MPI in TNT for parallel processing
0 likes474 views
The document discusses the Message Passing Interface (MPI) as it relates to parallel computing, highlighting its role in executing processes simultaneously on either clusters or multi-core computers. It explains that TNT, an existing platform with Parallel Virtual Machine (PVM), is integrating MPI due to its dominance in high-performance computing and its standardization for communication among parallel processes. The document also details the setup requirements and command functionalities for implementing MPI in the TNT environment, noting that development is nearing completion with a focus on compatibility and testing.
MPI in TNT for parallel processing
- 1. The Message Passing Interface (MPI) in TNT Martín Morales Pablo Goloboff UEL (CONICET – FML) 1/5
- 2. Parallel computing. Parallel computers. Parallel computing: simultaneous execution of computational processes that allow the division of a work or problem. Parallel computers: special hardware systems to perform parallel computing. • Clusters: group of individual computers connected by a local area network (LAN). Most common: Beowulf. • Multi-core computers: computer with a processor with two or more processing units. Most computers today are multi-core. That's means that we can run parallel apps (as TNT!) in a real parallel hardware without a Cluster; just with our single computer! MPI. The standard. The implementation. MPI: Message Passing Interface. Standard. It defines a communication protocol for parallel computing execution. An MPI implementation is the software that applies (implement) the standard. There are many implementations made of MPI. We choosed Open MPI. Why TNT-MPI? TNT already has PVM (Parallel Virtual Machine) for parallel jobs handling. However: • MPI remains the dominant model used in high-performance computing today. It has become a de facto standard for communication among processes that model a parallel program. • Actual supercomputers such as computer Clusters often run such programs. • Many sysadmins do not allow PVM to be run on their Clusters because if the way in which PVM uses resources (Pablo dixit.) TNT-MPI The project goal was to integrate MPI onto TNT and next to PVM. The user decides one or the other. Syntax and commands are practically the same. Hosts in MPI (Clusters) 2/5 Hosts in MPI are loaded from a simple text file usually called hostfile. Column 1: host name. Column 2: slots: how many processes can potentially run in this host? Column 3: max_slots: are we going to limit slots use in this host? master: usually, computer that starts processes (a job) on workers hosts and coordinates all of their work. worker: usually, computer that executes part of that job. Oversubscription: when the number of processes in the host is greater than cores or processors in it.
- 3. Install will be... MPI implementation binaries must be in all nodes (if Cluster) and pointed to them in env variables. Then it must define hostfile (if Cluster). Recommended implementation (at the moment): Open MPI version 4.0.1. Release date: March 2019. TNT-MPI Stage PARALLEL COMMANDS PVM MPI again, begin, cleansig, close, get, gwait, kill, mnemonic, pause, reset, resume, setram, setdata, spy, status, stop, timeout, wait IMPLEMENTED goto, host commands, load, skipto, tagset soon... Scripts As PVM, MPI scripts runs directly in TNT built-in script interpreter. But, if we want to use MPI with the new C interpreter we can do it “indirectly” with tnt() function (example below). Of course, we can do this with PVM too. Next • Full testing and probably... bug fixes. • Windows. This development is at the moment just Linux and OS X (Mac) compatible. 3/5 • Development is in a final stage. About 90%. • Main functionalities are already done. • We can run jobs and get results; run swaps, scripts...
- 4. Example. mult/swap searches. > tnt p coetal2.tnt, ptnt mpi, ptnt begin JOB 8 =mult2=ho1, return, ptnt wait . > tnt p coetal2.tnt, ra1, ptnt mpi, ptnt begin JOB 8 /swap 0= , return, ptnt wait . 4/5
- 5. Example. C-script with MPI parallel instructions. tnt() function. 5/5