 high performance computing
 
|
Single interconnect independent MPI across high performance clusters
Scali MPI Connect software is the only message passing interface on the market that enables companies to take advantage of the leading interconnect hardware through a single MPI implementation. Unlike other solutions, Scali’s integrated architecture enables third-party applications to be compiled once to run on the various leading interconnect technologies. The MPI allows binary programs that are linked with Scali MPI Connect to run on any of the supported interconnects without recompilation or relinking. Whether the cluster is built using Gigabit Ethernet, Myrinet, SCI, InfiniBand or a combination thereof, applications and users need only interact with Scali MPI Connect.
Scali MPI Connect highlights
Key interconnects and standards supported
- TCP/IP
- Fast Ethernet
- Gigabit Ethernet
- SCI
- Myrinet
|
High-End PerformanceScali has consistently demonstrated outstanding performance with low message latency for Linux clusters. Highly optimized algorithms for message passing through a direct data transport mechanism eliminates overhead and reduces latency. The MPI is thread-safe and -hot, enabling any mixture of message passing and shared memory programming.
Interconnect interoperability
Scali leverages the Direct Access Transport (DAT) interface to run Scali MPI Connect on the leading interconnect platforms and standards in the industry. Interconnect interoperability enables Scali MPI Connect to dynamically select the transport medium at runtime.
Reliability, fault tolerance and failover
Scali MPI Connect is designed for extensive reliability, availability and serviceability. The MPI operates transparently to transient network errors, interconnect resets or change of routing tables. It also supports failover between interfaces by automatically switching to an alternate network if available.
Scali MPI Connect Features
- DET - Direct Ethernet Transport
The DET is a Scali DAT module that runs over any communication interface with a MAC address, which offers substantially lower latency and overhead than standard TCP/IP. Channel bonding and out-of-order package handling enable linear scalability with the number of adapters used.
- Heterogeneous Cluster Support
The MPI supports applications with processes hosted on IA32, IA64 and AMD64.
- Automatic selection of transport mechanism
Dynamic selection of the optimal transport medium at runtime enables failover between individual connections and integration of clusters.
- Multithread safe and hot
Multithreaded applications can fully exploit Scali MPI Connect and multiple threads can simultaneously request services and conduct communication.
- UNIX Command Line Replication
The command line arguments are automatically provided to all MPI processes.
- MIMD Support
The Multiple Instruction - Multiple Data (MIMD) model is supported by having provisions for launching different executables.
- Tracing and monitoring
The system includes built-in tracing and timing facilities for performance tuning and debugging of parallel codes Environment variables determine which and how MPI related timing and trace information is presented. Using regular expressions, single or groups of
functions can be selected for monitoring.
- Support for debuggers
Scali MPI Connect supports Pallas’ Vampir MPI and GNU gdb.
|
 |
system description | |
|
|
| |
computenodes | |
|
|
| |
printversion | |
|
|
| |
communication network | |
|
|
|
|
