Nimble's 'Scale to Fit' Storage Architecture Can Scale Up or Scale Out

Enterprises are often forced to scale up storage to meet capacity needs rather than just scaling out. Nimble Storage is addressing the problem with what it calls a "scale to fit" approach, a storage architecture that allows customers to choose performance or capacity as required.

Nimble's new architecture gives enterprises the ability to scale the capacity and performance of their storage arrays individually based on immediate and near future requirements and budgets, says Dan Leary, the company's VP of marketing. Often, companies often have to predict capacity and performance needs several years in advance.

The company's new line of ES-Series storage expansion shelves allows enterprises to add capacity to the company's high-performance CS200 series arrays. Previously, customers could only buy additional arrays to expand capacity. The company also released a new line of extreme-performance CS400 arrays, which offers higher CPU performance. Customers can expect two to three times the rate of performance on the CS400 series compared with its predecessor, says Leary.

Nimble has also upgraded its software to support clustering, which wasn't available with the CS200 series arrays. Nimble 2.0, which comes free with Nimble's hardware, now enables customers to cluster both CS200 and CS400 controller arrays together into single pools of storage, says Leary. "It essentially gives you linear scaling of both performance and capacity. You don't lose anything because you are clustering."

The new offerings allow customers to extend their storage capacity without having to abandon their existing investments, he says. In addition to being able to cluster CS200 series and CS400 series arrays together, existing CS200 series array controllers can be upgraded to CS400 arrays. "It lets our customers scale their storage deployments exactly along the dimensions they need at the lowest possible cost, and do it all non-disruptively."

A new shelf or array can be added without affecting live applications, Leary adds. The software handles moves, adds and changes automatically behind the scenes without administrator intervention. It also balances storage space across the arrays. "You're not going to end up with one array almost full and one empty," he says. "A volume can also span multiple arrays." A cluster can be treated as one system, he adds. "It removes the complexity of having to manage a separate set of storage silos."

Mark Peters, a senior analyst at Enterprise Strategy Group, says the ability to acquire capacity in the form of expansion shelves based on business requirements is appealing, especially to small and midsize businesses, but it's the flexibility and scalability of clustering that delivers value because it simplifies storage management through automatic configuration and provisioning. "There is no real reward in managing storage," he says. "The more the system can do that for you, the less you have to think about it, the better."

While this is particularly important for smaller enterprises that often don't have IT staff members dedicated to storage, even larger enterprises can benefit, says Peters.

Howard Marks, founder and chief scientist at DeepStorage, says introducing expansion shelves has moved Nimble away from being a fixed-performance, fixed-capacity system. Its storage architecture is now more flexible, he explains, giving customers the ability to add performance or capacity that scales out--two shelves can be added while still appearing as one device, for example. "That's the secret sauce," Marks says.

Next: A Hybrid Approach to Storage ArchitectureLeary explains that Nimble has taken a hybrid approach to its storage architecture. Dubbed Cache Accelerated Sequential Layout Architecture (CASL), it combines the performance of flash memory and low-cost disk capacity. CASL takes a new approach to laying out data, by taking random writes coming into the system and writing them on a new location on disk sequentially, making more efficient use of low-cost disk, he explained. "We turn slow disk into something that's very fast because we're not asking it do something that it's bad at," Leary says.

Meanwhile, data that is likely to read again quickly, such as new email messages in Microsoft Exchange, is put on flash because it's good at performing reads fast, says Leary. Frequently requested data sits on the flash layer but also resides on disk. "We've purpose-built our fundamental architecture around the fact that there's always going to be flash and disk in our system."

Marks says many vendors have bolted on flash to existing storage systems, but there are limitations to how far that can go because they weren't designed to take advantage of flash from inception. "The way you design a system for spinning disks and the way you design a system for flash are very different," he says. "Trying to make one system that does both well while remaining backwards compatible is too hard. Eventually, you have to start with a clean sheet of paper."

He says another unique differentiator for Nimble is its efficient snapshot system built into its file system, which lets the system use snapshots as backups. "It's what makes it use flash effectively, and it's what makes it use snapshots effectively."

Foster Pepper, a Seattle-based general practice law firm with 350 employees and a second location in Spokane, Wash., originally adopted Nimble because it was struggling to back up its discovery data. "We were running into 72-plus hour backup windows," says IT director Lucas Clara.

After adopting Nimble for that purpose alone, the firm moved its entire storage infrastructure over from EMC and has seen improved performance in other areas such as its Microsoft Exchange environment and data warehousing. "We've never run into any storage bottleneck since we've implemented Nimble," Clara says.

Foster Pepper standardized on Nimble two years ago to run its own private cloud. The firm has eight Nimble units in production--both CS240G and CS260G models--which run in its virtualization environments and provide its core infrastructure for storage, with 200 Tbytes of data.

Clara says Foster Pepper does a lot of litigation work that involves e-discovery, which has the potential to bring in a lot of data. Because the amount of data is unpredictable and very case-centric, it's hard to plan out storage capacity. "Today it may be no growth; tomorrow it may be several terabytes of data that comes in."

With Nimble's new storage architecture, Foster Pepper could purchase an additional shelf that includes disk and flash to increase capacity or address performance by clustering more controller units or increasing the cache cards on the system. "It's modular, based on what my business needs happen to be," says Clara. Prior to adopting Nimble, he would have to "drastically" over-provision.

Leary says existing customers can have a new array or shelf shipped to them within a couple of business days, if necessary, and have them up and running within 30 minutes.

Nimble's CS400 series arrays, ES-Series expansion shelves and flash expansion options will be available in the third quarter; Nimble OS 2.0 will be available in the fourth quarter. Pricing for the CS400 series array starts at $80,000, while pricing for CS200 series starts at $38,000. Expansion shelves start at $35,000, with 15 Tbytes capacity and up.