Gartner Maintains Some Confusion on Primary Storage

We read a recent document published by Gartner and we have to admit that we were disappointed. The recognized analyst firm continues to create confusion and promote wrong information. We refer here to the report published April 26, 2023 entitled Voice of the Customer for Primary Storage (ID G00752726) you can find on various places on Internet with licensed reprint versions to vendors or directly from Gartner web site. This time, again, is about primary storage and Gartner gave its own definition but this one is just not accurate. How can such a big analyst firm say such things?

So to address this significant error, let me first copy here their definition of primary storage given at the first chapter of the document: “Gartner defines the primary storage market as vendors that offer dedicated products or product lines for solid-state arrays (SSAs) or hybrid storage arrays, or both, as well as software-defined storage (SDS) software.” It continues with some considerations on the media itself and also claims that “A primary storage product’s foremost purpose is to support response time and IO/s sensitive structured data workloads.”.

So what is the right definition of primary storage?
The first key aspect of this definition is, it is completely independent of the media, the access methods, protocols or nature of the data. It is just about the role of the storage within an organization. When Gartner insists on structured data, SSA or response time or IO/s, it just misses the point and create confusion on the market as finally primary storage equals block storage for the analyst firm. This report should be called block storage or primary block storage or something like that. So you get the point.

As said the definition is just about answering the question about the role of the storage. We have to understand where data is generated, where data is copied… The storage where data is generated, we mean created, instantiated… is THE primary storage, this storage entity stores production data that supports the business, the mission or the activity of the organization. It could be block storage, file storage or object storage. It means implicitly that that storage environment exposes a block, file or object interface such as SCSI, NVMe, NFS, SMB or S3 among others.

Now having said that, secondary storage is defined as a destination where copy data is stored in order to support IT and not the business. Data is copied from primary to secondary storage for protection, data is mirrored, replicated, backup-ed to this target. Data should be recovered, copied back, restored… from secondary storage to primary storage to enable the business to continue or restart depending on the process itself. Again this is about the role of the storage, not access methods or media, we can find block, file and object storage here with HDD or SSD.

Use cases also are important to select storage technologies. And it’s true, block storage is deployed with structured data such as databases, unstructured data on file or object storage. And then you can consider various vertical applications with these basic considerations.

We also need to make a clarification about block storage. The vast majority of applications and databases rely on a file system but here, a disk file system, on top of a block device. Only a few databases support native or raw devices and it is more a historical approach, some of us remember this, when Oracle or Sybase were deployed on raw devices, without any disk file system. But again today, it is very very common that a file system is involved in the storage stack. Even with a file system deployed, I should say a disk file system, we don’t call it file storage as this term is reserved for (network) file server or NAS, we all name this model a block storage as it refers to the exchange between a localhost and its storage connected units based on SCSI, NVMe…

For file server or NAS, this is a disk file system on top of a block device exposed on the network with a network file access protocol such as NFS or SMB consumed by other client machines. We even find today some file servers without any disk file system. File servers and NAS belong to the file storage category as the exchange between the client and the server is based on a file-oriented protocol and not a block one. We can potentially say the same thing for object storage even if we find some implementation with a S3 stack on top of raw device but also several of them with a S3 layer on top of disk file system itself on top of raw device. Sure at the end, everything lands on a HDD or SSD and is written as block or page depending on the vocabulary used with the media technology. But again flash and NVMe have invited several companies to design a new native storage layer directly on top of these units.

This also is why vendors choose to offer unified storage as a consolidation of various independent solutions. These offer a mix of interfaces exposed to applications machines with a large repository.

Let’s wait and see if Gartner will adapt and change their definition.