Press Room

SwiftStack CPO: ‘If you take a filesystem and bolt on an object API’… it’s upside down

Joe Arnold on Ethernet accessed object drives and more

Interview Open source OpenStack-focused, object storage startup SwiftStack has has early involvement with Seagate’s object storage-focussed Kinetic disk drives, the ones needing server-resident software to manage their IO.

We had an email conversation with Joe Arnold, the founder, president and chief product officer of SwiftStack, about this and other matters.

Kinetic drive use is indistinguishable from normal disk drive use for users according to Joe. Here are his views explaining that:

El Reg I’d like to understand how SwiftStack users can use Kinetic (and other direct-addressed key:value disk drives).

Joe Arnold To the user of SwiftStack object storage, using either standard HDDs or Kinetic drives is virtually the same experience. SwiftStack was the first software company involved with Seagate specific to writing to the Kinetic API before it was launched in October 2013 and made open source in May 2014. The SwiftStack product is the only software today that supports management of Kinetic drives in a scale-out storage cluster.

We just got back from the Linux Foundation Kinetic PlugFest. SwiftStack was one of the few who is up and running with Kinetic API drives from Seagate, Toshiba and WD. So now that the device compatibility has been worked through, the next stage is to deliver full solutions. To this end, Seagate and SwiftStack have commercial opportunities, which are in progress.

El Reg Does SwiftStack address raw (non-Kinetic) disk drives or talk to them via file system. If so will Kinetic drive access by SwiftStack be faster than standard HDD access?

Joe Arnold SwiftStack consumes raw block devices (HDDs) and creates a storage system around the pool of storage. Individual Kinetic drives respond to many API commands from Swift in much the same way a SwiftStack node with non-Kinetic disk drives does.

El Reg Why should SwiftStack users choose Kinetic drives rather than standard HDDs?

Joe Arnold SwiftStack believes that users should have freedom of choice. This in includes servers, drives, and even drive technologies. In fact, we architected SwiftStack so that Kinetic and standard drives can be used within the same cluster and namespace. This is a huge enabler for adoption because it means that users can implement this new technology incrementally.

I have written about and presented on Kinetic before – about how the technology makes new storage topologies possible. There are near-term efficiencies which allow the necessary compute and networking to be much more easily tuned for the storage workload. Instead of running on a standard server filled with HDDs – the storage devices can be connected to a network via an enclosure with an embedded switch. This means a smaller footprint of servers can be used to run the storage services.

El Reg What do you think of the idea that object storage is a feature and not a product?

Joe Arnold If you take a filesystem and bolt on an object API, you have the architecture upside down!

Object storage with a unified namespace can serve as a foundation for filesystem access, but not the other way around; filesystems do not infinitely scale. It reminds me of the saying – you can’t turn a sausage into a pig.

Sure, an object gateway can be layered on top of a filesystem to provide an object API. But the API isn’t the point. Object storage is an architecture – not just an access method. The architecture enables high-throughput scale-out, high-capacity workloads with a unified namespace, and the leveraging of metadata. In this way, traditional file access via CIFS or NFS is an access feature for scale-out storage with a unified namespace.

2015 proved many traditional applications in backup have refactored to support object APIs (e.g. Veritas NetBackup 7.7) for both public cloud storage and scalable on-premises storage under a unified namespace. Like us, other object storage vendors provide filesystem access as a feature, but the industry at large has not been able to find a category name for what we do other than “object storage”. Maybe El Reg can coin a better term for all of us!

El Reg Aren’t Amazon S3 and file-level access driving object storage development more than open source?

Joe Arnold Application workloads are driving development. Object storage wouldn’t exist if there wasn’t a need for it – applications would still be using filers! File-level access is for the applications of yesterday and today, and object API access is for the applications of today and tomorrow.

OpenStack Swift contributor numbers growth

OpenStack Swift contributor numbers growth