Hewlett Packard Enterprise Development LP

Nimble Storage, Inc.

Murali Krishna Vishnumolakala

Anil Nanduri

Chunqi Han

Umesh Maheshwari

Hy Dinh Vu

Ying Hu

Yihong Xu

In some examples, a system allocates a plurality of partitions of a shared storage to respective data services. Based on respective utilizations of the plurality of partitions, the system selects, for a given data service of the data services, between a global storage apportionment process to rebalance shares of the shared storage among the data services, and a local storage apportionment process, where the rebalancing includes releasing a partition of the given data service, and the local storage apportionment process comprising freeing segments within the partition of the given data service.

Shared storage allocation among data services

According to examples, a storage node may include storage devices and a controller that may determine whether all of a plurality of data chunks of a first intra-node portion of a stripe have been stored on the storage node. Based on a determination that all of the data chunks have been stored, a first intra-node parity chunk may be stored at a second one of the storage devices, in which the first intra-node parity chunk may be determined from at least one of the data chunks of the first intra-node portion. Based on a determination that at least one of the data chunks has not been stored, storage of a first intra-node parity chunk of the stripe on the storage node may be delayed until a determination is made that all of the data chunks of the first intra-node portion have been stored at the storage node.

Storing data in a distributed storage system

The failure of a storage unit in a storage array of a storage system may render the storage unit unresponsive to any requests. Any writes to the storage system that occur during the failure of the storage unit will not be reflected on the failed unit, rendering some of the failed unit's data stale. Assuming the failed unit's data is not corrupted but is just stale, a partial rebuild may be performed on the failed unit, selectively reconstructing only data that is needed to replace the stale data. Described herein are techniques for storing information that identifies the data that needs to be rebuilt. When the storage unit fails, the segment identifier associated with the last data segment written to the storage system may be stored. Upon the storage unit recovering, the storage system can rebuild only those data segments whose identifier is greater than the stored segment identifier.

Methods and systems for storing information that facilitates the reconstruction of at least some of the contents of a storage unit on a storage system

Described herein are techniques for storing data in a redundant manner on a plurality of storage units of a storage system. While all of the storage units are operating without failure, only error-correction blocks are stored on a first one of the storage units, while a combination of data blocks and error-correction blocks are stored on a second one of the storage units. Upon failure of the second storage unit, one or more data blocks and one or more error-correction blocks formerly stored on the second storage unit are reconstructed, and the one or more reconstructed data blocks and the one or more reconstructed error-correction blocks are stored on the first storage unit.

Methods and systems for rebuilding data subsequent to the failure of a storage unit

In an all-flash storage array, write requests can take about 9 to 10 times longer than a read request of the same size. There could be several problems when reading or writing from all-flash storage, such as a large write request slowing down small read requests, or other write requests. Also, a large read request may slow down smaller read requests by filling the incoming requests queue. In one implementation, a determination is made on what is the maximum size of a request to flash storage that improves the throughput of a flash chip (e.g., write requests beyond a certain size do not improve throughput). A chunklet is defined as a block of data having the calculated maximum size. As write requests come in, the write requests are broken into chunklets, and then the chunklets are queued for processing by the flash chip. One chunklet is processed at a time per write request. This way, one write request does not monopolize the use of the flash chip for a period of time, allowing other requests to be queued while the chunklet is being processed by the all-flash storage.

Segmenting read requests and interleaving segmented read and write requests to reduce latency and maximize throughput in a flash storage device

Interleaving read and write requests to reduce latency and maximize throughput in a flash storage device

Described herein are techniques for cancelling I/O requests. Initially, virtual memory of an application is assigned to a first portion of memory. The application may issue a read request to an external device. The external device is instructed to record any response to the read request in the first portion of memory. The read request may be cancelled as follows. The virtual memory of the application may be re-assigned to a second portion of the memory. If and when the external device finishes processing the read request, the external device's response to the read request may still be saved in the first portion of memory, even though the read request has been cancelled. Such action of the external device would ordinarily corrupt the virtual memory of the application, but due to the memory re-assignment, no corruption of the virtual memory occurs. Similar techniques may be applied to cancel write requests.

Growing community of inventors

San Jose, CA, United States of America

Murali Krishna Vishnumolakala