Emc Corporation

Emc IP Holding Company LLC

Dell Products Inc.

Peng Yin

Rong Yu

Stephen Richard Ives

Adi Ofer

Jiahui Wang

Ningdong Li

Dan Aharoni

Yechiel Yochai

Gilad Sade

Orit Levin-Michael

Ran Margalit

Daniel E Rabinovich

Qun Fan

John W Lefferts

Pei-Ching Hwang

Cynthia Burns

Barak Bejerano

Robert Lucey

Daohong Wang

James McGillis

Marik Marshak

Ramesh Doddaiah

Peng Wu

Michael J Scharland

Amnon Naamad

Robert S Mason, Jr

Kuolin Hua

Kunxiu Gao

Jeffrey Wilson

Seema G Pai

Earl Medeiros

Michael Garvey

Christopher Monti

Shao Hu

Gabriel Hershkovitz

Stephen F Modica

Sanjay Narahari

Li Lang

William P Powers, Jr

Timothy Johnson

Manickavasasaham M Senghuden

William N Eagle

Akash Agrawal

Bing Liu

Amit Kumar

Thomas Rogers

Sean Hu

Emulation modules running on a single-board compute node have allocations of cores of a multi-core processor. The emulation modules are configured to donate underutilized cores or available CPU cycle units to a shared pool that can be used by other emulation modules running on the board. Emulation modules that require additional processing resources borrow cores or CPU cycle units from the shared pool. The emulation modules dynamically donate and retract cores or CPU cycle units from the shared pool based on utilization of allocated cores.

Dynamic CPU core sharing

A method comprising: detecting a storage device failure in a storage array, the storage array including a plurality of storage devices, the storage device failure being a failure of one or more of the plurality of storage devices; wherein: the plurality of storage devices are configured to implement a supergroup including a plurality of subgroups; identifying ones of the plurality of subgroups that are affected by the storage device failure; and rebuilding the identified subgroups, wherein: (A) when a count of failed storage device is less than a first threshold, each of the identified subgroups is rebuilt based only on subgroup parity components for the subgroup without the use of supergroup parity components, and (B) when the count of failed storage devices is between the first threshold and the second threshold, each of the identified subgroups is rebuilt based on both subgroup and supergroup parity components.

Hierarchical parity protection groups for efficient data recovery

Backend processes in a storage system are implemented using a plurality of worker threads that are divided into two thread groups: thread group A and thread group B. Threads of thread group A are able to be used to process both Random Read Miss (RRM) workload items and Random Write Miss (RWM) workload items. Threads of thread group B are divided into two thread sub-groups: thread sub-group B1 and thread sub-group B2. Threads of thread sub-group B1 are able to be used to process both RRM and RWM workload items, while threads of thread sub-group B2 are reserved to be used to process only RRM workload items. The size of thread sub-group B2 relative to the size of thread sub-group B1 changes over time (from 0 threads to all threads of thread group B), depending on the duty cycle of the threads of group A.

Optimizing backend workload processing in a storage system

A Random Read Miss (RRM) distribution process monitors execution parameters of first, second, and third emulations of a storage engine, and distributes newly received read operations between the emulations. The RRM distribution process assigns newly received read operations to the first emulation, unless the CPU thread usage of the first emulation or the response time of the first emulation meet a first set of criteria. The RRM distribution process secondarily assigns newly received read operations to the second emulation, unless the CPU thread usage of the second emulation or the response time of the second emulation meet a second set of criteria. The RRM distribution process assigns all other newly received newly received read operations, that are not assigned to the first emulation or to the second emulation, to the third emulation. Distribution of read IOs between the emulations enables the storage engine to increase IOPs while minimizing response time.

Method and apparatus for distributing read operations between emulations in a storage engine

Distributed failover of a failed BE, and failback of the failed BE when it recovers, are described. A separate pool of memory may be reserved for each BE on a storage system. Each such pool may be further divided into sub-pools. When a BE fails, the one or more other active BEs on the storage system may be notified, and each such active BE may attempt to take ownership of one or more sub-pools of the BE and execute the one or more pending write operations of the failed BE. Upon completing execution of the pending write operations in a failed-over sub-pool, the ownership of the sub-pool may be returned to the failed BE. When the failed BE recovers, the recovered BE may attempt to take back control of one or more of its sub-pools.

Distributed failover of a back-end storage director

Techniques for performing configuration may include: determining an initial distribution of logical devices among disk adapters, wherein the initial distribution assigns each logical device a designated disk adapter as a primary active disk adapter servicing I/Os directed to the logical device, wherein the logical devices have storage provisioned from a RAID group including physical storage devices, wherein the disk adapters concurrently issue I/Os to each of the physical storage devices; detecting a configuration change; and responsive to detecting a configuration change, performing first processing. The first processing may include determining, in accordance with the configuration change, a redistribution of the logical devices among the disk adapters; and draining pending I/Os. The configuration change may include adding or removing a logical device provisioned from the RAID group. Performing a first operation on a storage device of the RAID group may include synchronous draining of I/Os of the storage device.

Techniques for multiple directors concurrently servicing I/Os directed to same physical storage devices

A SAN storage node has multiple engines, each including at least one computing node with at least one drive adapter for accessing arrays of drives. All drives of the SAN node are accessible by all drive adapters. The drives are organized into hypers. Each hyper is accessible to all drive adapters. Each drive is temporarily associated with a preferred owner for spin-up. A drive adapter that initializes spin-down of one of the drives takes a spindle lock on the drive being spun-down and prompts all other drive adapters to complete outstanding IOs on the drive being spun-down.

Storage array with N-way active-active backend

A method of distributed management of recovery of multi-controller NVMe drives includes detecting a path failure of a PCIe path from a first storage node to a first controller on the multi-controller NVMe drive, and initially attempting to correct the path failure using a controller level reset. If the controller level reset is unsuccessful, an alternative path to the controller is sought, and if that is unsuccessful a drive level reset operation is coordinated by all storage nodes with controllers executing on the NVMe drive. To coordinate reset of the NVMe drive, one storage node is elected master. Each node (both slave and master) initiates quiescing of IO operations on its respective controller, and after quiescing has completed, initiates shutdown of its respective controller. Once all controllers are shut down, the master initiates reset of the NVMe drive. Timeouts are used to constrain completion of the quiescing and shutdown operations.

System and method for managing recovery of multi-controller NVMe drives

Described are techniques for processing a data operation in a data storage system. A front-end component of the data storage system receives the data operation. In response to receiving the data operation, the front-end component performs first processing. The first processing includes determining whether the data operation is a read operation requesting to read a data portion which results in a cache miss; and if said determining determines that the data operation is a read operation resulting in a cache miss, performing read miss processing. Read miss processing includes sequential stream recognition processing performed by the front-end component to determine whether the data portion is included in a sequential stream.

Techniques for data prefetching

Described are techniques for processing data operations. A read request for first data is received at a data storage system. It is determined whether the read request results in a cache hit whereby the first data is stored in a cache of the data storage system, or whether the read request otherwise results in a cache miss. If the read request results in a cache miss, processing is performed to determine determining whether to perform cacheless read processing or deferred caching processing to service the read request. Determining whether to perform cacheless read processing or deferred caching processing is performed in accordance with criteria including a measurement indicating a level of busyness of a back-end component used to retrieve from physical storage any portion of the first data not currently stored in the cache.

Techniques using cacheless read and deferred caching

Scheduling jobs for a plurality of logical devices associated with physical devices includes assigning a physical run count value and a physical skip count value to each of the physical devices, at each iteration, examining the physical skip count value and the physical run count value for each of the physical devices, and scheduling a number of jobs up to the physical run count value for logical devices associated with a particular one of the physical devices at each iteration corresponding to the physical skip count value for the particular one of the physical devices. The physical skip count value and the physical run count value for a particular one of the physical devices may vary according to a total load of the particular physical device. The total load of the particular physical device may vary according to a number of jobs for all logical devices associated with the particular physical device and a total number of all jobs for all physical devices associated with a particular physical device controller.

Physical based scheduler using run count and skip count values

Scheduling jobs for a plurality of devices includes assigning a run count value to each of the devices, scheduling a number of consecutive jobs up to the run count value, and decrementing the run count value according to the number of consecutive jobs scheduled. The run count value for a particular one of the devices may vary according to a total load of the particular one of the devices. The total load of the particular device may vary according to a number of pending jobs for the particular device and a total number of jobs for all devices coupled to a controller for the devices. The devices may be disk drive devices. Only one read job may be scheduled for a device irrespective of the run count value for the device and devices having pending read jobs may be given precedence over other devices. Multiple read jobs may be scheduled for a device according to the run count value and devices having pending read jobs may be given precedence over other devices. The disk drive devices may be subdivided into a plurality of logical devices.

Consecutive scheduling of jobs for a device using run count values

Data units are prefetched into a cache memory by executing a first prefetch task to prefetch a first prefetch series of data units from off-cache. A first prefetch operation is executed to prefetch and store a first selected set of data units. Decisions are made about the first prefetch task, subsequent prefetch operations and prefetch tasks based on in-cache data units associated with the first prefetch task and on other data units that are read from off-cache but that are not associated with the first prefetch task. A determination is made whether an additional data unit that is read from an off-cache logical location is a first prefetch series member. If so, it is associated with the first prefetch task, and, for decision making, is treated as having been prefetched into cache in accordance with the first prefetch task.

Methods and systems for incorporating sequential stream read requests into prefetch management for data storage having a cache memory

A scheduling method and apparatus for use by a processor that controls storage devices of a data storage system is presented. The method allocates processing time between I/O operations and background operations for predetermined time slots based on an indicator of processor workload.

Management of background copy task for point-in-time copies

Growing community of inventors

South Grafton, MA, United States of America

Peng Yin