NVRAM data organization using self-describing entities for predictable recovery after power-loss

摘要

In one embodiment, a node coupled to a plurality of solid state drives (SSDs) executes a storage input/output (I/O) stack having a plurality of layers. Write data associated with one or more write requests to the SSDs is stored in a volatile log. The write data is organized into one or more extents that are copied to the SSDs. The volatile log has a front-end and a set of records with metadata. The metadata includes a head offset referencing an initial record and a tail offset referencing a final record. A portion of the one or more write requests including the write data is copied to a non-volatile log maintained in a non-volatile random access memory (NVRAM). The front-end and the set of records from the volatile log are copied, but the head offset and the tail offset are not, to reduce an amount of metadata copied to the NVRAM.

主权项

1. A system comprising:
a central processing unit (CPU) of a node coupled to a plurality of solid state drives (SSDs); a memory coupled to the CPU and configured to store a storage input/output (I/O) stack having a plurality of layers including a persistence layer executable by the CPU, a portion of the memory configured as a volatile log to store write data associated with one or more write requests, the persistence layer configured to organize the write data into one or more extents that are copied to the SSDs, the volatile log organized as a data structure having a front-end and a set of records with metadata, the metadata including a head offset referencing an initial record and a tail offset referencing a final record; and a non-volatile random access memory (NVRAM) coupled to the CPU and including a non-volatile log configured to store the one or more write requests including the write data, the non-volatile log further configured to store the front-end and the set of records when copied from the volatile log, without storing the head offset and the tail offset, to eliminate copying of the head offset and the tail offset to reduce an amount of metadata copied to the NVRAM.