Abstract: Parallel training of a machine learning model on a computerized system is described. Computing tasks of a system can be assigned to multiple workers of the system. Training data can be accessed. The machine learning model is trained, whereby the training data accessed are dynamically partitioned across the workers of the system by shuffling subsets of the training data through the workers. As a result, different subsets of the training data are used by the workers over time as training proceeds. Related computerized systems and computer program products are also provided.

Abstract: A computer program product, according to one embodiment, includes a computer readable storage medium having program instructions embodied therewith. The computer readable storage medium is not a transitory signal per se. Moreover, the program instructions are readable and/or executable by a controller to cause the controller to perform a method which includes: assigning data having a first heat to a first data stream, assigning data having a second heat to a second data stream, and writing the data streams simultaneously, in parallel, to page-stripes having a same index across a series of planes of memory. The writing of the first data stream begins at an opposite end of the series of planes as the writing of the second data stream, the writing of the streams being toward one another. Other systems, methods, and computer program products are described in additional embodiments.

Abstract: A controller sets an error count margin for each of multiple units of a non-volatile memory and detects whether the error count margin of any of the multiple units has been exceeded. In response to detecting that the error count margin of a memory unit is exceeded, the controller determines whether calibration of the memory unit would improve a bit error rate of the memory unit sufficiently to warrant calibration. If so, the controller performs calibration of the memory unit. In some implementations, the controller refrains from performing the calibration in response to determining that calibration of the memory unit would not improve the bit error rate of the memory unit sufficiently to warrant calibration, but instead relocates a desired part or all valid data within the memory unit and, if all valid data has been relocated from it, erases the memory unit.

Abstract: A computer-implemented method, according to one embodiment, includes: performing a first read of one or more pages in a first page region of a first block. In response to determining that the highest RBER experienced during the first read is not in a first predetermined range, a first calibration procedure is performed on the one or more pages. A second read of the one or more pages is performed. In response to determining that the highest RBER experienced during the second read is not in a second predetermined range, a second calibration procedure on the one or more pages is performed, and a third read of the one or more pages is performed. In response to determining that the highest RBER experienced during the third read is not in the second predetermined range, a reliability counter which corresponds to the first page region of the first block is incremented.

Abstract: A computer-implemented method, according to one embodiment, includes: detecting that an error count resulting from reading a first page in a block of storage space in memory is above a first threshold, and reading a second page in the block of storage space. The second page is one which had a highest error count of the pages in the block of storage space following a last calibration of the block of storage space. Moreover, a determination is made as to whether an error count resulting from reading the second page is above the first threshold. In response to determining that the error count resulting from reading the second page is above the first threshold, the block of storage space is calibrated. Other systems, methods, and computer program products are described in additional embodiments.

Abstract: A non-volatile memory includes a plurality of blocks of physical memory, including a target block and at least one source block containing at least some valid data and some invalid data. Responsive to determining to perform garbage collection for the non-volatile memory, the controller transfers valid data from the at least one source block to the target block. The controller ends garbage collection on the at least one source block with at least some valid data present in the at least one source block and all interfaces of the target block closed at the boundary of independent layers. In at least some embodiments, the target block may be configured to store more bits per cell than the at least one source block.

Abstract: A computer-implemented method according to one embodiment includes determining, after writing data to a non-volatile memory block, one or more delta threshold voltage shift (TVS?) values. One or more overall threshold voltage shift values for the data written to the non-volatile memory block are calculated, the values being a function of the one or more TVS? values to be used when writing data to the non-volatile memory block. The overall threshold voltage shift values are stored. A base threshold voltage shift (TVSBASE) value, the one or more TVS? values, or both the TVSBASE value and the one or more TVS? values are re-calibrated during a background health check after a predetermined number of background health checks without calibration are performed.

Abstract: Performance of a data cache is controlled; the cache implements a garbage collection process for maintaining free storage blocks in a data store of the cache and an eviction policy for selecting data to be evicted from the cache. A cache performance control method defines a performance target for operation of the cache and, in operation of the cache, monitors performance of the cache in relation to the performance target. The garbage collection process is selectively performed in a relocation mode and an eviction mode so as to promote compliance with the performance target. In the relocation mode, data contained in a set of storage blocks selected for garbage collection is relocated in the data store. In the eviction mode, a set of storage blocks for garbage collection is selected in dependence on the eviction policy and data contained in each selected storage block is evicted from the cache.

Abstract: A non-volatile memory includes a plurality of physical pages each assigned to one of a plurality of page groups. A controller of the non-volatile memory performs a first calibration read of a sample physical page of a page group of the non-volatile memory. The controller determines if an error metric observed for the first calibration read of the sample physical page satisfies a calibration threshold. The controller calibrates read voltage thresholds of the page group utilizing a first calibration technique based on a determination that the error metric satisfies the calibration threshold and calibrates read voltage thresholds of the page group utilizing a different second calibration technique based on a determination that the error metric does not satisfy the calibration threshold.

Abstract: Methods, computing systems and computer program products implement embodiments of the present invention that include identifying, in a storage system including multiple storage devices having respective sets of storage regions, respective default low storage region thresholds that are used for garbage collection. For each given storage region, a time threshold and an alternative low storage region threshold greater than the default low storage region threshold for the given storage device are defined. While processing input/output operations for each given storage device, a count of unused storage regions in the given storage device is maintained, a timer is initialized, and upon the timer matching the time threshold for the given storage device, a garbage collection operation is initiated. In some embodiments, processing the input/output operations includes using a log-structured array format.

Abstract: A computer-implemented method, according to one embodiment, includes: receiving a multi-page read request and predicting whether using a multi-plane read operation to read pages of storage space in memory which correspond to the multi-page read request will result in a bit error rate that is in a predetermined range. In response to predicting that using the multi-plane read operation to read the pages will not result in a bit error rate that is in the predetermined range, a threshold voltage shift (TVS) value is computed for the multi-plane read operation. Furthermore, the pages are read using the multi-plane read operation with the computed TVS. Other systems, methods, and computer program products are described in additional embodiments.

Abstract: Embodiments for adaptive placement of parity information within Redundant Array of Independent Disks (RAID) stripes in a computer storage environment. A RAID controller periodically collects a physical capacity usage of each of a plurality of storage devices within the RAID. The RAID controller determines a placement of data and the parity information within at least one of the plurality of storage devices according to at least one of a plurality of factors associated with the physical capacity usage. The RAID controller writes the data and the parity information to the at least one of the plurality of storage devices according to the determined placement.

Abstract: A computer-implemented method, according to one embodiment, includes: calibrating a first block of storage space in memory, identifying a page in the calibrated first block having a highest RBER, and determining whether the RBER of the identified page is greater than an error correction code limit. In response to determining that the RBER of the identified page is not greater than the error correction code limit, a determination is made as to whether the RBER of the identified page is greater than a relocation limit. In response to determining that the RBER of the identified page is not greater than a relocation limit, another determination is made as to whether the first block has been excessively calibrated. Furthermore, in response to determining that the first block has been excessively calibrated, data in the first block relocated to a second block of storage space in the memory.

Abstract: A controller performs background reads of multiple physical pages of a selected physical block of a non-volatile memory. The controller detects asymmetric transient errors in a physical page among the multiple physical pages based on a bit error rate (BER) observed in the background read of the physical page. In response to detecting the asymmetric transient errors, the controller mitigates the detected asymmetric transient errors by relocating valid logical pages of data from the physical page to another physical block of the non-volatile memory and by retaining valid logical pages of data programmed into other physical pages of the selected physical block.

Abstract: A method for optimizing a read threshold voltage shift value in a NAND flash memory may be provided. The method comprises selecting a group of memory pages, determining a current threshold voltage shift (TVS) value, and determining a negative and a positive threshold voltage shift offset value. Then, the method comprises repeating a loop process comprising reading all memory pages with different read TVS values, determining maximum raw bit error rates for the group of memory pages, determining a direction of change for the current TVS value, determining a new current TVS value by applying a function to the current TVS value using as parameters the current threshold voltage, the direction of change and the positive and the negative TVS value, until a stop condition is fulfilled such that a lowest possible number of read errors per group of memory pages is reached.

Abstract: A computer-implemented method, according to one embodiment, includes: detecting that an error count resulting from reading a first page in a block of storage space in memory is above a first threshold, and reading a second page in the block of storage space. The second page is one which had a highest error count of the pages in the block of storage space following a last calibration of the block of storage space. Moreover, a determination is made as to whether an error count resulting from reading the second page is above the first threshold. In response to determining that the error count resulting from reading the second page is above the first threshold, the block of storage space is calibrated. Other systems, methods, and computer program products are described in additional embodiments.

Abstract: A controller-implemented method, according to one embodiment, includes: examining, by the controller, each of a plurality of journal entries from at least one journal beginning with a most recent one of the journal entries in a most recent one of the at least one journal and working towards an oldest one of the journal entries in an oldest one of the at least one journal, the journal entries corresponding to one or more updates made to one or more logical to physical table (LPT) entries of a LPT; determining, by the controller, whether a current LPT entry, which corresponds to a currently examined journal entry, has already been updated; and discarding, by the controller, the currently examined journal entry in response to determining that the current LPT entry has already been updated.

Abstract: Performing a calibration of a NAND flash memory block that is in an open state. An open state of the NAND flash memory block is detected, the NAND flash memory block comprising a plurality of memory pages, each of which comprising a plurality of memory cells. A group of pages of the NAND flash memory block being in an open state having comparable characteristics is identified. A calibration of read voltage values to pages of the group of identified pages is performed.

Abstract: A computer-implemented method is provided, which includes: assigning data having a first heat to a first data stream, assigning data having a second heat to a second data stream, determining an anticipated throughput of each of the first and second data streams, assigning a first number of logical erase blocks of non-volatile memory to the first data stream based on the anticipated throughput of the first data stream, and assigning a second number of logical erase blocks of non-volatile memory to the second data stream based on the anticipated throughput of the second data stream. Wear-leveling may be performed on open logical erase blocks, including assigning at least some open of the logical erase blocks to a queue. Open logical erase blocks having health values less than a health value of other logical erase blocks by a predetermined amount may be skipped over during the assigning of open logical erase blocks to the queue.

Abstract: A technique for adapting over-provisioning space in a storage system includes determining one or more workload characteristics in the storage system. Over-provisioning space in the storage system is then adjusted to achieve a target write amplification for the storage system, based on the workload characteristics.