什么是对象存储？

Unlike file storage systems that organize data into a hierarchy, object storage stores data objects in a flat system. These objects then are referenced and retrieved using a Universally Unique Identifier (UUID) or Globally Unique Identifier (GUID) of 128-bit integers, significantly large enough to allow for a wide range of unique IDs, important because of the large number of objects stored.

Also unlike file storage systems, object storage uses metadata linked to data objects via UUIDs to find user requests. Files on the other hand retain very little metadata, relying on the hierarchy to locate requested data. The main advantage, object metadata allows for extensive (unlimited) description of the object, for example, video files can include lists of cast and crew. So, a user can locate video files by actor name.

Hierarchical file storage systems run into difficulties when scaled to capacities ranging in the petabytes. At this size, file storage performance degrades considerably, and the common solution is to split data into logical unit numbers (LUN), or collections of physical or virtual storage devices. While this improves storage performance, it also adds more complexity, until the complexity begins to cause difficulties as well. Instead, metadata search capabilities are used to overcome the scaling of data, and the ensuing technical challenges that crop up in file storage.

Object storage reliability is further enhanced by the technique of Erasure Coding. In essence, Erasure Coding volumes divide data into fragments, or shards, each is packed with error correction data, and then each are placed on different disks. By diversifying the data and packing it with redundant information, it ensures that files can be restored even after many disk failures. In short, Erasure Coding is a modern RAID system.

Object storage use cases tend towards two general kinds, those workloads traditionally fulfilled by file, block, or tape storage devices, and the new innovative solutions that programmatically access object storage to derive insights. Traditional applications include backups, archiving, content storage, enterprise file servers, and collaboration. New innovations include mobile, social, IoT, cloud native apps, and AI/ML or cognitive apps.

• Active Archiving — Active archiving is a space between deep storage archiving and real-time transparent data access, which tend to cause conflicting complications. Active archiving is typically configured as a tier architecture using tape storage behind disk storage, and replication. Object storage is used to flatten the tier architecture while adding resilience and simplifying administration.
• Backup Repositories — Backup data is a practical concern for many organizations, which requires capacity management, space reclamation, and data replication. These data backup practices tended to use costly storage arrays and tape silos. With object storage, resilience features allow organizations to move away from replication and use low cost storage, with the benefit of a seemingly “endless” pool of storage.
• Cognitive and analytic systems — With the advent of AI and machine learning followed the rise of cognitive systems that learn and reason based on human, experience, and environmental interactions. To do this, cognitive systems have been designed to interpret and understand unstructured data, found in images, speech, social media, etc. However, current data storage systems suffer performance issues with the type and scale of data that cognitive systems consume to determine their predictive responses. Object storage provides the metadata architecture that helps to facilitate the performance requirements of these complex and heavy workloads.
• Enterprise Files Services — Enterprise file services have been a necessary tool for organizations for many decades, however, the dependence on RAID schemes using block storage has steadily introduced complexity in large-scale data protection efforts. Object storage removes this older, siloed approach in favor of a unified back-end, with metadata that allows easy access without locking into one solution, effectively achieving economies of scale.
• Internet of Things Data Repository — The Internet of Things refers to the connection of a wide variety of devices that penetrate many aspects of life, each with a unique identifier, to eliminate the human element in gathering data. The amount of data generated by IoT systems tends towards the Big Data size, and storing this data requires a more robust system than what block or file storage can support. Object storage supplies the robust storage back-end where data of all varieties can be stored, retrieved, and archived.

Object storage can be compared to two other common storage formats, block, and file storage. These formats aim to store, organize, and allow access to data in specific ways that benefit certain data applications. For instance, file storage, commonly seen on desktop computers as a file and folder hierarchy, presents information intuitively to users. This intuitive format, though, can hamper operations when data becomes voluminous. Block storage and object storage both help to overcome the scaling of data in their own ways. Block storage does this by “chunking” data into arbitrarily sized data blocks that can be easily managed by software, but provides little data about file contents, leaving that to the application to determine. Object storage decouples the data from the application, using metadata as a file organization method which then allows object stores to span multiple systems, but still be easily located and accessed.