Abstract: A change in workload characteristics detected at one tier of a multi-tiered cache is communicated to another tier of the multi-tiered cache. Multiple caching elements exist at different tiers, and at least one tier includes a cache element that is dynamically resizable. The communicated change in workload characteristics causes the receiving tier to adjust at least one aspect of cache performance in the multi-tiered cache. In one aspect, at least one dynamically resizable element in the multi-tiered cache is resized responsive to the change in workload characteristics.

Abstract: Technology for operating a cache sizing system is disclosed. In various embodiments, the technology monitors input/output (IO) accesses to a storage system within a monitor period; tracks an access map for storage addresses within the storage system during the monitor period; and counts a particular access condition of the IO accesses based on the access map during the monitor period. When sizing a cache of the storage system that enables the storage system to provide a specified level of service, the counting is for computing a working set size (WSS) estimate of the storage system.

Abstract: The techniques introduced here provide for efficient management of storage resources in a modern, dynamic data center through the use of virtual storage appliances. Virtual storage appliances perform storage operations and execute in or as a virtual machine on a hypervisor. A storage management system monitors a storage system to determine whether the storage system is satisfying a service level objective for an application. The storage management system then manages (e.g., instantiates, shuts down, or reconfigures) a virtual storage appliance on a physical server. The virtual storage appliance uses resources of the physical server to meet the storage related needs of the application that the storage system cannot provide. This automatic and dynamic management of virtual storage appliances by the storage management system allows storage systems to quickly react to changing storage needs of applications without requiring expensive excess storage capacity.

Abstract: Method and system for distributing a notification of an event occurring at a one entity to an affected entity. In one embodiment, entities are grouped into one or more groups so that an event occurring at one entity of the group is distributed to the other entities of the group (“affected entities”). If an event occurs at an entity managed by a local entity manager, a notification of the event can be routed to a remote entity manager that manages an affected entity. The local entity manager can use a distributed mapping to determine where to route the event notification. In one embodiment the mapping associates an entity to an entity manager.

Abstract: Graph transformations are used by a data management system to correct violations of service-level objectives (SLOs) in a data center. In one aspect, a process is provided to manage a data center by receiving an indication of a violation of a service-level objective associated with the data center from a server in the data center. A graph representation and a transformations data container are retrieved by the data management system from data storage accessible to the data management system. The transformations data container includes one or more transformations. The transformation is processed to create a mutated graph from a data center representation from the graph representation. An option for managing the data center is determined as a result of evaluating the mutated graphs.

Abstract: A change in workload characteristics detected at one tier of a multi-tiered cache is communicated to another tier of the multi-tiered cache. Multiple caching elements exist at different tiers, and at least one tier includes a cache element that is dynamically resizable. The communicated change in workload characteristics causes the receiving tier to adjust at least one aspect of cache performance in the multi-tiered cache. In one aspect, at least one dynamically resizable element in the multi-tiered cache is resized responsive to the change in workload characteristics.

Abstract: Embodiments of the systems and techniques described here can leverage several insights into the nature of workload access patterns and the working-set behavior to reduce the memory overheads. As a result, various embodiments make it feasible to maintain running estimates of a workload's cacheability in current storage systems with limited resources. For example, some embodiments provide for a method comprising estimating cacheability of a workload based on a first working-set size estimate generated from the workload over a first monitoring interval. Then, based on the cacheability of the workload, a workload cache size can be determined. A cache then can be dynamically allocated (e.g., change, possibly frequently, the cache allocation for the workload when the current allocation and the desired workload cache size differ), within a storage system for example, in accordance with the workload cache size.

Abstract: Collaborative management of shared resources is implemented by a storage server receiving, from a first resource manager, notification of a violation for a service provided by the storage server or device coupled to the storage server. The storage server further receives, from each of a plurality of resource managers, an estimated cost of taking a corrective action to mitigate the violation and selects a corrective action proposed by one of the plurality of resource managers based upon the estimated cost. The storage server directs the resource manager that proposed the selected corrective action to perform the selected corrective action.

Abstract: Embodiments of the systems and techniques described here can leverage several insights into the nature of workload access patterns and the working-set behavior to reduce the memory overheads. As a result, various embodiments make it feasible to maintain running estimates of a workload's cacheability in current storage systems with limited resources. For example, some embodiments provide for a method comprising estimating cacheability of a workload based on a first working-set size estimate generated from the workload over a first monitoring interval. Then, based on the cacheability of the workload, a workload cache size can be determined. A cache then can be dynamically allocated (e.g., change, possibly frequently, the cache allocation for the workload when the current allocation and the desired workload cache size differ), within a storage system for example, in accordance with the workload cache size.

Abstract: Graph transformations are used by a data management system to correct violations of service-level objectives (SLOs) in a data center. In one aspect, a process is provided to manage a data center by receiving an indication of a violation of a service-level objective associated with the data center from a server in the data center. A graph representation and a transformations data container are retrieved by the data management system from data storage accessible to the data management system. The transformations data container includes one or more transformations. The transformation is processed to create a mutated graph from a data center representation from the graph representation. An option for managing the data center is determined as a result of evaluating the mutated graphs.

Abstract: Collaborative management of shared resources is implemented by a storage server receiving, from a first resource manager, notification of a violation for a service provided by the storage server or device coupled to the storage server. The storage server further receives, from each of a plurality of resource managers, an estimated cost of taking a corrective action to mitigate the violation and selects a corrective action proposed by one of the plurality of resource managers based upon the estimated cost. The storage server directs the resource manager that proposed the selected corrective action to perform the selected corrective action.

Abstract: Graph transformations are used by a data management system to correct violations of service-level objectives (SLOs) in a data center. In one aspect, a process is provided to manage a data center by receiving an indication of a violation of a service-level objective associated with the data center from a server in the data center. A graph representation and a transformations data container are retrieved by the data management system from data storage accessible to the data management system. The transformations data container includes one or more transformations. The transformation is processed to create a mutated graph from a data center representation from the graph representation. An option for managing the data center is determined as a result of evaluating the mutated graphs.

Abstract: A service level objective (SLO) violation is detected for a workload of a networked storage system, based on a performance metric not being satisfied for the workload. In response to detecting the SLO violation, a controller determines that changing a level of caching at a node of the networked storage system will improve the performance metric for the workload. The controller implements the change by adjusting an operation of a virtual cache appliance (VCA) of the networked storage system. The adjusting can be instantiating a new VCA, or adjusting the level of caching at an existing VCA. The adjusting can be for caching related to the workload itself, or it can be caching for an intertering workload.

Abstract: Methods, systems, and computer executable instructions for performing distributed data analytics are provided. In one exemplary embodiment, a method of performing a distributed data analytics job includes collecting application-specific information in a processing node assigned to perform a task to identify data necessary to perform the task. The method also includes requesting a chunk of the necessary data from a storage server based on location information indicating one or more locations of the data chunk and prioritizing the request relative to other data requests associated with the job. The method also includes receiving the data chunk from the storage server in response to the request and storing the data chunk in a memory cache of the processing node which uses a same file system as the storage server.

Abstract: Systems and methods for providing resource isolation in a shared computing environment using reinforcement learning (RL) techniques are disclosed. A resource isolation mechanism may be applied in a shared storage system, or database service, that limits the resource utilization of each namespace to its specified allocation. For example, the resource isolation mechanism may be used to limit the I/O utilization of database applications in a shared computing system (e.g., a system supporting a database service) to a specified limit. In such embodiments, RL techniques may be applied to the system to automatically control the rate of queries made by an application. RL techniques, such as those based on the State-Action-Reward-State-Action (SARSA) method may be effective in controlling the I/O utilization of database applications for different workloads. RL techniques may be applied globally by the service, or may be applied to particular subscribers, applications, shared resources, namespaces, or query types.

Abstract: A change in workload characteristics detected at one tier of a multi-tiered cache is communicated to another tier of the multi-tiered cache. Multiple caching elements exist at different tiers, and at least one tier includes a cache element that is dynamically resizable. The communicated change in workload characteristics causes the receiving tier to adjust at least one aspect of cache performance in the multi-tiered cache. In one aspect, at least one dynamically resizable element in the multi-tiered cache is resized responsive to the change in workload characteristics.

Abstract: Collaborative management of shared resources is implemented by a storage server receiving, from a first resource manager, notification of a violation for a service provided by the storage server or device coupled to the storage server. The storage server further receives, from each of a plurality of resource managers, an estimated cost of taking a corrective action to mitigate the violation and selects a corrective action proposed by one of the plurality of resource managers based upon the estimated cost. The storage server directs the resource manager that proposed the selected corrective action to perform the selected corrective action.

Abstract: A hybrid storage device uses a write cache such as a hard disk drive, for example, to cache data to a solid state drive (SSD). Data is logged sequentially to the write cache and later migrated to the SSD. The SSD is a primary storage that stores data permanently. The write cache is a persistent durable cache that may store data of disk write operations temporarily in a log structured fashion. A migration policy may be used to determine how long to cache the data in the write cache before migrating the data to the SDD. The migration policy may be implemented using one or more migration triggers that cause the contents of the write cache to be flushed to the SSD. Migration triggers may include a timeout trigger, a read threshold trigger, and a migration size trigger, for example.

Abstract: A hybrid storage device uses a write cache such as a hard disk drive, for example, to cache data to a solid state drive (SSD). Data is logged sequentially to the write cache and later migrated to the SSD. The SSD is a primary storage that stores data permanently. The write cache is a persistent durable cache that may store data of disk write operations temporarily in a log structured fashion. A migration policy may be used to determine how long to cache the data in the write cache before migrating the data to the SDD. The migration policy may be implemented using one or more migration triggers that cause the contents of the write cache to be flushed to the SSD. Migration triggers may include a timeout trigger, a read threshold trigger, and a migration size trigger, for example.