Abstract: A network storage controller uses a non-volatile solid-state memory (NVSSM) subsystem which includes raw flash memory as stable storage for data, and uses remote direct memory access (RDMA) to access the NVSSM subsystem, including to access the flash memory. Storage of data in the NVSSM subsystem is controlled by an external storage operating system in the storage controller. The storage operating system uses scatter-gather lists to specify the RDMA read and write operations. Multiple client-initiated reads or writes can be combined in the storage controller into a single RDMA read or write, respectively, which can then be decomposed and executed as multiple reads or writes, respectively, in the NVSSM subsystem. Memory accesses generated by a single RDMA read or write may be directed to different memory devices in the NVSSM subsystem, which may include different forms of non-volatile solid-state memory.

Abstract: A network storage controller uses a non-volatile solid-state memory (NVSSM) subsystem which includes raw flash memory as stable storage for data, and uses remote direct memory access (RDMA) to access the NVSSM subsystem, including to access the flash memory. Storage of data in the NVSSM subsystem is controlled by an external storage operating system in the storage controller. The storage operating system uses scatter-gather lists to specify the RDMA read and write operations. Multiple client-initiated reads or writes can be combined in the storage controller into a single RDMA read or write, respectively, which can then be decomposed and executed as multiple reads or writes, respectively, in the NVSSM subsystem. Memory accesses generated by a single RDMA read or write may be directed to different memory devices in the NVSSM subsystem, which may include different forms of non-volatile solid-state memory.

Abstract: A technique for achieving consistent read latency from an array of non-volatile solid-state memories involves an external entity determining the “busy” or “not busy” status of non-volatile solid-state memory elements in a RAID group. An external data layout engine then uses parity based RAID data reconstruction to avoid having to read from any memory element that is busy in a RAID group, along with careful scheduling of writes and erasures.

Abstract: A network storage system includes “raw” flash memory, and storage of data in that flash memory is controlled by an external, log structured, write out-of-place data layout engine of a storage server. By avoiding a separate, onboard data layout engine on the flash devices, the latency associated with operation of such a data layout engine is also avoided. The flash memory can be used as the main persistent storage of a storage server and/or as buffer cache of a storage server, or both. The flash memory can be accessible to multiple storage servers in a storage cluster. To reduce variability in read latency, each flash device provides its status (“busy” or not) to the data layout engine. The data layout engine uses RAID data reconstruction to avoid having to read from a busy flash device.

Abstract: In a computing system, cache coherency is performed by selecting one of a plurality of coherency protocols for a first memory transaction. Each of the plurality of coherency protocols has a unique set of cache states that may be applied to cached data for the first memory transaction. Cache coherency is performed on appropriate caches in the computing system by applying the set of cache states of the selected one of the plurality of coherency protocols.

Abstract: The data path in a network storage system is streamlined by sharing a memory among multiple functional modules (e.g., N-module and D-module) of a storage server that facilitates symmetric access to data from multiple clients. The shared memory stores data from clients or storage devices to facilitate communication of data between clients and storage devices and/or between functional modules, and reduces redundant copies necessary for data transport. It reduces latency and improves throughput efficiencies by minimizing data copies and using hardware assisted mechanisms such as DMA directly from host bus adapters over an interconnection, e.g. switched PCI-e “network”. This scheme is well suited for a “SAN array” architecture, but also can be applied to NAS protocols or in a unified protocol-agnostic storage system. The storage system can provide a range of configurations ranging from dual module to many modules with redundant switched fabrics for I/O, CPU, memory, and disk connectivity.

Abstract: A network storage system includes “raw” flash memory, and storage of data in that flash memory is controlled by an external, log structured, write out-of-place data layout engine of a storage server. By avoiding a separate, onboard data layout engine on the flash devices, the latency associated with operation of such a data layout engine is also avoided. The flash memory can be used as the main persistent storage of a storage server and/or as buffer cache of a storage server, or both. The flash memory can be accessible to multiple storage servers in a storage cluster. To reduce variability in read latency, each flash device provides its status (“busy” or not) to the data layout engine. The data layout engine uses RAID data reconstruction to avoid having to read from a busy flash device.

Abstract: A technique for achieving consistent read latency from an array of non-volatile solid-state memories involves an external entity determining the “busy” or “not busy” status of non-volatile solid-state memory elements in a RAID group. An external data layout engine then uses parity based RAID data reconstruction to avoid having to read from any memory element that is busy in a RAID group, along with careful scheduling of writes and erasures.

Abstract: The data path in a network storage system is streamlined by sharing a memory among multiple functional modules (e.g., N-module and D-module) of a storage server that facilitates symmetric access to data from multiple clients. The shared memory stores data from clients or storage devices to facilitate communication of data between clients and storage devices and/or between functional modules, and reduces redundant copies necessary for data transport. It reduces latency and improves throughput efficiencies by minimizing data copies and using hardware assisted mechanisms such as DMA directly from host bus adapters over an interconnection, e.g. switched PCI-e “network”. This scheme is well suited for a “SAN array” architecture, but also can be applied to NAS protocols or in a unified protocol-agnostic storage system. The storage system can provide a range of configurations ranging from dual module to many modules with redundant switched fabrics for I/O, CPU, memory, and disk connectivity.

Abstract: A network storage controller uses a non-volatile solid-state memory (NVSSM) subsystem which includes raw flash memory as stable storage for data, and uses remote direct memory access (RDMA) to access the NVSSM subsystem, including to access the flash memory. Storage of data in the NVSSM subsystem is controlled by an external storage operating system in the storage controller. The storage operating system uses scatter-gather lists to specify the RDMA read and write operations. Multiple client-initiated reads or writes can be combined in the storage controller into a single RDMA read or write, respectively, which can then be decomposed and executed as multiple reads or writes, respectively, in the NVSSM subsystem. Memory accesses generated by a single RDMA read or write may be directed to different memory devices in the NVSSM subsystem, which may include different forms of non-volatile solid-state memory.