I’ve just finished to read this article from Chris Mellor about object storage and I would like to add my 2 cents to this conversation.
A brief market recap
If you take a look at Object storage industry today you can easily find three kinds of products:
- good for high volumes and ISPs: great scalability, throughput, reliability, based on standard X86 HW.
- good for enterprises: less scalability and throughput but with good end-to-end integration from single (primary) vendors.
- good for nothing: the ones that can’t succeed because out-of-scope with end users needs or without a clear roadmap.
Splitting the market in these segments helps to find the players, the products and even their actual results.
Moreover, from another point of view, If object storage was comparable to block storage it would be divided in three segments: high-end, midrange and low end. Where High-end is for big CSP/ISPs, midrange is for small SPs and Large enterprises while low end is for medium sized enterprises… and yes, you can easily see a remarkably shift here: the new high end is no more the large enterprise but the CSP and the small enterprise doesn’t need to deploy object storage locally.
Moreover, on the contrary of block storage, the market is turned upside down: the number of object storage “boxes” (or even better Petabytes) delivered to smaller enterprises are less than the number of boxes delivered to the high end segment of the market!
Use cases
Market adoption of storage is driven by different factors and it strongly depends on use cases.
For example, some Italian service providers adopted Object storage for storing customers’ emails (millions of mailboxes) and then they used the same platform to build other services (e.g.: Dropobox-like and S3-like offerings).
On the enterprise side, I have direct experience with end users (thousands of work seats) that are evaluating these platforms to consolidate file servers, dispersed all around the world, in a single big central repository with the goal to deliver file and sync&share services in a private cloud-like fashion!
A horizontal platform
In practice you can look at Object storage as a horizontal platform capable of managing many different data types.
If the first driver for object storage is the use case (or application) the quantity of managed data comes right after it. An example? Last week, during Storage Field Day 3, Cleversafe showed a success story about a customer deploying a 70PB system that is targeting 150PB. Some Object Storage vendors are also providing HDFS interfaces on top of their storage solution.
At the same time, large enterprises don’t have tens of PBs but they have many types of data to manage: ordinary storage solutions are good to solve one (or two?) problem at a time adding complexity to the infrastructure while object storage could be seen as a single backend platform.
It’s all about gateways…
If it is true that object storage is a horizontal platform to manage data, it’s also true that the end user has vertical problems to solve! Different problems mean different approaches, protocols and solutions. APIs aren’t an out-of-the-box solution while gateways, especially in the midrange segment of this market, are the only way to go!
In other words, APIs look sexy to CSPs but ugly to enterprises… this kind of end users are used to listen terms like “unified” or “converged” and they don’t want to change: they can change storage but they can’t change the way they operate it!
Once again, similarities with the general purpose disks continue to arise: midrange solutions are the richest in terms of features and functionalities (providing the highest number of gateways or third party certifications), while the high end ones are more focused on a single protocol (APIs in this case).
..and TCO
The third adoption driver is TCO (and value of data). I’ve already talked about value of data stored in an object storage system and I’ve already mentioned that object storage could be a good solution for Big Data needs.
The cost of storing data differs from the cost of managing and maintaing it, even if the data is cold and never accessed!
Backups, archiving, versioning, retention, disaster recovery, compliance (only to name a few) actively contribute to the TCO of your data. Object storage platforms are policy based and have many embedded functionalities to provide most of these capabilities without needing any user intervention.
Bottom line
Object storage is gaining relevance from many points of view, but in different forms!
CSPs are deploying PBs and PBs of object storage and Amazon’s S3 is growing exponentially.
The number of customers here is very limited but I don’t think that we should count the customers but rather the volume. If Nimble is the most successful enterprise storage startup ever in terms of systems sold, Object storage startups are outpacing it with a few deployments in terms of installed storage!
These customers buy object storage and they want APIs to build their solutions on top of it.
Large enterprises and small CSPs are pretty new to this market. They are seeing the exponential growth of their unstructured data and different ways to access them right now! They are sick of traditional storage because they would like it cheaper, easier, more manageable and more scalable! Primary vendors are reading their solutions right now and sooner or later we will see more competition in this space.
These customers are buying (or will buy) “concealed object storage” (object storage + gateway + integrations) only if it is comparable to their traditional storage and if it will give them the opportunity to solve more problems!
SMB are not relevant for this market… or it could be better to say that they are contributing in a different way.
They aren’t buying on premise Object Storage (it’s hard to say if solutions like Exablox will change it but, in any case, it won’t be soon). These kind of customers are buying the full service (e.g.: Dropbox is stored on S3), or they buy a gateway to have access to a back-end object storage service… they are somehow contributing to the success of this market when buying from CSPs.
Coming back to Chris’ article, now it’s easy to analyze what is happening to object storage… isn’t it?
Interesting, Enrico. Here’s one expert on how to build your own object store: http://www.datacenteracceleration.com/author.asp?section_id=2559&doc_id=262371
Enrico great perspective and spot on.
However, a variation as it’s not always about the gateways with object storage platforms/repositories/services (e.g. some don’t have gateways) is it is about the personalities that enable the business/application use case. This is something that I touched on in Cloud and Virtual Data Storage Networking (CRC Press).
On the surface personalities might cause people to jump to protocols, interfaces (e.g. block, file, API, etc. However, it is also, about how those protocols and interfaces are used or supported for doing different things.
For example just because something has an NFS or CIFS access method does not mean it is good for general file sharing as its personality might be optimized for backup/restore or archive or some other function.
In addition, as noted, some object platforms/services/solutions have different access methods more closely integrated (e.g. if you prefer to call that an internal gateway ok) while others have discrete ad-ons for doing different things.
However getting to what I think your point is and if so I concur is that the value to the business of object storage platforms has been in the past and will remain in the future is interoperability and coexistence with existing applications. Like it or not, probably the best example of that has been EMC Centera (yes some of the object purist may assert that is not object however lets sidestep object obstacles for now ;).
If you recall EMC made Centera compatible with existing applications and customer environments with APIs as well as NFS/gateways etc, thus usable.
I am seeing some similar with the current generation of products where some vendors are doing more to make their solutions coexist and compatible with the customers’ business/applications vs. the other way around. By providing the customer with option of S3 or other API along with programmatic bindings, along with IOS, JSON, HTTP, XAM, CDMI, NFS, CIFS, DICOM, iSCSI or FCoE, HDFS, etc etc etc the object solutions become opportunities vs. creating object obstacles.
I hope that we see and hear more around how the platforms can be accessed and used. That and their ability to take on or assume various personalities to meet diverse business/application needs. If that occurs result will be enabling object opportunities vs. what is occurring in some instances now, creating object obstacles. Likewise, do we need to wait for a compromise by committee of vendors, industry consortiums, or special interest groups to create a new access methods, protocol, personalities or API? tough to say however in the meantime there are those that exist and usable from HDFS, iSCSI, IOS, NFS, CIFS, HTTP, S3, CDMI and so forth.
Btw, if you do not mind Enrico, will post a link to this post from my http://objectstoragecenter.com page.
Hope all is well, ok, nuff said (for now).
Cheers
GS @storageio
Greg,
Thank you very much for the insightful comment!
Enrico
This is a very good article, Enrico. You capture something that seems to allude many: understanding the difference and purpose between a gateway and a platform. Nice job.
Bob
Bob,
Thank you for the nice comment.