Introduction

Solid-state Storage Adoption and Use

The use of solid-state, both in terms of adoption and application, is broadening. Recent ESG research[1] endorsed this statement and added a lot of specificity. At a high level, the situation is this:

• Adoption: The use of solid-state storage is increasing, with ESG’s research finding that just over a half of the enterprise-class end-users surveyed either already have implemented solid-state in some form or plan to implement it by mid-2012.
• Application: The research found that solid-state storage is making a move from being “simply” a specific application “problem fixer” for things such as databases, to being a broader horizontal infrastructural play for things such as server virtualization. Although 62% of current solid-state storage users report that their organizations purchased it in order to alleviate performance challenges associated with a specific application, more than half of potential solid-state storage adopters do not believe that their organizations will deploy the technology as a means to address specific application performance challenges.

As the choice, economics, and understanding of solid-state are all increasing, so the opportunity for new, innovative vendors such as Pure Storage is excellent.

Pure Progress

Pure Storage positions its FlashArray storage system as a next-generation tier-1 storage array. As such, it is aimed very squarely at replacing current—mainly-HDD based—high-performance tier-1 storage systems. Pure is not looking to offer an all-out performance-at-any-price flash appliance (such devices often employ DRAM as well as NAND flash and are focused on extreme application acceleration needs), nor is it looking to provide a hybrid system. It is instead aiming to present a viable—indeed preferable—alternative to contemporary performance enterprise disk systems, not only by offering higher performance and management ease at a comparable price, but also by providing all the crucial functionality such as HA that is expected for mission-critical workloads … yet with far better operational characteristics.

On paper and in early testing, the overall package is one that could provide a real challenge to the status quo. Pure is now well into its beta program. It is largely competing with “mainstream” systems (such as those from EMC or NetApp), and it has updated ESG on its progress.

• Beta commentary: Pure Storage has more than 25 beta sites up and running, covering an extremely varied range of industries (from manufacturing and advertising to government, education, and IT) and workloads (from virtualization to specific specialist applications). Most sites were operational within hours and, despite only having had their Pure devices for a few weeks, many of the beta customers have already agreed to purchase their units. The early results are still confidential[2] (but impressive) and are not only proving the product but also providing valuable reality lessons for the IT industry (for example, advice on testing solid-state, tuning applications, and appreciating that most users don’t actually have 4k all-read workloads).

Implementation and Operational Lessons

Early Pure Storage users have revealed a number of insights regarding the ways that solid-state should be treated differently from—and can provide added value compared with—spinning disks. Deploying flash differs from disk in different application environments. The balance of this brief examines this idea in terms of two crucial workloads, and it reviews what ESG research discovered about the market relevance of solid-state for these two environments:

• VMware: A prime example of server virtualization environments driving solid-state storage adoption.
• Oracle: A prime example of database/OLTP workloads that lead the specific applications for which solid-state storage has been, and is expected to continue to be, deployed.

Solid-state in VMware Environments

ESG Research

Storage performance is a widely recognized challenge stemming from deployments of server virtualization technology. In fact, ESG research has revealed that more than 20% of organizations have experienced performance challenges in the disk systems that support their virtual servers.[3] As such, it is not surprising (see Figure 1) that more than one-third (38%) of current solid-state storage users identified the alleviation of IO bottlenecks caused by server virtualization as the primary reason for their organization’s initial deployment of solid-state storage. Similarly, more than half of potential adopters (59%) divulged that it was at least one of the reasons for deploying or considering solid-state storage.

Pure Insights

With server virtualization being such a strong element contributing to deploying solid-state storage—it was a reason, primary or otherwise, for an impressive four out of five (81%) current users—any suggestions for optimum implementation are welcome. Here is what Pure Storage, working with its beta customers, has found to be best-practice advice:

• Larger LUNs/datastores are possible. In the disk world, users have generally kept datastores small (<1TB) in order to limit the chances of performance contention. That has meant that they have to manage a lot of LUNs and datastores, and they constantly end up moving virtual machines among them. With flash and the new vSphere 5 release, it is entirely possible to use up to the full 64 TB datastore sizing (which also allows Fibre Channel deployments to benefit from the same kind of “everything in one bucket” benefits that NFS already provides).
• Use “EagerZeroThick” instead of “LazyZeroThick” VMs. When using full/thick provisioning, users used to have a choice: Either have VMware write zeros to the entire VM space (“eager,” which yields better performance but takes up to 30 minutes per VM when provisioning), or reserve the space but only write to it when needed (“lazy,” which is faster to provision but slows performance). With a system like Pure Storage that offers zero elimination and VAAI block zero support, users can use EagerZeroThick to get the best VMware performance, and VM deployments can still be fast (in the range of seconds to minutes).
• VM/disk block alignment becomes a non-issue. Many users go to great lengths to ensure that the block alignment of their VMs matches their disk array in order to ensure that a single-block read on the VM side of things doesn’t trigger a double-block read on the array side. This does not matter in a Pure Storage environment. Pure’s operating system, the Purity Operating Environment, virtualizes at a 512-byte level. This means users can choose any block size on the host. Unlike some of the other capabilities mentioned in this section, it is important to note that this ability is specific to Pure Storage, not general to all flash storage.
• No need for host-side SSD cache. Recent VMware VDI best practices suggest putting solid-state storage in each VDI server that a user has in order to serve as a local cache to offload IO. Naturally this can help in virtualized environments, but it comes at the cost of a lot of waste of solid-state storage. Using a shared flash array allows users to avoid this waste. And indeed, ESG’s research found that roughly two-thirds of respondents viewed having their solid-state resources shareable as important.
• More consolidation, mixed workloads. Today, VMware environments tend to limit their consolidation when IO is a bottleneck, and they often create “application islands” to ensure there is no performance contention for critical applications. Neither of these is necessary when using a flash array. Flash invariably serves the IO demands—both in terms of IO bandwidth and randomness—of multiple workloads well.
• Accelerate VMware operations. Almost all VMware operations (snapshots, clone deployments, vMotions, etc.) can easily become storage IO bottlenecks. Naturally such operations happen significantly faster (maybe 5-10x) on platforms such as Pure’s FlashArray, with the result that IT managers are free to take more advantage of such advanced VMware functions, since they perform better and complete faster.

Solid-state in Oracle Environments

ESG Research

While the move to solid-state storage being applied across the whole infrastructure is important, it is still true that, to date, most users have purchased solid-state to alleviate a performance issue with a specific application.

Figure  shows the specific application types that users acquired solid-state storage to support, as well as those that potential adopters expect to support with the technology. While a range of applications are cited, the most significant in terms of influencing solid-state adoption were clearly—and not surprisingly—databases/OLTP environments and financial applications/ERPs.

It is interesting to observe some of the differences between current users and potential adopters, where the overall increase in the number of applications mentioned by potential adopters serves to reinforce the concept of a shift toward a more horizontal usage of solid-state storage. In other words, future users are going to leverage the technology across a broad spectrum of their IT and business operations, not just for database and financial applications.

Pure Insights

Nonetheless, in terms of applications, what has driven the market so far—and looks set to continue so to do with an increasing dominance—are exactly the sorts of things that Oracle provides. And that is the second environment where Pure and its beta customers have discovered a range of comments, best practices, and advice for flash users to consider.

• Greater consolidation is achievable. Put simply, faster storage tends to mean higher transactional throughputs for users’ database instances, which in the end will naturally tend to mean that fewer of them will be needed (with the concomitant need for less hardware and less software license expense).
• Simpler provisioning. Today, many users find that a large part of the pain of managing storage in an Oracle environment is centered in the need to “hand-craft” the storage for each of the Oracle objects (such as DB, tmp, redo, etc.). Each of these objects has a different performance and sizing need, so each will typically therefore have a different LUN with different RAID/spindle-geometry trade-offs. Provisioning Oracle on flash (as Oracle itself is promoting) can be much simpler: “Create one LUN. Put everything in it. Done.”
• Reduced index dependencies. When addressing Oracle performance problems, a common solution is to create indexes. These improve performance but consume a lot of space and create CPU/IO overhead. By using flash, users can generally reduce the number of indexes and make the database more efficient.
• Any Oracle block size. Similar to VMware, users can pick any block size without performance worries. In the spinning disk world, the name of the game is often to try and move to larger block sizes to allow disks to be more efficient. (Although this can make the database less efficient.) With flash, the ideal tuning is often to “dial down” the block size significantly (4K or 8K), thus allowing the database to be more efficient.
• Mix workloads. Traditional disk often doesn’t serve “conflicting” workloads well, so users make many copies of their database for various operations (online transactions, analytics, or backup, for instance). Using flash, the need for multiple copies goes away, and these “conflicting” workloads can all run on the same database instance. This also means that analytics can run off live data, delivering more real-time and relevant results.

A Further Note on Performance

The way the solid-state market is currently moving is all too often over-focused on performance “specs” alone, and not enough on what the real-world implications of solid-state can be. There has always been—and likely always will be—a segment of the market that is totally committed to the ultimate, marginal, ever-improving performance at any price. But this market segment is very application-specific and limited in scope. The big opportunity for flash is not only for it to become mainstream in terms of acceptance, but also to become a part of the mainstream general storage market in terms of broad adoption.

Some of the current performance-hype surrounding solid-state in all its forms definitely needs to be taken with a dose of reality and at times even a pinch of salt. After all, there is not an over-abundance of users or applications that are demanding many millions of continual IOPS. But there is plenty of value to be reaped. The performance of flash is a means to a number of ends as much as it is an end in itself. Therefore the pragmatic advice that Pure Storage and others are sharing about the practicalities and considerations of flash implementation is very helpful.

Flash may not be a panacea, but it categorically has its place. And the economical performance that it can deliver—when packaged and managed well—is capable of delivering enormous value to much of the mainstream IT community for whom serving multiple applications while driving consolidation and demanding improved value is the order of the day.

The Bigger Truth

As the IT world moves to adopt solid-state more widely and to use it across a growing range of applications (horizontal and infrastructural, vertical and specific), it behooves users to consider not only performing their current operations better, but also looking at changing the ways they do things. Solid-state is likely to improve their results in any case, but those results will be sub-optimal if the additional opportunities that solid-state storage offers are not harvested.

Some of the early implementations of flash storage and even some of the first new market entrants over the last few years have been very focused on just enhancing what is already in existence. That’s no bad thing per se. But there’s a new breed of vendor and approach, and Pure Storage is a prime example, that is actively seeking to replace the current tools and take even greater advantage of the opportunities that solid-state can confer. In other words, Pure is able to compete in, if you will, the “regular” world. Users should be prepared to be a little more flexible in their approach than they perhaps imagined … in order to benefit a lot more than they perhaps expected.

[2] ESG expects to interview a number of Pure Storage users early in Q1 2012 and will report its findings then.

[3] Source: ESG Research Report, The Evolution of Server Virtualization, November 2010.


]]> http://www.enterprisestrategygroup.com/2012/01/pure-storage-makes-solid-progress-and-provides-valuable-insights-for-all/feed/ 0 http://www.enterprisestrategygroup.com/2012/01/marvell-unveils-first-scalable-pcie-nand-flash-controller-computerworld-2/ http://www.enterprisestrategygroup.com/2012/01/marvell-unveils-first-scalable-pcie-nand-flash-controller-computerworld-2/#comments Tue, 10 Jan 2012 14:24:07 +0000 cwhitehouse http://www.enterprisestrategygroup.com/?p=27624 Mark Peters, an analyst with Enterprise Strategy Group, said what sets Marvell’s new flash module apart from others on the market today is its ability to scale to address multiple business applications.

He criticized other NAND flash system manufacturers for being mostly focused on performance rather than tailoring their products to meet more general needs.

“People are saying, I don’t need a 6.9-liter twin-turbo intercooled engine in my car to go to [the grocery store]. Please sell me the 4.2-liter gas hybrid to do that,” he said. “And, Marvell is saying if you want, you can also scale it to a 6.9 liter engine when you want it. That’s really smart.”

via Marvell unveils first scalable PCIe NAND flash controller – Computerworld.