With vendors beginning to offer storage virtualization features that obscure object storage and make it look like...
other storage types, one has to question whether object storage can become a "one-size-fits-all" choice for all of an organization's storage requirements. If object storage devices can be configured to emulate NAS storage, for example, is NAS storage still necessary?
Although object storage devices may eventually become a standard storage type for all workloads, it is not yet going to be practical for most organizations to use object storage exclusively. There are several reasons for this.
One of the primary reasons why we probably won't soon see many organizations going all in on enterprise object storage is economics. The object storage virtualization products that are available today are based on proprietary hardware -- although some vendors do allow the use of drives from any vendor. Given the investment that most organizations have already made in storage, it probably doesn't make sense for an organization to rip and replace existing storage hardware in favor of proprietary object storage. Instead, those organizations that are interested in transitioning to an object storage devices will probably do so gradually, in accordance with their hardware lifecycle roadmap and available budget.
Another factor that may hold organizations back from using enterprise object storage as a universal storage platform is that of performance. Traditionally, object storage devices have been used primarily as high-capacity storage for unstructured data. It is this high capacity that makes object storage such a popular choice for public cloud-based file storage -- Microsoft Azure BLOB storage and Amazon's S3, or Simple Storage Service, for example, as based on object storage.
Object storage is also known for being inexpensive, with a very low cost per gigabyte of data. While scale certainly contributes to this low cost per gigabyte, part of the reason why object storage tends to be inexpensive is because of its historical reliance on commodity drives. Conventional wisdom has long held that if object storage is being used primarily for data archiving, cloud storage and off-site data replication, then storage capacity and the cost per gigabyte become far more important than performance. Hence, object storage architecture is commonly regarded as a poor choice for performance-dependent workloads.
In spite of the fact that object storage devices have a reputation for being slow, vendors are beginning to improve object storage performance. There are, for example, object storage products that are based solely on flash storage. There are also tiered storage options in which recently accessed data is placed on a high-speed flash tier, while less frequently accessed data resides on commodity storage. Such solutions can provide flash-like performance at a fraction of the cost.
Regardless of whether an object storage architecture implementation is all flash, or tiered, there is still a significant Capex cost associated with the storage. This is another reason why the adoption of object storage-based implementations will be largely driven by economics.
A third reason why organizations are unlikely to use object storage as a universal storage platform is that object storage devices are best suited to storing unstructured data. The reason for this has to do with the way that data attributes are stored. A traditional file system such as NTFS stores file attributes within the file system. For each file, the file system stores a date and time stamp, a set of attributes, an access control list and other information. In contrast, object storage is database driven. File attributes are stored in a database to make indexing more efficient -- although there are some object storage vendors that store attributes with files rather than within a database. Remember, object storage can achieve a massive scale, and part of the reason why this scalability is possible is because database queries make it easy to locate data.
As previously mentioned, object storage vendors commonly recommend using object storage for unstructured data. Structured data contains its own indexes, and tends to be better suited to block storage. Although there are ways of placing structured data on top of object storage, doing so creates excessive overhead because the storage and the structured data both perform indexing. When you also factor in the fact that non-flash object storage tends to be slow, it becomes increasingly clear that object storage is a poor choice for storing structured data.
The virtualization of object storage in a way that allows for the emulation of other storage types will no doubt drive enterprise adoption of object storage. Even so, it seems unlikely that object storage architecture will completely replace other storage types in the near future.
Object storage in the mix of NAS replacement candidates
Scale-out NAS or object storage for unstructured data?
Vendors say object storage is ready to take over for NAS