Remember Seagate’s Kinetic open storage vision? Turns out there is a shipping product embodying the same ideas – but not from Seagate.
Huawei’s UDS – Universal Distributed Storage – system launched two years ago with a home-grown smart disk. Each UDS drive has a daughter board with an ARM processor, memory, two Ethernet ports and software.
The software implements a distributed hash table, a key-value store and two clusters, an object-storage service controller node (OSCN) and a universal distributed storage node (UDSN).
The OSCN provides an access interface for object-based storage services, which is mainly used to process and control access requests from clients, establish an object transmission channel, and manage metadata, while the UDSN is for storing and replicating both data and metadata, and guaranteeing data consistency.
Huawei can pack 75 of these drives into a 4u, top-loading box with two switches so each drive can talk to two fabrics and the two logical clusters. Power consumption is about 3.9W per terabyte.
Besides Seagate, HGST is reportedly experimenting with smart disks as well, running Linux directly on the disk’s controller. But there’s a tough trade-off between software richness and predictable latency.
The StorageMojo take
Globally routable storage. Advanced erasure coding across drives. A standardized object storage interface. Organic rolling upgrades without forklifts. All on a mass-produced, low-cost brick: the smart disk drive.
As StorageMojo noted last year:
Getting RAID controllers out of the stack reduces latency and eliminates a major cost and bug pool – a Very Good Thing. It also allows drive vendors to reclaim margin lost with falling enterprise drive sales. . . .
Distributing the needed intelligence to the lowest possible level – the drive – should be more scalable and economic than current DAS-based server models. The tipping point is the value of the aggregation, caching and low-cost disk connectivity – network bandwidth is way more costly than internal DAS bandwidth – of storage servers versus the advantages of removing the storage server tier.
Instead of tens of thousands of RAID controllers, hundreds of millions of smart disks, with all the advantages of object stores. That would go far in reducing operating costs and bugs.
The key is a systems-level – not drive level – approach to the architecture, which drive vendors aren’t well-equipped to provide. Huawei’s approach is promising and shows what can be done – if drive vendors can summon the courage and the smarts to do it.
Courteous comments welcome, of course.