FLASH MEMORY STORAGE ARRAY

Transcription

1 2013 The Insider s Guide to Evaluating Memory Storage Arrays By Ken Clipperton and Joshua Konkle DCIG, LLC. All rights reserved.

2 DCIG 2013 The Insider s Guide to Evaluating Memory Storage Arrays Table of Contents 1 Introduction 2 Executive Summary 4 How to Use This Buyer s Guide 5 Disclosures 5 Inclusion and Exclusion Criteria 6 The 8-Step Process Used to Score and Rank Products 7 DCIG Comments and Thoughts 7 Memory as a Storage Category 7 SAN Replacement Versus Niche Player in the Data Center 7 Memory Storage Arrays Include All Software Features at No Extra Cost 8 Memory Optimization is Important 8 VMware 8 Memory Storage Arrays are Data Center Capacity and Efficiency Multipliers 9 Performance and Pricing 9 Reliability 10 Feature Areas Where DCIG Expects to See Improvement 10 DCIG Observations and Recommendations on the Rankings 10 Best-in-Class Ranking 11 Recommended Ranking 12 Excellent Ranking 13 Good Ranking 13 Basic Ranking 15 Memory Scores and Rankings 16 Overall Scores and Rankings 18 Memory Storage Array Buyer s Guide Models 19 Astute Networks ViSX G Astute Networks ViSX G Astute Networks ViSX G GreenBytes IO Offload Engine (5,000 VDI Model) 23 Huawei Enterprise OceanStor Dorado Huawei Enterprise OceanStor Dorado Kaminario K2 F 26 Nimbus E Class Memory System E6400M+E1000X 27 Nimbus E Class Memory System E6400M+E2000X 28 Nimbus S Class Memory System S255M / S255X 29 Nimbus S Class Memory System S505M / S505X 30 Nimbus S Class Memory System S1005M / S1005X 31 PureStorage FA PureStorage FA Tegile Zebi HA Texas Memory Systems, an IBM company, RamSan Texas Memory Systems, an IBM company, RamSan Texas Memory Systems, an IBM company, RamSan Violin Memory Violin Memory Violin Memory Violin Memory Violin Memory Violin Memory Violin Memory Violin Memory Violin Memory Whiptail Accela ASSA Whiptail Accela ASSA Whiptail Accela ASSA Whiptail Accela ASSA Whiptail Invicta ISSA 6 51 Whiptail Invicta ISSA Whiptail Invicta ISSA Product Rankings Dashboard Appendix A-1 Definitions, Explanations and Terminology B-1 Storage Provider Contact Information C-1 Author Contact Information 2013 DCIG, LLC. All rights reserved. i

3 DCIG 2013 The Insider s Guide to Evaluating Memory Storage Arrays Introduction Welcome to DCIG s inaugural Memory Storage Array Buyer s Guide. DCIG created this Buyer s Guide to help organizations make faster and better informed purchasing decisions. DCIG accomplishes this goal by helping organizations identify the features that should matter most to them and then more quickly get to a short list of products that they want to test and/or acquire. The DCIG 2013 Memory Storage Array Buyer s Guide includes a scoring category called that reflects the support of each array for the most important flash optimization techniques. Every type of storage media requires optimizations for strengths and weaknesses. One weakness of flash memory is that writing any data requires that an entire block be erased and then re-written. Another weakness is that flash supports a limited number of write-cycles before it can no longer reliably store data. These weaknesses can be addressed through a variety of flash management optimization techniques. Like all prior DCIG Buyer s Guides, the DCIG 2013 Memory Storage Array Buyer s Guide continues to do the heavy lifting for organizations as they look to purchase a flash memory storage array by: Listing each individual flash memory storage array model by storage provider. Listing different flash memory storage array features, whether or not they are supported and (where appropriate) how they are implemented. Weighting these features according to what organizations consider most important. Scoring these features. Ranking each flash memory storage array model. Creating a standardized data sheet for each flash memory storage array model. The end result is that the DCIG 2013 Memory Storage Array Buyer s Guide will give organizations the opportunity to do at-a-glance comparisons between many different flash memory storage array models. Using these results, they can quickly come up with a short list of products that meet their specific needs and either go directly to a product evaluation stage with select flash memory storage arrays or even opt to get competitive bids on these products. Note that this Buyer s Guide is not intended to be a substitute for bringing individual flash memory storage arrays in-house and testing them with specific applications and data center environments. That function should still be done, if possible, since every storage array will perform differently under different application workloads. The purpose of the DCIG 2013 Memory Storage Array Buyer s Guide is to help organizations identify the features that should matter most to them and then get to that short list of products that they want to test much more quickly. We hope you find that this Buyer s Guide meets this intended purpose in your environment. Ken & Joshua 2013 DCIG, LLC. All rights reserved. 1

4 DCIG 2013 The Insider s Guide to Evaluating Memory Storage Arrays Executive Summary memory storage arrays are getting attention because they help organizations achieve results that matter to them. For example, the success of corporate initiatives like client virtualization and the opportunity to fix application performance issues that could not be cost-effectively addressed by traditional spinning disk storage systems. Memory Storage Arrays bring performance improvements to the I/O subsystem that finally keep pace with the other performance elements in the data center. memory storage arrays support a new generation of servers. These servers are built for virtualization by supporting up to 16 processor cores and 768GB of RAM in just 2U of rack space. They achieve 4x the processor capacity and 4x to 20x the RAM capacity per rack unit. Storage systems had trouble feeding enough data to the older generation of servers to keep CPU's busy and represent the single biggest obstacle to data center efficiency. The new generation of servers demands a new storage paradigm. Memory Storage Arrays bring the following compelling benefits to the data center: 10x to 100x improvement in IOPS. memory storage arrays provide up to 1,000,000 IOPS and much lower latencies than spinning disk storage arrays. The result is a substantial improvement in application performance and user experience. For example, unpredictable virtual server workloads or VDI boot storms. 2x virtual machines per physical machine. CPUs spend much of their time waiting on responses from the underlying storage subsystem. By using flash memory arrays, these wait times are minimally reduced by a factor of two (2). Reduced wait times improve application performance and user experience. For example, applications perform faster and users are not waiting for tasks to complete. Reduced power consumption and heat per TB and per rack unit that multiplies the effective capacity of existing data centers. For example, one can achieve 1,000,000 IOPS with fewer spinning disks by using flash. Reducing total number of disk devices reduces power and cooling requirements in the data center. Reduced management overhead through simplified deployment and management tools. memory storage arrays include flash memory storage module/disk management software. For example, flash memory modules can wear out after many writes. memory storage arrays include software to manage the wear leveling of the flash memory. Storage and data center system buyers are experiencing ROI and a substantial improvement in their overall return on assets by adding flash memory storage arrays to their data center storage environment. The following are examples of ROI experienced by buyers: Elements of ROI: Increased performance which achieve business goals. Increased data protection through integrated snapshot/replication. Increased flexibility and automation through VMware integration DCIG, LLC. All rights reserved. 2

5 DCIG 2013 The Insider s Guide to Evaluating Memory Storage Arrays Executive Summary continued Decreased data center footprint using high IOPS/rack unit flash memory. Decreased power and cooling requirements as compared to spinning disks. It is in this context that DCIG presents its DCIG 2013 Memory Storage Array Buyer s Guide. As prior Buyer s Guides have done, it puts at the fingertips of organizations a comprehensive list of flash memory storage arrays and standardized data sheets that can assist them in this important buying decision. The DCIG 2013 Memory Storage Array Buyer s Guide accomplishes the following objectives: Provides an objective, third party evaluation of Memory Storage Arrays that evaluates and scores their features from a end user s viewpoint. Scores and ranks the features on each Memory Storage Array based upon the criteria that matter most to end users and presents the results in easy to understand tables so they can quickly know which Memory Storage Arrays are the most appropriate for their needs. Provides a standardized data sheet for each of the 34 Memory Storage Arrays from 10 different storage providers so end users can do quick comparisons of the features that are supported and not supported on each Memory Storage Array. Provides insights into what VMware APIs the Memory Storage Arrays offer to optimize their VMware environments. Gives any organization a solid foundation for getting competitive bids from different providers of Memory Storage Arrays that are apples-to-apples comparisons DCIG, LLC. All rights reserved. 3

6 DCIG 2013 The Insider s Guide to Evaluating Memory Storage Arrays How to Use This Buyer s Guide In determining how to best use the information contained in this Buyer s Guide, it is important to note that it is intended to help users accelerate the product research and selection process driving cost out of the research process while simultaneously increasing confidence in the results. The purpose of this Buyer s Guide is NOT to tell users exactly which product(s) to purchase. Rather it is to help guide them in coming up with a short list of competitive products that have comparable features that meet their specific needs. It is also important to note that just because a product scored the highest in a particular category or is ranked a certain way does not automatically mean that it is the right product for their organization. If anything, because of the scope of the products evaluated and analyzed, it may have features that are too robust for the needs of an individual department or organization. However what this Buyer s Guide does is give users some sense of how each product compares to other products classified in this guide, as well as offers additional insight into what product offerings are available on the market. DCIG recommends that you use this Buyer s Guide in the following seven ways: Eliminate the painstaking research associated with coming up with a short list of products that meet their needs. This Buyer s Guide ranks, scores and contains data sheets for 34 different products from 10 different vendors. Product is scored and then ranked as Best-in-Class, Recommended, Excellent, Good and Basic based upon its score. In each product, more than 80 different features were evaluated, weighted, scored and then ranked. All an organization has to do is look at the scores and features of each product in order to come up with a short list of products for consideration. Do apples-to-apples comparisons of products from different vendors. In today s crowded Memory Storage Array market, it behooves organization to get competitive bids from multiple vendors. After all, when they compete, you win! But that tactic only works well when organizations know that they are receiving competitive bids on products that are roughly comparable. Using this Buyer s Guide, organizations can do a better job of accomplishing that objective. Separate the apples from the oranges. Just as important as doing apples-to-apples comparisons is identifying when an orange is thrown into the mix. Sometimes it is very difficult for an organization to know if it is truly getting a good deal when bids come in from vendors that include different products. Now organizations can refer to the scores and rankings of each product on this guide so they know when they are getting a good deal, a great deal or just a so-so one. Provides perspective on how products from less well known vendors compare against established and better known brands. Anyone involved with high performance storage systems has probably at least heard of Texas Memory Systems, now an IBM company, but many of the companies providing these innovative products are less well known. This Buyer s Guide helps to remove some of that apprehension about buying from a less well known vendor or even a less well known model from an established vendor. Using this Buyer s Guide organizations can see how these products from lesser known vendors as well as lesser known products from established vendors stack up. Normalizes complex terminology. Every segment across industries has a proclivity to adopt acronyms and jargon that is specific to it. The flash memory storage industry is no exception. In this dynamic and relatively young market, vendors often refer to the same capability in different ways. This Buyer s Guide sifts through the acronyms and jargon and then normalizes these terms, providing a foundation for meaningful comparisons. Definitions for these normalized terms are provided in the Glossary, included as Appendix A. Creates a standardized data sheet. The product data sheets available from the different vendors are rarely laid out in the same way or contain the same information. Some vendors even have data sheet formats that vary from product to product within their own portfolio. This Buyer s Guide tackles this problem by creating a standard, easy to read data sheet for every product. In this way, product data sheets for individual 2013 DCIG, LLC. All rights reserved. 4

7 DCIG 2013 The Insider s Guide to Evaluating Memory Storage Arrays products can be printed out, laid down side by side and then the features on them quickly compared. Helps justify buying recommendations to business teams. Nothing is easier for those on the business side to understand than a number when doing comparisons. So at the top of every product data sheet, a product score is included so the business side of the house can quickly see how the different product models compare. Disclosures Over the last few years the general trend in the US has been for both large and boutique analyst firms to receive some or all of their revenue from vendors. DCIG is no different in that respect as it also receives payment for the different services it performs for vendors. The services that DCIG provides include blogging, case studies, product reviews, executive white papers and special reports. So in the interest of being fully transparent, a number of the vendors included in this DCIG Buyer s Guide are or have been DCIG clients. This is not to imply that they were given preferential treatment in the Buyer s Guide. All it meant was that DCIG had more knowledge of their products and that DCIG would consider their product for inclusion in this Buyer s Guide. In that vein, there are a number of important facts to keep in mind when considering the information contained in this Buyer s Guide and its merit. No vendor paid DCIG any fee to develop this Buyer s Guide. DCIG did not guarantee any vendor that its product would be included in this Buyer s Guide. DCIG did not imply or guarantee that a specific product would receive a good score on this Buyer s Guide, before or after completion. All research was based upon publicly available information, information provided by the vendor and the expertise of those evaluating the information Because of the number of features analyzed, how these features were weighted and how these products were then scored and ranked, there was no way for DCIG to predict at the outset how individual products would end up scoring or ranking. DCIG wants to emphasize that no vendor was privy to how DCIG did the scoring and ranking of the products. In every case the vendor only found out the scores and rankings of its product after the analysis was complete. Inclusion and Exclusion Criteria This inaugural edition of the DCIG 2013 Memory Storage Array Buyer s Guide attempts to take into consideration as many variables as possible though likely fails to account for all of them. This may have even resulted in a specific model not being covered in this Buyer s Guide when it rightfully belonged. DCIG will refine both the variables for which we collect data and the inclusion criteria for future editions of the DCIG Memory Storage Array Buyer s Guide. The rationale that was used to determine the inclusion and exclusion of specific models in this Buyer s Guide is as follows: Available as an appliance. Incorporated into rack-based data centers in a quick and familiar way. Fitting into data center design as a replacement for another storage array. Is NOT based on Microsoft Windows Storage Server. This requirement was put in place to eliminate products that are just standard server hardware with flash drives plus MS Storage Server. memory has specific weaknesses that must be addressed if it is to be responsibly incorporated into storage arrays as primary data storage. We were looking for vendors who have taken steps to address flash memory s weaknesses. Must support a configuration containing only SSDs. A number of establish storage vendors have begun to offer flash as a cache or tier in their traditional arrays. To be included in this buyer s guide, vendors had to support an all-flash configuration of their array. May support mixed configurations of HDDs and SSDs. for a hybrid SSD/HDD did not disqualify a product for the buyer s guide as long as an all-flash configuration was supported and available as a specific SKU. Must use SSDs as primary storage. memory is being incorporated into the data center in many different 2013 DCIG, LLC. All rights reserved. 5

8 DCIG 2013 The Insider s Guide to Evaluating Memory Storage Arrays ways and in different roles. We were looking for arrays that supported flash as the permanent home for data. Must support a minimum of sixteen (16) disks and/or modules. Must support one or more SAN protocols (, Fibre Channel, FCoE, Infiniband). May support one more NAS protocols (CIFS,NFS, REST). for SAN protocols is required, but support for NAS protocols in addition to SAN did not disqualify a product from being included. Indeed, 13 of the 34 arrays do provide concurrent SAN/NAS capabilities. There had to be sufficient information available to DCIG to make meaningful decisions. DCIG made a good faith effort to reach out and obtain information from as many array providers as possible by providing surveys to them for them to complete. However, some products still had to be excluded because of a lack of reliable data. Must ship by October 31, A cut-off date had to be put in place or this Buyer s Guide would never be published. The 8-Step Process Used to Score and Rank Products To score and rank each product, DCIG went through an eight step process to come to the most objective conclusion possible. 1. DCIG listed out all of the features available on all of the products. Prior to selecting the features that were included in the final evaluation in the Buyer s Guide, DCIG went through and quantified what features the products possessed. As part of this process, DCIG normalized the list of features available on products such that a common name for each feature was established. 2. DCIG established which features would be included in the Buyer s Guide and which ones would not. One of the goals of this Buyer s Guide was to try to only include features in each product that could be objectively and authoritatively analyzed. For example, Maximum Raw Storage Capacity was evaluated as a feature instead of Maximum Useable Storage Capacity. While useable storage capacity is what users ultimately care about, a consistent objective answer cannot be arrived at as most arrays offer multiple RAID options and some offer deduplication and compression. So Maximum Raw Storage Capacity was selected as the feature to be evaluated since an objective answer could be ascertained and supported. 3. Each feature had a weighting associated with it. The weightings were used to reflect if a feature was supported and potentially how well it was implemented. For example, the feature is more of a Yes or No type of response whereas the can be one of several different options including Active-Active, Active-Standby, Dual Active, and Single Controller. Each of these features was weighted and scored differently. 4. A survey that asked about all of the features scored in this Buyer s Guide was sent to each vendor. In addition to using the information that was publicly available on each vendor s website, each vendor included in this Buyer s Guide had the opportunity to respond to a survey sent to it by DCIG. This was done to both verify that the information DCIG found on the vendors website was correct as well as remove any ambiguities that existed regarding how some of the features were implemented. 5. All vendors were given the opportunity to review their data sheets before the final scores and rankings were determined. To ensure the information presented in this Buyer s Guide is up-to-date and reliable, DCIG provided each vendor a copy or copies of their data sheets that appear in this Buyer s Guide without the scores and rankings on them. In this way they had the opportunity to validate the information and correct it before it was publicly released. 6. All of the features were scored based upon the information that was gathered. The weighting and scoring was done both by DCIG and by a select group of end-users who assisted DCIG in selecting the features to be evaluated in this Buyer s Guide. 7. The features were broken down into general categories. The features included in this Buyer s Guide broke down into a total of six broad categories that are reflected on each array data sheet. These six categories 2013 DCIG, LLC. All rights reserved. 6

9 DCIG 2013 The Insider s Guide to Evaluating Memory Storage Arrays include and Replication, VMware, Storage Array Hardware,,, and. scores were not displayed on the data sheets as there was more variation among scores in the other categories. 8. The vendors were ranked using standard scoring techniques. One of the goals of this Buyer s Guide is to establish clear lines of differentiation between products with conclusions that are arrived at objectively. To accomplish this goal, the mean or average score for each classification was first determined and then the standard deviation. Using the mean of the scores from all of the products from which the standard deviation was calculated, DCIG developed a ranking for each product based upon the following in each classification: Those models that were more than.425 standard deviations or more below the mean were given the rank of Basic. Those models that were.425 standard deviations above or below the mean were ranked as Good. Those models that were standard deviations above the mean were ranked as Excellent. Those models that were greater than standard deviations above the mean were ranked as Recommended. The model with the highest score was given the designation of Best-in-Class. Because the mean score and standard deviation was calculated separately for each classification, the number of products that achieved a particular ranking varied. DCIG feels confident that using this scoring and ranking method provides a good foundation for meaningful comparisons among all of the products included in this Buyer s Guide. DCIG Comments and Thoughts Memory as a Storage Category memory-based storage has emerged as a viable, cost effective, and success-enabling storage technology for the data center. Because the flash memory storage marketplace is relatively young and because vendors are promoting various approaches to bringing flash memory storage to the data center there is some confusion about where flash memory storage fits into the data center and about the product features that are likely to be most important to end users. In this context, this inaugural edition of the DCIG 2013 Memory Storage Array Buyer s Guide aims to deliver value to prospective storage purchasers by helping them make faster and better-informed purchasing decisions. SAN Replacement Versus Niche Player in the Data Center DCIG believes that Memory Storage Arrays are poised to address not only special I/O-intensive use cases, but to begin displacing traditional storage arrays in many data centers where performance is a particular concern. The arrays included in this Buyer s Guide support the use of flash as the permanent home for data, and many of the vendors do intend for their flash memory storage arrays to function as a full replacement for the traditional SAN storage array. However some vendors, including GreenBytes and Astute Networks, are focused on a specialized niche role in the data center. These vendors intend their arrays or appliances to function as single-purpose storage systems. GreenBytes focuses on large-scale VDI deployments, while Astute zeros in on fixing application performance problems. These solutions are intended to complement rather than replace existing data center storage. The fact that a vendor markets an all-flash memory array for a particular purpose does not rule out its use for other purposes. However, arrays from vendors who expect their products to fill a niche role may be missing some features that would facilitate use across the full spectrum of use cases. For example, Astute eschews Fibre Channel connectivity and thin provisioning because these features are not relevant to its use case. Memory Storage Arrays Include All Software Features at No Extra Cost The majority (70%) of arrays included in this Buyer s Guide come with array management software included, and with all features fully licensed as part of the standard array package. If a Memory Storage Array will replace an existing array, purchasers must weigh the lower cost of software 2013 DCIG, LLC. All rights reserved. 7

10 DCIG 2013 The Insider s Guide to Evaluating Memory Storage Arrays against existing storage vendors who may charge separately for each feature a purchaser requires. The decision by the vendors to include the full software suite reduces buying overhead. Buyers are not distracted by separate purchase decision and justification required to license array management software functionality. In that regard, cost of hardware and software maintenance for a legacy storage system versus the acquisition and maintenance cost for a replacement Memory Storage Array should be incorporated into overall ROI calculations. Memory Optimization is Important Like all types of storage media, flash memory has both strengths and weaknesses. memory s strengths include low latency and low power requirements, but weaknesses like low write endurance were initially a cause for skepticism about its use in enterprise storage systems. Some storage vendors have successfully addressed flash memory's weaknesses through a variety of techniques. Those techniques, and the percentage of arrays in this Buyer s Guide that incorporate them, are: (88%) (85%) (38%) (35%). (18%) (12%) the most implemented flash memory optimization technique involves buffering and/or coalescing of writes to match erasure block size (EBS) of the underlying flash media. This technique reduces the number of write-to-flash operations and program-erase cycles that are incurred in order to store a given amount of data. As a result, write coalescing extends the life of flash memory AND maintains its high performance across time. memory optimization techniques have enabled storage vendors to create storage systems that are both high-performing and reliable; possessing a useful life that easily equals or exceeds traditional storage systems. VMware In general DCIG puts a lot of emphasis on advanced software features in the DCIG 2013 Memory Storage Array Buyer s Guide. This is especially true of integration with VMware storage-related APIs such as (vstorage APIs for Array ) and ( Storage APIs for Storage Awareness). These APIs that can dramatically improve overall performance when implemented by storage vendors. Given the wide adoption of VMware in small, midsize and enterprise organizations it follows that they are seeking hardware that can take advantage of the force multiplication these API s provide for their existing and future VMware deployments. The good news is that 74% of the arrays support all of the 4.1 API s, and 47% support and v5.0 s capabilities. However, only 26% support the full set of v5.0 features. Those organizations embracing VMware as their primary hypervisor will want to pay particular attention to how an array s VMware support maps to their requirements. Memory Storage Arrays are Data Center Capacity and Efficiency Multipliers Every data center is designed to provide three primary resources: rack space, power capacity, and cooling capacity. As a company s requirements for computational and storage resources grows, they eventually run into the capacity limits for one or more of these resources. Memory Storage Arrays especially when combined with the latest generation of servers can act as data center capacity and efficiency multipliers. Rack space. Memory Storage Arrays provide many times the IOPS performance per rack unit (IOPS/rack unit) than is possible with traditional storage arrays. Some flash memory storage arrays can also provide more raw and/or effective capacity per rack unit. Either way, flash memory storage arrays can free up considerable rack space in the data center. Power. Memory Storage Arrays can be more power efficient than traditional storage arrays. This is especially true with regard to I/O performance (IOPS/watt). Some 2013 DCIG, LLC. All rights reserved. 8

11 DCIG 2013 The Insider s Guide to Evaluating Memory Storage Arrays comparisons show it is possible to meet high I/O requirements with a Memory Storage Array while using just 1/10th the power of an equivalent performing legacy storage array. Cooling. Power and cooling requirements move together. As power utilization increases, so does the required cooling capacity. A storage array that draws less power will also throw off less heat and require less cooling. For companies that are considering building another data center because they are approaching space, power, and/ or cooling capacity in current facilities, the flash memory multiplier effect may enable some companies to delay or entirely eliminate the need for a new data center. The avoidance of such a large capital expense, and the management time associated with locating and building and operating another data center, could make for a very attractive ROI. Performance and Pricing Two factors that strongly influence buying decisions are performance and cost. So it may come as a surprise to those who look at this DCIG 2013 Memory Storage Array Buyer s Guide to see no performance benchmarks as to how any of the flash memory arrays performed and only high level pricing information. There are three reasons why DCIG elected not to include performance or detailed pricing information in this Buyer s Guide. Performance will vary according to data center environment, the data being stored and implementation decisions. Vendors make IOPS claims ranging from the sub-100k to beyond 1,000,000 IOPS. Some vendors readily admit that they chose tests that provide an optimal IOPS number, while others claim to base their numbers on actual production workloads at client sites. Similarly, most arrays support multiple RAID options that affect both storage capacity and throughput. The same is true of the potentially large impact of deduplication and compression on effective versus raw storage capacity. So to introduce any type of performance metric would only result in the analysis and evaluations of the flash memory arrays included in this Buyer s Guide becoming more subjective, not less. Too many products to try to test to produce a Buyer s Guide in a timely manner. The intent of this Buyer s Guide is to provide a near real time snapshot of the flash memory arrays currently available in the marketplace. If DCIG had tried to test and establish performance benchmarks for all of these flash memory arrays, the next generation of flash memory arrays from all of these storage providers could well be available before the testing was completed, making this Buyer s Guide obsolete before it ever saw the light of day. DCIG did make some allowance in this DCIG 2013 Memory Storage Array Buyer s Guide for pricing as each flash memory array data sheet provides a starting list price for that flash memory array. These list prices had no bearing in the weighting of any flash memory array and are provided for reference purposes only. DCIG recognizes that features like price and performance are relevant and even key considerations when buying a flash memory storage array. However it is also almost impossible for a third party like DCIG to objectively measure these features on a large scale. Therefore testing is a part of the buying process that that DCIG believes is still best left to end users. It should be noted that the scoring method DCIG awards points for features we believe will matter to many end users. The arrays that score better are likely to be a good fit for a broader range of use cases and integrate well into a wider range of data center environments. It is also logical to conclude they will generally cost more. Reliability The Buyer s Guide only attempts to provide some insight into a flash memory array s reliability by examining whether or not it has redundant hardware, hot swappable components and what type of technical support it provides. All of the arrays include redundant fans, power supplies and flash memory storage modules. Nearly all arrays in the Buyer s Guide support hot-swap of all three components, though a few featured redundant, but not hot-swap fans. The combination of redundant hardware and flash optimization techniques, has enabled storage vendors to create storage systems that are not only high-performing, but that are also highly reliable; possessing a useful life that easily equals and exceeds traditional storage systems DCIG, LLC. All rights reserved. 9

12 DCIG 2013 The Insider s Guide to Evaluating Memory Storage Arrays Feature Areas Where DCIG Expects to See Improvement Less than half (43%) of the arrays support thin provisioning. Because flash memory is more expensive per gigabyte than traditional storage, the ability of thin provisioning to allocate storage capacity to those volumes that actually need them on a just-in-time basis should make the arrays that support thin provisioning more cost effective to acquire and require less management effort over time. A review of list prices indicates that arrays priced under $50,000 are just as likely to support thin provisioning as their more expensive brethren. VMware Although 74% of the arrays support all of the 4.1 API s, less than half support and only 26% support the full set of v5.0 features. As previously noted, this integration is key to the software defined data center and to minimizing ongoing management overhead for the large number of data centers that utilize VMware. Robust VMware support is, therefore, a product differentiator that will matter to many potential array purchasers. Memory Optimization While 85% of the arrays support and, two important techniques that enhance both long term performance and the life expectancy of flash memory; only 35% implement in-line deduplication, and just 4 arrays (12%) support block level compression. provides both performance and cost savings by only storing one copy of identical data. Alone, or in combination with, deduplication can achieve an effective storage capacity that is multiples of the raw capacity. If deduplication and compression occur in front of the cache, they can also multiply the effective capacity of RAM caches or other performance optimized data spaces, increasing the hit rate for that cache and the overall performance of the array. Cloud Storage The last area DCIG expects to see growth is in cloud storage integration. Although seven arrays (20%) integrate with cloud storage, they are all from a single vendor Whiptail. Whiptail s Accela and Invicta arrays support both the Amazon S3 cloud and their own proprietary cloud storage solution. When it came to structured object storage, although 60% of the arrays support proprietary REST APIs, none currently support either the Cloud Data Interface (CDMI) or Content Interoperability Services (CMIS). The initial release of the CDMI and CMIS standards occurred in 2010, and both are still in their early 1.x releases. for one or both of these standards may become an important purchase criterion in the years ahead. DCIG Observations and Recommendations on the Rankings Observations: Best-in-Class Ranking The Tegile Zebi HA2800 earned the Best-in-Class ranking among the Memory Storage Arrays that DCIG evaluated. In comparing this solution to its counterparts, this solution stood out in the following ways: The best balance of strengths acrossall scoring categories. Recommended or Excellent in five of the six scoring categories. for all evaluated VMware features. Large RAM cache of 192 GB,matched only by runner-up PureStorage. Recommendations The Tegile Zebi HA2800 is the latest array from Tegile, and the first to support what Tegile calls Vols. Vols are volumes that are pinned in flash memory so applications run at maximum performance without the potential for delay due to caching algorithms or tiering policies. Although the Zebi HA2800 is an all-flash array, it does have a hybrid twist in that Vols are automatically replicated to Zebi hard drives in a (required) expansion chassis. Tegile combines a Redirect on Write (ROW) file system with a proprietary Metadata Accelerated Storage System 2013 DCIG, LLC. All rights reserved. 10

13 DCIG 2013 The Insider s Guide to Evaluating Memory Storage Arrays (MASS) technology that separates metadata from the primary data path and stores the metadata on highspeed memory devices with optimized retrieval paths. This optimizes functions such as deduplication, compression, RAID and snapshot pointers, while simultaneously accelerating all I/O. Tegile claims their existing clients see a real world 3x-5x effective capacity from deduplication and compression. Because deduplication and compression occur in front of the cache, they also boost the effective capacity of the already impressive 192 GB of cache. The Zebi HA2800 includes excellent data protection features that support per-lun automated snapshot policies and multiple replication scenarios. The Zebi s web-based management interface speeds deployment by making it relatively simple to define a particular LUN as backup target, virtualization store, database store or file server store with a single click. Another important aspect of simplified storage administration is VMware integration. The Zebi HA2800 proved excellent in this category--supporting all VMware features evaluated by DCIG. The Zebi HA2800 provides unified storage through concurrent support for both SAN and NAS protocols. This fits with Tegile s goal, to deliver a new generation of storage arrays that address all of the demands of the modern data center, backed by expert customer support. Tegile Company Profile: The Tegile management team has a track record of working together to build successful companies and products. The CEO and CTO worked together in similar roles at Perfigo. The management team also has experience at HP/3PAR, Pillar Data Systems, Cisco and Cobalt Networks. Tegile was founded in 2010, and is privately funded by August Capital. Recommended Ranking Observations: The PureStorage FA 320 shares the Recommended label with Nimbus E-Class and Nimbus S-Class array. The flash memory storage arrays ranked as Recommended in this Buyer s Guide generally shared the following characteristics: Top of the heap and Hardware scores. Thin provisioning and multiple snapshot options. Good VMware. all evaluated 4.1 APIs, but only some 5.0 APIs. for at least the top four flash memory management techniques, including in-line deduplication. Recommendation: The PureStorage FA 320 leads the overall Recommended group. It captured Best-in-Class in three of the six scoring categories, but was hurt by a lack of and v5.0 API support. The five year warranty that PureStorage includes with the FA 320 is the longest of any array included in this Buyer s Guide. Like the overall Best-in-Class Tegile Zebi HA2800, the PureStorage FA320 supports both inline deduplication and compression along with 192 GB of cache in front of its 22TB of raw flash capacity. The PureStorage FA 320 is strictly a SAN. It does not support NAS file protocols. PureStorage Company Profile: The PureStorage management team has extensive executive and technical leadership experience at technology companies from A-to-Z including Akamai, NetApp, Quantum, Veritas, Xiotech, and Zimbra. The PureStorage team has expertise in flash memory, file systems, data structures, I/O optimization, and in building successful products and companies. Founded in 2009, the company is venture funded. The Nimbus Data S-Class Memory arrays captured Best-in-Class honors for support and Hardware. They were beaten in only by PureStorage, tying in that category with the Tegile Zebi HA2800. The Nimbus arrays were also outscored by PureStorage in VMware integration and SSD DCIG, LLC. All rights reserved. 11

14 DCIG 2013 The Insider s Guide to Evaluating Memory Storage Arrays The Nimbus Data E-Class and S-Class Memory arrays provide unified storage through concurrent support for both SAN and NAS protocols. This facilitates their broadest possible use in the data center. Nimbus arrays include support for host attached Infiniband storage networking, unlike other Recommended products. Nimbus Data Company Profile: The Nimbus Data web site lists just the CEO/Founder and several investors/ advisors as the management team. The CEO has executive experience as the founder of TrueSAN Networks and in marketing at Network Computer. Nimbus Data was founded in The company is internally funded and has a track record of profitability. Excellent Ranking Observations: An Excellent overall ranking was led by the PureStorage FA310, followed by the Violin Memory 6000 Series arrays and the GreenBytes IO Offload Engine (5,000 VDI Model). The flash memory storage arrays ranked as Excellent in this Buyer s Guide generally shared the following characteristics: Scored as well as the overall Recommended arrays in the and VMware categories, but had lower Hardware scores. Achieved only a Good or Basic ranking in at least three of the six scoring categories. and multiple options. At least 96GB of Cache. Both Fibre Channel and connectivity. Concurrent FC/, except for PureStorage. SAN only, though Nimbus provides unified SAN/NAS storage. Recommendation: The PureStorage FA310 is the single-controller sibling of the dual-controller FA320. It is identical to the FA320 except that it supports exactly half as much cache, flash memory capacity, and storage network port counts as the FA320. Being a single-controller array, the FA310 is appropriate where high availability is not required or is being achieved above the array. Inline deduplication and compression should multiply the effective capacity of the maximum 11 TB raw flash memory capacity. The Violin Memory 6000 Series arrays picked up extra points for their support of VMware vstorage APIs for Storage Awareness (). These APIs give vcenter visibility into the capabilities of the underlying storage hardware. Infiniband support also gave their hardware scores a lift. Violin Memory Company Profile: The Violin Memory management team includes several former executives from Fusion-IO, including the former CEO. Like PureStorage, the team has an impressive technical and management background and a track record of building successful products and companies. Violin Memory was founded in 2005 and has gone through four rounds of venture funding totaling $180 million. The GreenBytes IO Offload Engine (5,000 VDI Model) is marketed to a specific use case virtual desktop infrastructure as a complement to existing storage. Unique among the Excellent arrays, the IO Offload Engine supports VMware s feature. Also unique among the arrays DCIG evaluated, the IO Offload Engine is available as a standard CAPEX purchase or on a pay-per-use OPEX model. The OPEX approach may appeal to service providers and some enterprises that would be put off by the sticker price. GreenBytes Company Profile: The GreenBytes management team has executive experience at successful startups that were later acquired by recognized entities like Agfa, CompuWare and VMware. Founded in 2007, the company is venture funded. Notable investors include Generation Investment, a firm co-founded by Al Gore. Good Ranking Observations: Arrays achieving a Good overall ranking include the Huawei Enterprise OceanStor Dorado5100, Whiptail s Invicta and Accela arrays, and the Violin Memory 3000 series arrays DCIG, LLC. All rights reserved. 12

15 DCIG 2013 The Insider s Guide to Evaluating Memory Storage Arrays The flash memory storage arrays ranked as Good in each classification of this Buyer s Guide generally shared the following characteristics: Concurrent support for Fibre Channel and SAN protocols. No support. memory management includes and, but not Inline or. Recommendation: Whiptail s Invicta and Accela arrays took Best-in- Class honors in VMware integration by supporting all VMware storage-related APIs evaluated by DCIG. Whiptail arrays are also distinguished among the Good arrays by support for concurrent SAN and NAS protocols to provide a unified storage capability. The Violin Memory 3000 series also feature solid VMware integration, with support for all evaluated 4.1 APIs but not 5.0. The arrays from both Whiptail and Violin Memory also have Good to Excellent hardware scores, but are weak on and Replication. This combination of strengths and weaknesses may work well for organizations relying on VMware s native management and replication features. The Huawei OceanStor Dorado5100 strengths are a mirror image of the other Good arrays. The Dorado5100 scored in the Recommended range in three categories, but was held back by a lack of VMware integration and support. The Dorado5100 does not support deduplication or compression, but this is offset by a maximum raw flash capacity of 38.4 TB. Huawei Company Profile: Huawei is a Chinese multinational networking and telecommunications equipment and services company headquartered in Shenzhen, Guangdong, China. It is the largest telecommunications equipment maker in the world. Basic Ranking Observations: The flash memory storage arrays ranked as Basic in each classification of this Buyer s Guide generally shared the following characteristics: only a single SAN protocol ( or Fibre Channel) No support for NAS protocols Limited management and replication tools Weak application layer support The Basic ranking was shared by arrays from four vendors: Astute Networks, Huawei, Kaminario, and Texas Memory Systems, an IBM company. Astute Networks ended up with a Basic ranking in a sense on purpose. The Astute Networks ViSX G4 is marketed as an application acceleration appliance. The ViSX G4 was designed to work well in a virtual server environment, where nearly all of the storage related management would be done from the point of view of the server administrator. Indeed, the VMware score of the ViSX G4 is bested only by arrays from Whiptail and the Best-in-Class Tegile Zebi HA2800. According to Astute, the ViSX G4 does NOT replace or compete with full featured storage arrays by design. In that regard it was intended to be added where existing SAN storage wasn t meeting application performance needs. Therefore, Astute avoids supporting redundant controllers and advanced multi-pathing features on purpose. Those add cost without adding value in their use case. Indeed, their target customer has a multi-pathing solution in place and wants to keep using it. As a result,in a checklist comparison the ViSX doesn t stack up like a storage array but it might be a good and cost-effective solution to an otherwise stubborn SAN performance problem. Astute Networks Corporate Profile: The executive management team at Astute is comprised of leaders from the virtualization, server, storage, data protection and data center infrastructure markets. Astute was founded in 2000, and is venture funded DCIG, LLC. All rights reserved. 13

16 DCIG 2013 The Insider s Guide to Evaluating Memory Storage Arrays The arrays from Kaminario and Texas Memory Systems were held back by their absolute lack of VMware integration, a zero score in the, and for Texas Memory Systems the lowest scores in the and Replication category. If Kaminario and Texas Memory Systems intend their arrays to address niche applications, the status quo may be fine for them. If they want their arrays to be in the mainstream of the movement to replace legacy storage, we believe they need to do development work in VMware, and Replication, and support. Kaminario Corporate Profile: The Kaminario management team includes technical and management experience at a variety of tech companies including BlueArc, Dell, EMC, NetApp, IBM, XtremIO with expertise in storage systems, networking, operating systems, BI and data processing. Kaminario was founded in 2008, and is venture funded. Texas Memory Systems Corporate Profile: Texas Memory Systems built its reputation as a company owned and operated by engineers. TMS was founded in 1978, and has long prided itself on delivering to market The World s Fastest Storage. Texas Memory Systems was acquired by IBM in The Huawei Enterprise OceanStor Dorado2100 shares features with the Dorado5100, but is a Fibre Channel-only device that maxes out at 9.6 TB of flash memory capacity versus the Dorado5100 s 38.4 TB. Recommendation: Understand what you need, and then let your needs determine which solutions are the best for your data center. A Basic array that supports only a single SAN protocol ( or Fibre Channel) and no NAS protocols, or an appliance that is designed to integrate into rather than replace your existing SAN, may just address the need that prompted you to begin looking at Memory Storage Arrays in the first place DCIG, LLC. All rights reserved. 14