Abstract: A method for use in a storage system, comprising: detecting an event; and in response to the event, changing one or more replacement algorithms that are being used by the storage system, wherein changing the one or more replacement algorithms includes: (-)selecting a first segment in a cache of the storage system, the first segment being serviced with a first replacement algorithm; (-)generating a first signature representing a state of the storage system; (-)classifying the first signature with a selection engine to identify a second replacement algorithm, the second replacement algorithm being different from the first replacement algorithm; and (-)configurating the storage system to service the first segment with the second replacement algorithm.

Abstract: Aspects of the present disclosure relate to optimizing atomic operations in storage arrays by selectively enabling fabric-less mode based on device access patterns. In embodiments, whether multi-path software is enabled is determined, and device-level access control mappings are analyzed to identify candidates for fabric-less mode operation. For devices without multi-path software, a fabric-less mode is enabled when mapped to a single board, allowing atomic operations via software locks instead of fabric hardware. When multi-path software is enabled, the embodiments either leverage ALUA states to identify optimized paths or employ time series analysis using ARIMA models to predict IO access patterns. Fabric-less mode is activated when forecasting indicates localized access patterns, significantly improving response time and IOPS by avoiding fabric bandwidth constraints and hardware vendor limitations.

Abstract: One or more aspects of the present disclosure relate to dynamic battery-based cache size management. In embodiments, a charge level of a battery configured to provide power to a storage array is monitored. Further, a size of a mirrored write cache partition of system memory in the storage array is dynamically adjusted based on the charge level of the battery.

Abstract: An intelligent predictive battery replacement system to increase disaster recovery stability uses training samples created using battery report indexes and vault condition recovery reports to train a linear regression model learning process to learn a recursion between battery post recovery charge and a set of time series battery operational parameters. Once trained, the learning process is used in a predictive manner to predict the post recovery charge state of batteries deployed in storage controllers, to provide a predictive per-battery risk assessment. The per-battery risk assessment identify batteries that may be scheduled to be replaced to increase disaster recovery stability of the storage systems.

Abstract: A method, comprising: receiving write requests at a first storage system; destaging the write requests to a permanent storage of the first storage system and collecting one or more destage statistics that are associated with the write requests; generating a hint object corresponding to the write requests, the hint object being generated based at least in part on the destage statistics; transmitting the hint object to a second storage system; and transmitting at least some of the write requests to the second storage system for remote replication.

Abstract: One or more aspects of the present disclosure relate to optimizing the rebuild process of a persistent storage device in a storage array is disclosed. The embodiments detect rebuild events, identify affected back-end slices, and prioritize the rebuild order based on a calculated priority score for each slice. This score is derived from service level objectives (SLO) and input/output (IO) statistics of corresponding front-end logical tracks. The embodiments can generate SLO slice objects representing back-end slices, group them in a shared memory database, and update scores during write operations. Rebuild job queues with different priority levels are established, and back-end slices are queued based on their priority scores. This approach ensures efficient rebuilding of critical data, considering both SLOs and real-time IO statistics, thus minimizing performance degradation and enhancing overall system reliability.

Abstract: In some implementations, a control device may receive a plurality of front-end (FE) write pending (WP) tracks. The control device may cluster the plurality of FE WP tracks into one or more FE extent objects using spatial correlations. The control device may add the one or more FE extent objects to a tree data structure. The control device may form back-end (BE) slices using a mapping of the one or more FE extent objects from the tree data structure and to the BE slices based on an aging threshold.

Abstract: A system can obtain a first prompt as output from inputting an alert about a computer system to a first retrieval-augmented generation system (RAG). The system can obtain a first answer as output from inputting the first prompt to a first large language model (LLM). The system can obtain a value maintained by an entity associated with the computing system as output from inputting the alert to a second RAG. The system can obtain a second answer as output from inputting the first answer, the value, and a second prompt to a second LLM, wherein the second LLM comprises the first LLM or another LLM different from the first LLM. The system can obtain a third answer as output from inputting the second answer, user information associated with the entity, and a third prompt to a third LLM. The system can make the third answer available to the entity.

Abstract: One or more aspects of the present disclosure relate to enhancing storage device performance and longevity. In embodiments, Self-Monitoring, Analysis, and Reporting Technology (SMART) data collected from storage devices can be analyzed using one or more machine learning models. Additionally, the embodiments can employ Principal Component Analysis (PCA) to reduce data dimensionality and K-means clustering to group storage devices with similar characteristics. The embodiments can also predict potential thermal throttling events and proactively adjust read/write rates to prevent performance degradation. Optionally, the embodiments can transfer highly active data from hot to cooler devices. For example, the embodiments can control logical-to-physical track mapping based on SMART data and direct data to physical tracks accordingly. By correlating SMART data with input/output (IO) workload analysis, the embodiments can predict when device temperatures will reach thermal thresholds and take preventive actions.

Abstract: Methods and apparatuses are provided for utilizing additional and available system resources to cache one or more incoming I/O write requests. A storage processor may implement a policy including a Backend as a Service (BEaaS) to use a node pair backend, a fabric network, and other resources to absorb incoming bursts of I/O write requests from a host and direct the requests temporarily to NVMe storage devices. The policy may further use local and/or remote storage arrays connected over the fabric to alleviate strain on local processing resources. The policy may provide for using cut-through writes, including for example, dual-cast operations over PCIe connections, to send I/O requests directly to the NVMe storage devices.

Abstract: A method for adjusting data processing unit (DPU) utilization of a storage array, the method includes sampling input-output (IO) of the storage array, forecasting, based on the sampling, demand for the storage array, where the storage array include a DPU core matrix, and making a first determination, based on the forecasting, that at least one service level (SL) of multiple SLs associated with the DPU core matrix is out of compliance. The method further includes identifying, based on the first determination, at least one DPU of the DPU core matrix currently operating at a first SL, and adjusting, based on the first determination, operation of the at least one DPU to operate from the first SL to a second SL of the multiple SLs.

Abstract: One or more aspects of the present disclosure relate to managing data reduction in a storage system during burst workloads. In embodiments, input/output (IO) operations received by a storage array are monitored to identify a burst workload based on a predefined threshold of IO operations per second (IOPS). In addition, the compression of data corresponding to IO write operations in the burst workload is deferred during the identified burst workload by temporarily storing the data in an uncompressed format. Further, the data is compressed during a self-healing period after a delay period. For example, the delay period is after the identified burst workload.

Abstract: One or more aspects of the present disclosure relate to dynamic compression engine management. In embodiments, statistics corresponding to an input/output (IO) workload received by a storage array are collected. Additionally, statistics corresponding to one or more compression cards of the storage array are collected. Further, one or more compression engines within the one or more compression cards of the storage array are dynamically activated or deactivated based on the IO workload and compression hardware statistics.

Abstract: A method, comprising: identifying an extent corresponding to a metadata page that is currently stored in a cache; calculating a predicted temperature score for the extent by using a time series forecasting model; detecting whether the predicted temperature score exceeds a first threshold; and extending a stay of the metadata page in cache in response to detecting that the predicted temperature score exceeds the first threshold, wherein the predicted temperature score is a measure of respective frequencies at which at least two different types of input-output I/O operations are expected to be received for the extent during a future time window.

Abstract: A method, comprising: executing an embedded malware detector, the embedded malware detector being executed inside a first guest operating system, the first guest operating system being executed on a storage processor that is part of a storage system, the storage processor being configured to execute one or more second guest operating systems in addition to the first guest operating system, each of the second guest operating systems being arranged to execute software for reading and/or writing data to one or more storage devices that are provided in the storage system; identifying, by the embedded malware detector, a given one of the storage devices; obtaining, by the embedded malware detector, a snapshot of the given one of the storage devices; mounting, by the embedded malware detector, the snapshot in the first guest operating system; scanning the mounted snapshot for malware.

Abstract: A method for managing data packing in a storage includes: receiving data from a computing device at a first point-in-time; analyzing the data to: determine characteristics of the data, and generate an extent for the data, in which the extent includes sequential tracks; analyzing sequential tracks to infer how a distribution of compression sizes change for the sequential tracks to obtain track-level in the compression sizes; forecasting, based on the characteristics of the data and the track-level changes in the compression sizes, a stability score of the distribution at a second point-in-time, in which the second point-in-time is after the first point-in-time; making a determination, based on the stability score, that the distribution would be the same in the second point-in-time; and placing, based on the determination, the tracks to a first zone in the storage, in which the storage further comprises a second zone and a third zone.

Abstract: An information handling system may include at least one processor and a network interface adapter. The information handling system may be configured to: couple to a plurality of client systems via the network interface adapter; implement a quality of service (QoS) policy for each client system; adjust the QoS policies based on predictions regarding resource utilization for resources of the information handling system and input/output (I/O) demands associated with each client system during specified time windows; and service requests from the client systems in accordance with the adjusted QoS policies.

Abstract: Systems and methods for time-series based machine learning anomaly detection and prevention are described. In an illustrative, non-limiting embodiment, an Information Handling System (IHS) may include: a processor; and a memory coupled to the processor, where the memory includes program instructions store thereon that, upon execution by the processor, cause the IHS to: obtain communication data associated with the IHS for a plurality of time windows, including a particular time window, and previous time windows before the particular time window; determine, using a machine learning model, that the communication data for the particular time window includes an anomaly; and based on the determination, perform one or more actions. In some embodiments, the program instructions further cause the IHS to: based on the communication data, determine time-series data for a plurality of attributes of the communication data; and determine that an attribute includes an outlier in the particular time window.

Abstract: In a data storage system, the storage capacity of a partition of volatile memory configured to store write-pending data is only increased after increasing the charge state of a backup power battery to protect the additional storage capacity. A forecast of needed storage capacity is computed for the partition. If the forecast storage capacity is greater than a predetermined percentage of current storage capacity, then the charge state of the battery is increased by a fixed amount relative to full charge. The storage capacity protected by the charge state of the battery inclusive of the charge state increase is computed and the storage capacity of the partition is increased to the computed amount. The charge state of the battery and the storage capacity of the partition are increased iteratively to achieve a target storage capacity such that the forecast storage capacity is not greater than the predetermined percentage of the target storage capacity.

Abstract: One or more aspects of the present disclosure relate to paging metadata into memory. In embodiments, a metadata demand score is calculated for each extent of a storage array based on hit forecasts and read forecasts corresponding to input/output (IO) operations targeting each extent. Further, each extent can be ranked based on their respective metadata demand scores. Additionally, metadata can be paged into memory from a storage device of the storage array based on the ranking to reduce metadata page misses. For example, the memory can correspond to a global memory portion of the storage array.