A Game-Changer in Storage
OpenStack and Kinetic: Next-Generation Datacenter Storage Starts Now
What's this
OpenStack is poised to shake up computing throughout the enterprise world, but perhaps the most obvious and transformative change will appear in storage. The previously mentioned OpenStack Object Storage (Swift) is built to deliver the same “bucket” approach to storage as that used by Amazon S3. With object storage, rather than worrying about things like sector-level block management, the system treats the file as a whole unit, an object, that can be placed anywhere within the storage media — even spanning multiple drives, if necessary. A database handles all of the tracking on the back end. Admittedly, object storage systems tend to offer relatively few capabilities. Users can put, fetch, copy, and delete, and admins can dictate which users can do which of those functions. On the other hand, object storage delivers new capabilities, such as allowing an application to program the storage interface and handling data replication of objects by the storage platform rather than discrete, external resources.
“OpenStack releases all the constraints of what we’ll call the vendor lock with all the specialized hardware, all the heavy expensive SAN environment,” says Silicon Mechanic’s Hostetler. “It just says. ‘You’ve got a bunch of data that you want to store. Just give it to me. I’ll make sure there are multiple copies, and I’ll make sure that you can get them.’”
Amazon and Google have proven that commodity storage resources can serve perfectly well in such environments and applications. There is no need to take a performance hit on parity overhead when datacenter-scale redundancy is in play. In fact, many of storage’s accepted conventions are being reexamined and upended. One of the best examples may be Seagate’s Kinetic Open Storage Platform effort.
At its simplest, the Kinetic platform leverages OpenStack Swift’s object storage capabilities and marries them with conventional storage drives updated with Ethernet ports rather than the usual SAS interfaces, even though the drive’s connector physically matches SAS. Every drive has two globally unique IP address, which operate in active/active fashion. Seagate has not released performance numbers yet but expects Kinetic-ready drives to perform “very close to disk speeds.”
Seagate’s Kinetic platform relies on key/value architecture rather than logical blocks, an alternative programming concept to SAS commands that lets clients directly address objects rather than media blocks. Instead of having backplanes feed into a SAS switch, they here feed into a dual gigabit or 10 GbE Ethernet switch. The result is that applications can leverage the key/value API to communicate over gigabit Ethernet straight to the drive. As the following slide shows, this cuts out a vast swath of conventional architecture.

The ramifications of this seemingly simple change are huge. While a Swift/Kinetic deployment will still need a limited number of proxy servers, the old need for fleets of storage servers vanishes. There is no more need for storage CPUs, SAS adapters, ranks of memory, operating systems, and so on.
“Comparing a cloud storage deployment today to a future one based on Kinetic, deployers will have the option of reducing the total number of servers,” says SwiftStack’s Mario Blandini. “Kinetic provides the option of using lower cost chassis, which, in high-capacity clusters, is the most prevalent device in the storage node. Think of only having a few proxy servers tomorrow compared to dozens of total storage servers today.”
Obviously, the total cost of servers in the data center will plummet. This yields secondary benefit, as well. In storage, CPUs account for most of the heat output, so cooling costs fall. Similarly, CPU heatsink fans — not the storage drives — account for most of the vibration in storage racks. Eliminating this vibration will improve read/write accuracy, drop latencies, improve overall performance, and extend drive longevity throughout the datacenter. On top of all that, there’s still Swift’s tremendous reduction in metadata overhead — up to 50% by Seagate’s estimates. According to Seagate’s Fenn, this could yield random write performance improvement of up to 300% in some workloads.