message-Id: <7ask46$dhn$1@goldenapple.srv.cs.cmu.edu> Originator: bigrigg@ux6.sp.cs.cmu.edu Xref: ukc comp.parallel:15410 NetSolve version 1.2 is now available (http://www.cs.utk.edu/netsolve/). NetSolve is a software environment for network computing that combines distributed computational resources into a unified service for solving complex scientific problems remotely. Using a client-agent-server paradigm, it aggregates the hardware and software resources of any number of computers that are loosely connected across a network and offers up their collective power through client interfaces that are familiar from the world of uniprocessor computing. When a user wants a certain computational task performed, he or she can use any one of a number of conventional software clients (e.g. MATLAB, Mathematica, simple procedure calls) to contact an agent with the request. The agent keeps track of information about all the servers in its resource pool, including their availability, load, network accessibility, and the range of computational tasks they can perform. The agent then selects a server to perform the task, and the server responds to the client's request. A load-balancing policy is used to manage computational resources to ensure good performance and simple retry provides fault-tolerance. NetSolve has been ported to every major UNIX platform. The NetSolve client is now available from C, Matlab and Mathematica interfaces on the MS-Windows platform, and both the agent and server components will soon be available for NT. The following numerical and scientific libraries have been integrated into NetSolve computational servers: ARPACK, BLAS, FFTPACK, FitPack, ItPack, LAPACK, MinPack, PETSc, and ScaLAPACK. These numerical libraries cover several fields of scientific computing; Linear Algebra, Optimization, Fast Fourier Transforms, etc. This list includes only the software that has been integrated by the NetSolve team, but it continues to grow. Adding new software libraries to NetSolve servers is relatively straightforward and users have added special libraries for their own purposes. Version 1.2 continues the effort to integrate NetSolve with other metacomputing systems. Earlier versions already provided an interface to the Condor system (http://www.cs.wisc.edu/condor/), which is being developed at the University of Wisconsin, Madison. Condor is a High Throughput Computing environment that can manage very large networks of separately owned workstations to harness otherwise wasted CPU cycles. A NetSolve server can be started as a front-end to a Condor pool; Condor then handles the job/resource mapping of standard NetSolve requests. Integration with Globus (http://www.globus.org/) and Legion (http://www.cs.virginia.edu/~legion/) is ongoing but less mature. NetSolve 1.2 uses the Globus Heart Beat Monitor to detect resource failures. Future releases of NetSolve will contain interfaces to the Globus GRAM for job launching (e.g. for batch MPPs), as well as to other components of the Globus Toolkit. Most of the changes in NetSolve 1.2 are the result of completely re-implementing the internals and redesigning the internal data structures, so they are invisible at the user-level. With this redesign the client, agent, and server code are downloaded as a single unit rather than as separate modules. But at compile time, you can still compile only the components that you want to use. This makes the system a lot cleaner and easier to manage. The other significant improvements to the system include the following: * We have implemented a concept of request farming that can be used to have NetSolve efficiently manage numerous requests for a single NetSolve problem type. The "request farming" interface is meant to be a general interface that allows NetSolve to facilitate manage resources for all problems of that type. This becomes extremely useful for efficient parametric studies in parallel. * For the problem specification, the following object types are available: packed files, character strings, and string lists. * In the problem description files used to expand the server, the object specification now carries an additional argument (the name of the object.) The keyword @format has been changed to @calling sequence. * The NetSolve error codes have been completely changed, so any previous software using NetSolve that has these values hardcoded will need to be modified. -- ************************************************************** Jack Dongarra dongarra@cs.utk.edu 104 Ayres Hall 423-974-8295 fax: 423-974-8296 Knoxville TN, 37996 http://www.netlib.org/utk/people/JackDongarra/ O__ Graham. Raleigh ProRace / LeMond Reno ,>/ GSX550 /=========================================================()=\()=======\ Dr Graham E. Fagg \ Distributed and Meta-Computing \ fagg@cs.utk.edu Innovative Computing Lab | PVM 3.4 / HARNESS \ GRM / SNIPE \ MPI_Connect Computer Science Dept | 104 Ayres Hall, Knoxville, TN 37996-1301, USA University of Tennessee | Phone: +1(423)974-5790 Fax: +1(423)974-8296 \======================================================================/ -- Articles to bigrigg+parallel@cs.cmu.edu (Admin: bigrigg@cs.cmu.edu) Archive: http://www.hensa.ac.uk/parallel/internet/usenet/comp.parallel