whitehoune - Fotolia

Object-file system combo rapidly expands

Object-based storage adoption took longer than expected, so object vendors have added file systems to their products to make them look more familiar to users.

Object-based storage came to market with a lot of fanfare several years ago with the premise that it overcame the performance limitations of traditional file systems, but now more object storage vendors are combining the two technologies.

Some object-file system examples include:

  • Caringo last year introduced its FileFly for Caringo Swarm, a Windows-based application that plugs into Windows NTFS file system. It replaced the Caringo File Server, a Linux-based archiving tool that provided a Network File System (NFS) mount on top of Swarm. FileFly was marketed as way to migrate data off NetApp and Windows filers to large Swarm object storage repositories.
  • Scality has its Scale-Out-File-System, which the company claims can scale to billions of files compared to traditional file systems that suffer performance degradation when they hit 100,000 files. It's layered over the company's flagship Ring object storage software. Half of the company's revenues come from the file system sales, according to Scality CEO Jerome Lecat. "We built our own file system. It's our own IP," he said. "It's not a gateway. It's completely scale out and there is no node limitation."
  • All-flash pioneer Pure Storage has branched out from its original SAN array system with a scale-out FlashBlade system that uses a common object storage along with a file system, and data reduction, encryption, and erasure coding. FlashBlade's Elasticity software is one code base across object and file protocols, and both layers share a metadata engine.
  • Exablox offers a scale-out, object-based storage appliance. Exablox took an object-file system approach from the start. It works with a scale-out Ring architecture that supports multiple OneBlox appliances that present a single global file system. One Ring contains multiple OneBlox systems that scale to multiple terabytes of data.

Files add familiar interface, latency to objects

Object storage encountered headwind in the adoption process partly because many legacy applications don't talk to the HTTP REST API. That's prompted storage vendors to add file system functions and capabilities to their software and hardware. But the object-file system maneuver comes with a tradeoff in performance.

Marc Staimer, president at Dragon Slayer Consulting, said he sees the object storage market growing faster than external storage. But he doesn't agree with the move of combining object storage and file system functionality.

"They are putting a file system on object storage," Staimer said. "So, you are adding more latency. I don't think it's a smart thing to do because they are not telling me what operational problem they are solving. And it neuters some fantastic capabilities you get out of object storage. What's key is how you integrate with the application. Most backup applications talk to object storage. A lot of the data mover and migration software talk to object storage."

Arun Taneja, founder of Taneja Group, said all the original object storage vendors such as Caringo, Scality, Amplidata and Cleversafe started with pure object storage software and eventually added file system capabilities. Western Digital acquired Amplidata and IBM bought Cleversafe in 2015.

"All were pure object with their own API and some added Amazon S3 [Simple Storage Service] capabilities," Taneja said. "But these guys took quite some time to get any traction in the market. There was quite a lot of interest, but it was not translating into a real growth. It was not creating companies that were big enough to warrant IPO [initial public offering] activity."

Taneja said object vendors are making file interfaces available now to attract customers used to dealing with traditional file storage.

"But it's reasonable to say anytime you have to go through another layer, you are going to take a performance hit," he added.

Objects are much easier to use than files with containers. But the world as we know it today is files.
John Hayesfounder and chief architect, Pure Storage

Caringo's object-file system strategy required building a connector from legacy file and block storage for the object storage environment.

"The main reason is for the existing applications and workloads," said Adrian Herrera, Caringo's vice president of marketing. "That's the main reason. Then you can appeal to a broader audience."

Pure Founder and Chief Architect John Hayes said the idea behind FlashBlade was to combine the emerging object technology with file technology that users understand. FlashBlade, designed to scale to tens of petabytes and billions of files, is in an early access program.

"When you look at object, it's more forward-looking," Hayes said. "We asked, 'What kind of interface do we need for the future?' People say object. I think that's related to containers. Objects are much easier to use than files with containers. But the world as we know it today is files. If [FlashBlade] was object-only, we wouldn't be able to go in and replace existing filers.

"There is tons of software that was written for files," Hayes added. "Designing an object store in a file system, we wanted something optimized more for lots and lots of metadata operations than a typical file system."

Object storage technology is more suited for organizations that are geographically dispersed and need to scale to the petabyte level, particularly for companies depending on Web-based, cloud applications.

Files vs. objects: What's the difference?

Object storage stores a file as a single object that includes metadata and assigns it an ID number so when content is needed, the user just needs to present the ID to the system and the content will be assembled with all the metadata that include authentication and security. It's stored on commodity hardware.

File systems, however, store data to disk in thousands of pieces with each having its own address that becomes part of a hierarchy file path. When the file is needed, the user enters the server name, file directory and file name and the files are reassembled without much metadata.

NAS systems built for files eventually hit a performance bottleneck when the controller experiences I/O limitations. That problem is solved by adding another NAS system to the environment, which leads to complex configurations that need to be managed.

Objects plus file systems can reduce complexity

Liquid Robotics, based in Sunnyvale, Calif., uses Exablox to store telemetry data generated by its Wave Glider ocean robots. Wave Gliders gather data used for meteorology and other purposes from beneath the sea and send it back over Wi-Fi networks or satellites. Liquid robotics has four OneBlox systems to store Wave Glider data.

Olav Philips, senior DevOps engineer at Liquid Robotics, said he was looking for a NAS system that performed well enough but didn't bust his budget. That's when he came across Exablox's object-file system architecture.

"It's a feature-rich NFS server for us," he said of OneBlox. "For day-to-day usage, we just use it as a file system. I know it has some fairly sophisticated features and the idea of implementing object storage is attractive, but we use it just for NFS and some CIFS [Common Internet File System]."

We're in the data business. We don't want to lose any data.
Olav Philipssenior DevOps engineer, Liquid Robotics

Philips said Liquid Robotics uses a Nutanix hyper-converged system for virtual machine storage, but the Wave Gliders required more storage than his Nutanix box could store. Instead of scaling his hyper-converged system by buying an appliance with additional CPU and software, he went looking for a dedicated storage system for the telemetry data.

"Our CPU and memory were fine, we just needed storage," Philips said. "So, instead of buying another [Nutanix] blade, we bought an Exablox because it was more cost effective."

He said he likes the way OneBlox stripes data. He said the price was also right at the level of support he needed.

"We looked at a lot of NAS options, including FreeNAS and NetApp," Philips said. "NetApp is a very expensive box, and the cheaper ones don't necessarily have good support. I'm perfectly capable of building an NFS server, but having support in critical moments was important for us."

Liquid Robotics was an early beta Exablox customer, but did extensive testing before putting a unit into production. "We're in the data business," Philips said. "We don't want to lose any data."

Pendleton Grain Growers in Oregon also uses OneBlox mainly as a file server, but its object capabilities saved the company from a Ransomware scare late last year. Because object storage on OneBlox cannot be changed, IT manager Cary Ford could go back to versions of his files that were not infected and recover from them.

"I had a user call me one morning and [say] he couldn't get into files. I thought he had a virus," Ford said. "All his files that could be shared were locked, even those on Exablox. I wiped the files, did a clean install and deleted all the files on Exablox. Then I copied the files [from earlier snapshots] back to his user share. I had his laptop back to him the next day."

Dave Raffo contributed to this story.

Next Steps

When object vs. file storage makes sense for cloud apps

How object metadata makes smarter storage systems

Scale-out vs. object storage for large data sets

Dig Deeper on Storage management tools