The Scalable I/O (SIO) InitiativesLow-Level Application Programming Interface (SIO LLAPI) provides filesystem implementers with a simple low-Level interface to supporthigh-level parallel I/O interfaces efficiently and effectively. Thispaper describes a reference implementation and the evaluation of the SIOLLAPI on the Intel Paragon multicomputer. The implementation providesthe file system structure and striping algorithm compatible with theParallel File System (PFS) of Intel Paragon, and runs either inside thekernel or as a user level library. The scatter-gather addressingread/write, asynchronous I/O, client caching and prefetching mechanism,file access hint mechanism, collective I/O and highly efficient filecopy have been implemented. The preliminary experience shows that theSIO LLAPI provides opportunities of significant performance improvementand is easy to implement. Some high level file system interfaces andapplications, such as PFS, ADIO and Hartree-Fock application, are alsoimplemented on top of SIO. The performance of PFS is at least thesame as that of Intels native PFS, and in many cases, such as smallsequential file access, huge I/O requests and collective I/O, it isstable and much better. The SIO features help to support high levelinterfaces easily, quickly and more efficiently, and the cache,prefetching, hints are useful to get better performance based ondifferent access models. The scalability and performance of SIO arelimited by the network latency, network scalable bandwidth, memory copybandwidth, memory size and pattern of I/O requests. The tradeoff betweengenerality and efficiency should be considered in implementation.