Today, Nimbus Data Systems announced that eBay has deployed more than 100 TB of their flash storage capacity, putting the spotlight on a key win for the emerging storage vendor. This is interesting on a few levels. First, it’s good to see some life from the emerging storage vendor space, after the pool has been depleted by numerous acquisitions by major storage vendors over the last couple of years. This announcement will get Nimbus more attention and give it some headway in the race to survive and succeed in the roster of next-generation of storage system vendors that includes Arkologic, BridgeStor, Caringo, Cleversafe, Coraid, Nexenta, Promise Technology, RELData, Scale Computing, Scality and Stormagic, as well as the crop of cloud gateway providers, Nasuni, Panzura, StorSimple and TwinStrata (I’m always trying to expand this list, comments or suggestions are always welcome…).
More specifically, the Nimbus architecture has the potential to shift current thinking about the role of flash in enterprise storage, and turn notions of tiering upside down. A huge part of the value proposition that Nimbus is claiming (and eBay is validating) is that they can deliver flash-only storage systems at comparable or better price points as systems that use spinning disk. The current market for flash-based storage is confined to the tip-top of the performance spectrum, as most of the entrenched disk storage vendors offer flash capacity at a 5-10X premium over the cost of disk on a price per GB basis. These vendors are justifying the high cost of flash on a price per IOPS basis, saying correctly that when you care about performance, flash is a better solution than disk. But this requires knowing the performance requirements of your data, and making good decisions about which data to put on the expensive stuff, AKA tiering. But tiering is a complex process that most IT ops environments can’t and don't do well manually. Thus the high levels of interest in automated sub-volume tiering, pioneered by Compellent (now Dell), and more recently rolled out by EMC, IBM, HDS and HP/3PAR. NetApp is pushing a different strategy, using flash for cache rather than persistent storage, which eliminates the need for continuously moving persistent data, but in their model, flash still represents only a small portion of costly acceleration capacity with disk still making up the majority.
With Nimbus, they are claiming to bring the total cost (hardware, software & maintenance) of storing data on flash down to parity with disk systems, allowing buyers to get the benefits without having to make hard choices and without having to thrash system CPU constantly moving data around. I’m admittedly skeptical of this as it seems too good to be true, but that’s the claim they are making. Their math relies on the following points:
- Flash optimized architecture — Nimbus has designed their systems from the ground up to take advantage of flash, compared to the establishment that has retrofitted hardware and file systems designed for disk to switch over to flash. I get it that this makes for a more balanced architecture with potentially better results, but it’s hard to imagine that this provides a material reduction of cost.
- Lower overhead RAID and file systems — while most vendors use RAID 1 protection for their flash capacity and require significant capacity for the file system to work, Nimbus claims to use an efficient version of RAID 5 and requires minimal capacity for their file system to do its thing.
- Cheaper support — partly because they’re a smaller, more nimble vendor, and partly because they claim that flash is inherently more reliable and faster to recover from failures due to faster system throughput, Nimbus is charging less than the big guys for system maintenance by about half.
- No software licensing charges — most storage vendors have realized that the software they offer to do management, care and feeding of their storage systems is more valuable than the boxes of spinning drives, and they charge for it; Nimbus, on the other hand, is throwing in their software licensing with the price of the system. I’m not sure how sustainable this is in the long term, as they have to make margin like everybody else, but their current license model seems aimed at bridging the cost gap between flash and disk more than protecting high legacy margins
While the big headline of the Nimbus design is flash for the price of disk, they wouldn’t be viable as an alternative to EMC, NetApp and others without enterprise-class scalability, features and data protection. And while they don’t charge for their software, they do include these must-haves. The protocol support includes FC, iSCSI, NFS and CIFS, they can scale to 250 TB per system by combining modules and they have key features like synchronous and asynchronous replication, thin provisioning, snapshots, data at rest encryption and in-line block level deduplication (a feature that is generally reserved for backup to avoid performance issues, but with the low latency of flash, this becomes possible). All this is pretty impressive to me, and obviously was to eBay as well, as they chose to displace NetApp and HP/3PAR for a big virtual server environment, which is a huge vote of confidence from a well respected IT powerhouse. Nimbus is a big small company, with more than 200 current customers, they have been around since 2006 and they are currently profitable. It will be very interesting to see what kind of impact they can have on the market over time, especially as customers cut their teeth on automated tiering from big vendors; if their experience isn’t so good, it’s not hard to imagine that they will be looking for an alternative, and given the current momentum, Nimbus could present an attractive one.