Abstract: Method and apparatus for managing data in a distributed data storage system. In some embodiments, a plurality of storage devices define an overall available memory space. A control circuit stores a first copy of user data from a selected distributed data set in a working set of memory buffers, stores a duplicate, second copy of the user data in an alias set of memory buffers, generates parity data based on the second copy of the user data in the alias set of the memory buffers, and flushes the user data and the parity data from the alias set of memory buffers to the storage devices while the first copy of the user data remains in the working set of the memory buffers. In this way, subsequently received access commands can be serviced using the working set of the memory buffers.

Abstract: Applications that use non-volatile random access memory (NVRAM), such as those that apply file system journal writes and database log writes where write operations apply data sequentially over the NVRAM, map the available capacity of the NVRAM in a virtual address space without compromising performance. The NVRAM is segmented into regions with multiple such regions fitting within a volatile RAM element accessible to the application and the NVRAM. One or more regions are loaded in the volatile RAM and reflected in page tables that reference the regions. The page tables are managed on a host computer executing the application. One region space in the volatile RAM is unused and available for transferred information. Mechanisms are provided for dynamically transferring regions and interfacing with the host computer. As the application sequentially accesses information in the stored regions, older regions are removed and new regions loaded from NVRAM to the volatile RAM.

Abstract: An apparatus comprising a memory and a controller. The memory may be configured to (i) implement a cache and (ii) store meta-data. The cache comprises one or more cache windows. Each of the one or more cache windows comprises a plurality of cache-lines configured to store information. Each of the plurality of cache-lines may be associated with meta-data indicating one or more of a dirty state and an invalid state. The controller may be connected to the memory and configured to detect an input/output (I/O) operation directed to a file system. The controller may perform a read-fill based on a hint value when there is a read miss in the cache. The hint value may be based on the application access pattern. The hint value may be passed to a caching layer with a corresponding I/O.

Abstract: Systems and methods presented herein provide for SSD data storage via PCIe controllers configured with NVMe interfaces. In one embodiment, a PCIe controller includes a plurality of buffers, a Dynamic Random Access Memory (DRAM) device, and an I/O processor operable to partition the DRAM device into a plurality of logical blocks. The controller also includes virtual function logic communicatively coupled to the logical blocks of the DRAM device and to the buffers. The virtual function logic is coupled to a host system through the I/O processor to process an I/O request from the host system to a logical block of the DRAM device, to retrieve data from the logical block to at least one of the buffers, and to transfer the data from the buffer to the host system.

Abstract: Method and apparatus for managing data in a distributed data storage system. In some embodiments, a plurality of storage devices define an overall available memory space. A control circuit stores a first copy of user data from a selected distributed data set in a working set of memory buffers, stores a duplicate, second copy of the user data in an alias set of memory buffers, generates parity data based on the second copy of the user data in the alias set of the memory buffers, and flushes the user data and the parity data from the alias set of memory buffers to the storage devices while the first copy of the user data remains in the working set of the memory buffers. In this way, subsequently received access commands can be serviced using the working set of the memory buffers.

Abstract: A data storage system and methods for managing data to be transferred between a host and a data volume distributed across solid state storage modules are disclosed. A storage controller couples the host to the data volume and manages data transfers to and from the logical volume. The storage controller receives a set of parameters that define how an array of blocks and chunks of buffered data will be distributed across solid state storage modules. The storage controller receives and buffers data to be stored and transfers the same when the capacity of the buffered data will fill a set of arranged stripes in the defined array in a single write operation.

Abstract: Systems and methods presented herein provide for SSD data storage via PCIe controllers configured with NVMe interfaces. In one embodiment, a PCIe controller includes a plurality of buffers, a Dynamic Random Access Memory (DRAM) device, and an I/O processor operable to partition the DRAM device into a plurality of logical blocks. The controller also includes virtual function logic communicatively coupled to the logical blocks of the DRAM device and to the buffers. The virtual function logic is coupled to a host system through the I/O processor to process an I/O request from the host system to a logical block of the DRAM device, to retrieve data from the logical block to at least one of the buffers, and to transfer the data from the buffer to the host system.

Abstract: A RAID controller includes a cache memory in which write cache blocks (WCBs) are protected by a RAID-5 (striping plus parity) scheme while read cache blocks (RCBs) are not protected in such a manner. If a received cache block is an RCB, the RAID controller stores it in the cache memory without storing any corresponding parity information. When a sufficient number of WCBs to constitute a full stripe have been received but not yet stored in the cache memory, the RAID controller computes a corresponding parity block and stores the RCBs and parity block in the cache memory as a single stripe.

Abstract: A RAID storage system serializes data blocks to be stored in a RAID storage array and uses a primary map table and a number of secondary map tables to relate host addresses to logical block addresses in the storage array. Secondary map tables and other metadata can be cached from the storage array. The dual or two-tier map scheme and metadata caching promote scalability.

Abstract: A data storage system is provided that implements a command-push model that reduces latencies. The host system has access to a nonvolatile memory (NVM) device of the memory controller to allow the host system to push commands into a command queue located in the NVM device. The host system completes each IO without the need for intervention from the memory controller, thereby obviating the need for synchronization, or handshaking, between the host system and the memory controller. For write commands, the memory controller does not need to issue a completion interrupt to the host system upon completion of the command because the host system considers the write command completed at the time that the write command is pushed into the queue of the memory controller. The combination of all of these features results in a large reduction in overall latency.

Abstract: A block I/O interface for a HBA is disclosed that dynamically loads regions of a SSD of the HBA to a DRAM of the HBA. One embodiment is an apparatus that includes a host system and a HBA. The HBA includes a SSD and DRAM. The host identifies a block I/O read request for the SSD, identifies a region of the SSD that corresponds to the read request, and determines if the region is cached in the DRAM. If the region is cached in the DRAM, then the HBA copies data for the read request to the host memory and a response to the read request utilizes the host memory. If the region is not cached, then the HBA caches the region of the SSD in the DRAM, copies the data for the read request to the host memory, and a response to the read request utilizes the host memory.

Abstract: A RAID storage system serializes data blocks to be stored in a RAID storage array and uses a primary map table and a number of secondary map tables to relate host addresses to logical block addresses in the storage array. Secondary map tables and other metadata can be cached from the storage array. The dual or two-tier map scheme and metadata caching promote scalability.

Abstract: Applications that use non-volatile random access memory (NVRAM), such as those that apply file system journal writes and database log writes where write operations apply data sequentially over the NVRAM, map the available capacity of the NVRAM in a virtual address space without compromising performance. The NVRAM is segmented into regions with multiple such regions fitting within a volatile RAM element accessible to the application and the NVRAM. One or more regions are loaded in the volatile RAM and reflected in page tables that reference the regions. The page tables are managed on a host computer executing the application. One region space in the volatile RAM is unused and available for transferred information. Mechanisms are provided for dynamically transferring regions and interfacing with the host computer. As the application sequentially accesses information in the stored regions, older regions are removed and new regions loaded from NVRAM to the volatile RAM.

Abstract: An apparatus comprising a memory and a controller. The memory may be configured to (i) implement a cache and (ii) store meta-data. The cache comprises one or more cache windows. Each of the one or more cache windows comprises a plurality of cache-lines configured to store information. Each of the plurality of cache-lines may be associated with meta-data indicating one or more of a dirty state and an invalid state. The controller may be connected to the memory and configured to detect an input/output (I/O) operation directed to a file system. The controller may perform a read-fill based on a hint value when there is a read miss in the cache. The hint value may be based on the application access pattern. The hint value may be passed to a caching layer with a corresponding I/O.

Abstract: A method for managing redundancy of data in a solid-state cache system including at least three solid-state storage modules. The method may include designating one or more extents of each dirty mirror pair to be of a particular priority order of at least two priority orders. The at least two priority orders can include at least a highest priority order. The highest priority order can have a higher relative priority than the other priority orders. The method may also include performing at least one redundancy conversion iteration. Each redundancy conversion iteration includes converting extents of at least two dirty mirror pairs into at least one RAID 5 group and at least one unconverted extent. The extents of the at least two dirty mirror pairs can include extents designated to be of a highest remaining priority order. Each redundancy conversion iteration can also include deallocating the at least one unconverted extent.

Abstract: A data storage controller exposes information stored in a locally managed volatile memory store to a host system. The locally managed volatile memory store is mapped to a corresponding portion of a peripheral component interconnect express (PCIe) compliant memory space managed by the host system. Backup logic in the data storage controller responds to a power event detected at the interface between the data storage controller and the host system by copying the contents of the volatile memory store to a non-volatile memory store on the data storage controller. Restore logic restores a data storage controller state by copying the contents of the non-volatile memory store to the locally managed volatile memory store upon the application of power such that the data in the volatile memory store is persistent even in the event of a loss of power to the host system and or the data storage controller.

Abstract: A first I/O transaction request is sent to a storage controller for processing by firmware running on the storage controller. A second I/O transaction request is sent to storage hardware without further processing by the firmware running on the storage controller. Since the firmware did not process the second I/O transaction request, information associated with the second I/O transaction is stored in in a circular buffer accessible to the firmware running on the storage controller. The firmware running on the storage controller reads, from the circular buffer, the information associated with the second I/O transaction that was stored in the circular buffer.

Abstract: A data storage controller exposes information stored in a locally managed volatile memory store to a host system. The locally managed volatile memory store is mapped to a corresponding portion of a peripheral component interconnect express (PCIe) compliant memory space managed by the host system. Backup logic in the data storage controller responds to a power event detected at the interface between the data storage controller and the host system by copying the contents of the volatile memory store to a non-volatile memory store on the data storage controller. Restore logic restores a data storage controller state by copying the contents of the non-volatile memory store to the locally managed volatile memory store upon the application of power such that the data in the volatile memory store is persistent even in the event of a loss of power to the host system and or the data storage controller.

Abstract: An apparatus for elastic caching of redundant cache data. The apparatus may have a plurality of buffers and a circuit. The circuit may be configured to (i) receive a write request from a host to store write data in a storage volume, (ii) allocate a number of extents in the buffers based upon a redundant organization associated with the write request and (iii) store the write data in the number of extents, where (a) each of the number of extents is located in a different one of the buffers and (b) the number of extents are dynamically linked together in response to the write request.

Abstract: A first I/O transaction request is sent to a storage controller for processing by firmware running on the storage controller. A second I/O transaction request is sent to storage hardware without further processing by the firmware running on the storage controller. Since the firmware did not process the second I/O transaction request, information associated with the second I/O transaction is stored in in a circular buffer accessible to the firmware running on the storage controller. The firmware running on the storage controller reads, from the circular buffer, the information associated with the second I/O transaction that was stored in the circular buffer.