Object Storage can take on a lot of roles in the enterprise. This flexibility is one the key values of object storage, the ability to serve as storage for a variety of use cases. With most of those use cases the role of object storage emphasizes cost effective, long term data preservation. While access to the data is relatively fast it is not typically thought of as high performance. High performance processing of data is still the purview of Network Attached Storage (NAS) systems. But now several object storage systems come with all flash configurations, with specific tuning to take advantage of a flash only configuration. Does the emergence of all-flash object mean the elimination of NAS systems?
The Bottlenecks of All-Flash Object Storage
The question of how much NAS functionality can an all-flash object store eliminate is largely dependent on how much and what kind of performance a flash-based object storage system can deliver. First, let’s look at the remaining performance bottlenecks – assuming flash drives replace hard disks. The first bottleneck is creation of the object ID. Every file stored on object storage receives an assigned object ID which the system creates via a hash of that file’s contents. In a hard disk world, the application or user was not impacted by the latency of creating the ID because the hard disk itself was slower than the object ID creation process. In a flash world they may, since it, the media responds instantly, and exposes the time required to create the object ID. Of course a single object ID creation is a non-issue, but when the system is creating thousands of object ID while performing other tasks.
Second, one of the most powerful attributes of object storage is its rich metadata capability that enables powerful data management and search. The tagging of objects with this metadata, and the management of it, also takes time. Again in a hard disk world the latency from this process is not noticeable since the hard drive itself is more latent. But in a flash world, it might be noticeable and will cause the flash storage not to perform to its full potential.
The third potential bottleneck is the scale-out nature of object storage systems. Most object stores come from the clustering of a series of interconnected nodes that create a single virtual storage pool, aggregating the capacity of all the storage in each individual node. The network connection between nodes and the management of data placement also add latency. The scale-out latency will exist on a scale-out NAS or Block system too but most of those systems invest in high performance networking to lower latency from inter-node communications. Object storage systems on the other hand have commodity networking components to keep costs down.
It is important to note that the all-flash object storage should still outperform a hard disk based object store, but they will not be able to live up to the full potential of the flash storage within them. There are use cases where a flash object storage system makes sense, specifically analytics.
The Role of NAS
As a result of the remaining bottlenecks, after the installation of flash most data centers should consider a strategy of a small but high performance flash powered NAS front end, with a cost effective object storage system on the back end. Doing so enables a single scale-up NAS to deliver the performance the environment requires without having to create concerns over scalability or cost, since data will move to the object store when not in use. The key is to make the movement between high-performance NAS and object storage transparent and automatic. We discuss how to create this data flow in our webinar “Sizing Up Object Storage for the Enterprise“, now available on-demand.