System and method for increasing application compute client data I/O bandwidth performance from data file systems and/or data object storage systems by hosting/bundling all of the data file system storage servers and/or data object storage system servers in the same common global shared memory compute system as the application compute clients

摘要

Compute client processes are currently limited to a small percentage of the data I/O bandwidth available from a parallel file system's total aggregate I/O bandwidth. I/O bandwidth is limited by the pathways linking the parallel file system's data storage servers to the clients process's computer system and the number of stops the data makes in route before arriving in the client's memory space. Both of these limitations are overcome by hosting the entire set of file system storage servers or object storage servers within the same common global shared memory, parallel computer system, as the requesting client process. The data moves once, in parallel, from the storage devices, directly to the storage servers memory, which is the same memory address space as the compute client. This provides the compute client with low-latency access to the “Entire Aggregate I/O Bandwidth” provided by all of the File System Data Storage Servers.