Abstract: Techniques are described for performing data storage optimization. A first I/O workload for a first data portion of a first snapshot of a first logical device is tracked. First processing is performed by a data storage optimizer to determine a set of one or more data movement optimizations. The first processing uses the first I/O workload for the first snapshot. The set of one or more data movement optimizations include a first data movement that is any of a promotion to move data included in the first data portion from a first storage tier to a higher performance storage tier and a demotion to move data included in the first data portion from the first storage tier to a lower performance storage tier. The first data movement is performed.

Abstract: Techniques for obtaining consistent read performance are disclosed that may include: receiving measured read I/O (input/output) response times for flash storage devices; and determining, in accordance with a specified allowable variation, whether a first of the measured read I/O response times for a first of the flash storage devices is inconsistent with respect to other ones of the measured read I/O response times. Responsive to determining the first measured read I/O response time is inconsistent first processing may be performed that corrects or alleviates the inconsistency of the first measured read I/O response time. The first processing may include varying the first measured read I/O response time of the first flash storage device by enforcing, for the first flash storage device, a write I/O workload limit a read I/O workload limit and an idle capacity limit. Data portions may be ranked and selected for data movement based on read workload, write workload or idle capacity.

Abstract: A service level is assigned to each application that uses virtual memory. The service level is used to select a type of memory used when paging-in data. The service level is used to select a type of storage used when paging out data. The service level is used to select a page to evict from memory, e.g. based on service level probabilities. The service level is used to select a number of contiguous pages to page-in, e.g. based on a service level scalar. Accesses (hits) to the pages in memory may be tracked, including contiguous pages that are paged-in based on the scalar. Pages with low hit frequency may be evicted. The scalar for an application may be adjusted when at least some of the contiguous pages are infrequently accessed.

Abstract: Data storage optimization techniques determine predicted values for I/O statistics using an ARIMA (auto-regressive integrated moving average) model. The ARIMA model may be used to capture periodic patterns and trends of workload I/O access to predict the future load demand. A current set of I/O statistics is collected for a current time period T. Using the current set and one or more ARIMA models, a predicted set of I/O statistics is determined for a next time period T+1. Each of the ARIMA models is characterized by model parameters including P denoting a number of auto-regressive terms, D denoting a number of nonseasonal difference needed for stationarity, and Q denoting a number of lagged forecast errors of prediction. A data storage optimizer may determine one or more data portions for movement from a current storage tier to a target storage tier using the predicted set of I/O statistics.

Abstract: A method, computer program product, and computing system for determining a load value for each of a plurality of storage targets included within a RAID group, thus defining a plurality of load values. The plurality of load values are compared to determine if the RAID group is load balanced. If the RAID group is not load balanced, data is repositioned between the plurality of storage targets.

Abstract: An external storage resource pool associated with federated tiered storage is associated with at least one performance tier based on evaluated performance of the external storage resource pool. Performance of the external pool may be evaluated in terms of service level expectations or service level objectives. Workload pattern analysis or performance probability curves may be used to classify the external storage resource pool. Workload distribution may be determined by a margin analysis algorithm that matches workloads and storage resource pool performance characteristics.

Abstract: High performance storage resources are proactively deallocated when a storage group exceeds its service level objective. A load score is calculated for each unit of storage allocated from a high performance physical disk group. A minimum load score is determined that is the lowest load score calculated for the storage units allocated from the high performance disk group to the storage group. When the storage group service level objective is exceeded, a demotion threshold is calculated equal to a product of the minimum load score and a demotion range factor. A demotion set is determined made up of storage units allocated from the high performance disk group having load scores less than or equal to the demotion threshold. Host data stored on the demotion set is migrated to storage from a lower performance disk group, and the high performance units of storage in the demotion set are released for re-allocation.

Abstract: Described are techniques for processing service level objectives. A first service level objective specified for a storage group of devices may include a first value denoting a first target level of performance for I/O operations. A second service level objective specified for a first portion of the storage group may include a second value denoting a second target level of performance for I/O operations directed to the first portion. The second value may denote a higher level of performance than the first value. It may be determined whether there is a violation of any of the first service level objective and the second level objective. Responsive to determining there is the violation of any of the first service level objective and the second level objective, one or more data movements in accordance with the violation may be performed.

Abstract: Described are techniques user to determine expected service levels. A first set of one or more data portions are selected from one or more devices forming a first device set. The first set of data portions are stored on physical storage of a physical device set of one or more physical devices having an unknown service level expectation. I/O operations directed to the first set of data portions are received and serviced for a time period. First information is collected that characterizes performance of the physical device set during the time period. In accordance with the first information and using an adaptive learning technique, a first service level expectation for servicing I/Os directed to the physical device set is determined.

Abstract: Described are techniques for controlling performance. A forecast error Et for time period “t” denoting a difference between a predicted performance value Ft of a metric for time period “t” and an actual performance value At of the metric for time period “t” is determined. A tracking signal Tt for time period “t” is determined in accordance with “N” forecast errors. It is determined, in accordance with the tracking signal Tt for time period “t”, whether to modify a current value for alpha denoting a weight used in calculating predicted performance values for the metric. If it is determined to modify the current value for alpha, the current value for alpha is set to a revised value in accordance with the tracking signal Tt for time period “t”. A predicted performance value Ft+1 of the metric for time period “t+1” is determined in accordance with the current value for alpha.

Abstract: Promotion and demotion of extents for each storage pool is determined based on a storage pool state defined by demanded activity and demanded capacity. Promotion and demotion of extents for each storage group is determined based on whether the storage group is performing at SLO, above SLO or below SLO. Events are identified and stored in local records using hash tables. The locally stored events are used to update a database of historical extent activity. A moving average of extent activity is used to calculate the demanded activity.

Abstract: Techniques for grouping data portions are disclosed. Each group includes data portions determined to exhibit similar behavior. The techniques may include determining whether an affinity measurement with respect to two groups exceeds an affinity threshold; merging the two groups into a single group responsive to the affinity measurement exceeding the affinity threshold; modeling movement of at least one data portion of the single group between two storage tiers at a particular time of day using predicted workload metrics; and performing the data movement of the at least one data portion between the two storage tiers. Predicted workload metrics may be determined by revising first modeled workload metrics using a bias value, where bias values are associated with different times of day, and the bias value is selected based on the particular time of day that the predicted workload metrics are modeling.

Abstract: Techniques are described data storage optimization that determine predicted values for I/O statistics using an ARIMA (auto-regressive integrated moving average) model. The ARIMA model may be used to capture periodic patterns and trends of workload I/O access to predict the future load demand. A current set of I/O statistics is collected for a current time period T. Using the current set and one or more ARIMA models, a predicted set of I/O statistics is determined for a next time period T+1. Each of the ARIMA models is characterized by model parameters including P denoting a number of auto-regressive terms, D denoting a number of nonseasonal difference needed for stationarity, and Q denoting a number of lagged forecast errors of prediction. A data storage optimizer may determine one or more data portions for movement from a current storage tier to a target storage tier using the predicted set of I/O statistics.

Abstract: An efficient linear technique is used to determine allocation of tiered storage resources among data extents based on system performance and SLOs. Efficiency is achieved by first determining a system performance boundary condition via hardware performance modeling under desirable system performance zones. SLOs are then balanced and SLO achievement improved by exchanging workload activities among SG donors and SG receivers while system performance boundary conditions are maintained. Remaining unutilized capacity is the uniformly distributed to further improve SLO achievement.

Abstract: An intrusion detection system collects architectural level events from a Virtual Machine Monitor where the collected events represent operation of a corresponding Virtual Machine. The events are consolidated into features that are compared with features from a known normal operating system. If an amount of any differences between the collected features and the normal features exceeds a threshold value, a compromised Virtual Machine may be indicated. The comparison thresholds are determined by training on normal and abnormal systems and analyzing the collected events with machine learning algorithms to arrive at a model of normal operation.

Abstract: An intrusion detection system collects architectural level events from a Virtual Machine Monitor where the collected events represent operation of a corresponding Virtual Machine. The events are consolidated into features that are compared with features from a known normal operating system. If an amount of any differences between the collected features and the normal features exceeds a threshold value, a compromised Virtual Machine may be indicated. The comparison thresholds are determined by training on normal and abnormal systems and analyzing the collected events with machine learning algorithms to arrive at a model of normal operation.