What an Interface Says About an SSD

February 1, 2011

When an SSD manufacturer brings a product to market you don’t need to look any further than the interface between the SSD and the server to understand its target market. Solid state storage systems are available with a wide array of sizes, shapes, densities, media, performance, cost and interfaces. The interface used gives the best hints as to how the manufacturer predicted the product would be used and more specifically which market they are targeting.

Fibre Channel SSDs are aimed at the enterprise data center. For most of the last decade, Fibre Channel has been the interface of choice for Tier 1 disk drives and the main interface for attaching external storage arrays in most data centers. Interestingly, the Tier 1 disk drives are now migrating to SAS, but the predominant interface for the enterprise storage array to the server is still Fibre Channel. Companies developing Fibre Channel SSDs want to appeal to enterprise data centers who have made major investments in Fibre Channel based storage area networks. There are plenty of predictions about the demise of Fibre Channel in the data center, but if you were making a choice about an interface for the enterprise today, you would offer Fibre Channel first. If I were deciding the next interface for an SSD or a storage array, I might go with FCOE, but I would probably wait to see that market develop further first. The rapid introduction of converged network adapters (CNA) could translate into changes at the storage controller, but I would also wait to see what happens in that arena.

InfiniBand SSDs are aimed at the high performance computing (HPC) market. InfiniBand is touted for its high bandwidth per link and its low latency. For SSDs with large backplanes, an InfiniBand (IB) controller is a good way to tout your bandwidth capability. Yes, I know there are other companies using IB outside of the HPC market, but the bulk of big opportunities for IB SSD are in that space today. I broadly define HPC to also include oil & gas and entertainment industries. I do believe that IB attached SSDs are an interesting option for data warehousing applications where bandwidth is more important than IOPS.

 NAS SSDs are aimed at the middle of the enterprise. This segment is one of the more intriguing to watch. A couple of companies have made credible attempts to develop NAS caching solutions which sit in front of existing NAS and provide a read or read/write caching layer. In a future blog, I might examine the challenges these companies face. As with mainstream Fibre Channel attached storage, the NAS vendors have incorporated SSD as a storage tier. Only one vendor comes to mind that is doing a pure SSD NAS solution, but others are likely to follow. NAS solutions are so much about software that it is harder for a new company to enter this space and compete with the incumbent suppliers.

iSCSI SSDs are aimed at the low to middle of the enterprise. This has not been a terribly active segment for pure SSD solutions but interesting options are on the horizon. Clearly, existing iSCSI storage arrays have options for including hard disk based SSDs. My automatic expectation when I look at an iSCSI solution is that it will be less expensive than a Fibre Channel SSD. The main reason I would offer iSCSI is to target the cost-sensitive part of the market. Given the increased availability of 10Gbit Ethernet and advanced TCP off-load engines, it is quite reasonable for an iSCSI SSD to offer good performance.

Internal PCI SSD. There has to be an exception to every rule and PCI SSD may be the exception to my rule about an interface telling you about the application for an SSD. PCI SSDs cover a wide variety of price ranges, capacities, media, performance and reliability. On the high end, there are a bunch of applications, particularly scale-out applications, which are server-centric and not storage network centric. PCI SSD have had tremendous success in this category. Similarly, for companies with smaller data sets and budgets, PCI SSD can be alluring. It is not a stretch to pitch PCI SSD for prosumer or high end gaming customers.

External SAS SSD. There are very few externally attached SAS SSDs on the market today. I think the people who offer them were probably temporarily delusional about the future role of SAS in the market and its ability to get rid of Fibre Channel for storage networking. This is not to say that SAS is a bad interconnect, in fact it is being effectively used to replace Fibre Channel as the backplane for many modern storage arrays (i.e. the connections between a disk controller and its enclosures are increasingly SAS).

Hard Disk Drive (HDD) Form Factor SAS SSD.  With the help of solid state storage, SAS HDD has killed the Fibre Channel disk drive. Hard drive form factor SSDs with SAS interfaces are more likely than not intended to be sold to a storage or server OEMs. For the storage OEMs, they replace their Fibre Channel SSDs (if they were ever offered). For the server OEMs, SAS SSD may be used as a boot drive.

SATA SSDs are aimed at the consumer, prosumer, gaming and small business markets. I cannot currently see an enterprise market for SATA SSD. In enterprise storage arrays, SATA HDDs are only used to offer the 3.5” high density (slower) drives.

External PCI. There are a few varieties of external PCI offerings including devices that are ground up designed to offer external PCI SSD and others that are I/O expansion chassis that can be loaded up with PCI SSD. My personal opinion is that the genesis of the external PCI SSD was to serve as extended memory for servers at a time when server memory capacities were limited and at high densities extremely expensive. In my experience, the only way to make one of these devices useful for traditional data centers is to put the external PCI chassis behind some other storage gateway. The storage gateway attaches to the storage network with Fibre Channel. This is all good, but the gateway is now the main dictator of your performance characteristics.

The story on SSD interfaces is certainly not complete. Innovative companies will capitalize on new markets and new interfaces in ways that we cannot yet predict. For the innovators in these segments lie new markets and new opportunities.


Application Owners vs. Data Center Operators

January 12, 2011

This blog seeks to articulate a perceived divide between the application owners and the data center operators and then explain how this rift has impacted the solid state storage market’s past, present and future.  I will close by predicting the real winner in this market. 

The divide between these groups starts in their background, their education and their experience.  The application-side has generally come up through the ranks as either business analysts or developers.  If they completed a college degree, they are more likely to have come from business or management information systems types of programs.  It is often the business analyst whose job it is to understand the business and map the business requirements to a custom software project or to be used in software selection.  Some of these folks have risen through these roles to become project managers or product managers.  The people with the strongest business skills document requirements, make great testers and are often skilled implementers.  The people with stronger technical aptitudes usually become developers, programmers, database administrators and application architects.  Together, these business analysts and systems analysts collectively represent the application-side.  Most IT consultants one meets are on the application side of things.

The data center sort of person has, more often than not, come up through the ranks as a system, network or storage administrator.  If they have completed a college degree, they are more likely to have an engineering background though I have seen a wide range of backgrounds in this field.  It is the system administrators that understand, better than anyone else, the practical impacts of the hardware choices dictated by application choices.  The system administrators frequently move on to take titles like “infrastructure manager” and “data center manager”.

People from either side can move into CIO roles, but the biases of their previous experiences can be difficult to separate from their decision making strategies.

In addition to different career paths, the two sides tend to have different objectives.  At the most basic level, the application owner is often driven to generate profits by maximizing features (more powerful queries, preference for real-time, faster applications, and features that enable process re-engineering).  The data center manager, on the other hand, can be driven to generate profits by minimizing costs and reducing risks (simplified management, high reliability, and standardization).

In the context of this complicated background, where do solid state storage devices fit, how do they become a part of the story?  As you might suspect, the answer is related to performance.  The pain of performance bottlenecks is first felt by the application owners. The application owners receive complaints from internal and external customers any time performance is thought to be slow.  In order to improve performance, the application owners invest a great deal of time in their code – they hire DBAs, who tune their code, improve their SQL statements, and change priorities, restricting application features that are less important.  If these actions don’t work, the application owners pressure the data center operators to move the applications to bigger servers, in the process adding more processors, more RAM, more disk drives, or adding more storage caching memory.  Data center operators naturally trying to protect their positions, sometimes blame poor performance on poorly written code.  Application owners, on the other hand, tend to blame performance on the hardware.  So what happens when the hardware is optimized, the code is as tuned as it can be and performance is still poor?  Generally speaking, performance stagnates unless features are dropped or hardware is refreshed.

In the thirty years since Solid State Disks entered the market, most SSD manufacturers quickly learned that their customers were the application owners.  As a result, the manufacturers geared their pre-sales, sales, marketing and product features around serving application owners.  By necessity, pre-sales teams became expert in conducting performance analysis for operating systems, file systems and even databases.  Sales teams learned to develop “champions” on the application side of the business.  Advertising and marketing programs were aimed at database and application audiences more successfully than storage audiences.

Slowly, the data center operators started to become interested in solid state storage.  Some cared because these SSD products were adding complexity to their data centers.  Others cared because they were concerned about losing control of hardware decisions.  Some cared because they could see the benefits of SSD for their infrastructure.  A mix of strategies began to unfold in the data center as the more adept data center operators maintained tight controls on technology standards by serving as the testing ground for SSD options.  As the data center operators became more engaged in SSD analysis, the big storage manufacturers started paying attention.

The big storage manufacturers finally entered the SSD market in 2008, targeting their primary customer, the data center operator.  The data center operator is not usually focused on accelerating one application; they want to accelerate their infrastructure.  They want their centralized, reliable and easily managed storage environment to get faster.  The big storage manufacturers focused on making the introduction of SSD simple for the data center operator.  It has become a sort of mantra “just add Flash SSD hard drives to the hard disk enclosures and provide some tools to make it easy” for the data center operator to move data between tiers of storage.  Over time, the manufacturers have made it so that these systems can even dynamically migrate data between storage tiers.  Why focus on accelerating one application when we can make everything faster?

Both SSD manufacturers and big storage manufacturers remain true to their customers, but neither has done much to sway the other’s customers.  The application owners, who, if the truth be told, would rather not have their mission critical business application on virtualized servers and centralized storage are much happier with a dedicated SSD for their application.  In head-to-head testing, they can observe that the pure SSD manufacturers, who have always focused on decreasing latency and increasing throughput, have the edge when it comes to the number one thing that they care about – making their application faster.  The data center owners, who are the people called on to actually install and support SSD solutions, can see that the integrated solutions offer what they care about – lower risk and easier management.  They can also see that the big integrated solutions offer “good-enough” performance for many users.

Today and, I would predict, for the next several years, the SSD market will be split.  Pure SSD manufacturers will continue to grow by solving application performance problems better than integrated storage manufacturers.  The customers buying these solutions will continue to be led by application owners.  Integrated storage manufacturers will rapidly grow market share by offering solutions which accelerate entire infrastructures (think centralized storage environments with dozens of applications and virtualized server environments).  The customers buying these solutions will be led by data center operators.  Thus, the great divide between the application owners and the data center operators will continue for the foreseeable future.

The next few years of R&D could reduce the divide somewhat.  Will the pure SSD manufacturers add storage services and reliability features equivalent to the big storage manufacturers or do they maintain their place in the world by widening the performance gap?  Will the big storage manufacturers decrease controller, cache and backplane latency or increase the divide by offering more storage services?  My bet is on …  Well as you may have guessed, I am hedging my bets a bit by working in an environment where I can evaluate a customer’s requirements, determine the fit for either pure SSD or big storage with SSD, and recommend the right solution.  In the end, the end-user wins because the introduction of any sort of SSD into their enterprise will make their applications faster.


Welcome to AppICU

December 21, 2010

AppICU was conceptualized years ago.  It started as an idea for a TMS consulting practice.  It spent some time (in my head) as the name of a new business.  I suppose it is fitting that the name finally makes its first public appearance as the title for my blog.  If you are a marketer, which was my primary responsibility for the last ten years, it is almost impossible to find the time or energy to blog about products that you are already marketing.  If it is such a great blog idea, why isn’t it a whitepaper or submitted for publishing?  Do you blog to bash competitors products?  Do you blog to brag about your cool new product/service?  Do you blog to become famous?  My goal is to help provide guidance to the buyers of application acceleration solutions or solid state storage by providing clarity where there is chaos.  Along the way, I hope to influence the companies that develop and market to these buyers.

The market for application acceleration and solid state storage is well served by a variety of other writers/bloggers and analysts.   You won’t find anyone promoting the solid state storage industry better than Zsolt Kerekes at StorageSearch.com.  Zsolt publishes more original content about SSD than anyone I know.  Without his independent view of the market, SSD buyers would be lost.  If you are looking for analysts who know their stuff when it comes to SSD I encourage you to follow:  Jeff Janukowicz at IDC, Joseph Unsworth at Gartner, Greg Schulz at StorageIO, Robin Harris at Storage Mojo, Ray Luchessi at Silverton Consulting, Jeff Boles at Taneja Group, and George Crump at Storage Switzerland.  These guys all suffer through endless hours of vendor fluff to distill nuggets of useful information to pass on to their customers/readers.

Welcome to my blog.  I hope it adds to the discourse and proves to be a good use of your time.


START for SSD Marketing

December 20, 2010

As the United States looks to approve the START treaty, I thought it was time to propose that SSD manufacturers enter their own strategic arms reduction treaty to control the rampant and destructive proliferation of million IOPS marketing.

The road to an IOPS arms race began innocently enough, SSD manufacturers had a novel story to tell.  The IOPS from hard disk drives have been atrocious since the dawn of computer time.  Even today, the lowly hard drive can only squeak out 300 random IOPS.  From the earliest days of the SSD, IOPS marketing was a big part of the story.  The beauty of a solid state storage device was that it could move more data (IOPS) to a processor faster (less latency) than traditional disk.  This simple story has been at the core of the SSD value proposition for 30 years.

Admittedly, I fired the first shots (and probably the second, third, fourth….) in the escalating IOPS arms race in 2001 when Texas Memory Systems announced the RamSan-520 a 5U monster of a system with all of 128GB of RAM capacity.  This system with fifteen 1Gbit Fibre Channel ports was said to deliver 750,000 random IOPS.  Do you have any idea what kind of reaction that generated at storage conferences in 2001.  Wow!  Impossible, most would say.  This process led to TMS proudly declaring itself the “World’s Fastest Storage®”.   After firing this weapon for nearly ten years, I have to admit the time has come to stop the IOPS marketing arms race.  The challenge in 2001, as it is today, is finding the customer that can drive 1,000,000 IOPS.  Actually, in 2001 finding a server to drive that many IOPS was impossible.  The processors, operating systems, host bus adapters, etc. were all too slow.  Fortunately, the imposition of Moore’s law on electronics led to breakthrough after breakthrough enabling SSD manufacturers to demonstrate high IOPS with single server configurations.

I like to think, but hesitate to admit, that in addition to pioneering million IOPS marketing, TMS also drove the widespread use of IOMeter (a tool used to test storage devices by generating IO) at trade shows.   As a storage marketer my grandest dream was to find a customer that needed to run IOMeter as their business application.  What a perfect customer this would be.  I searched the world over.  Strangely, the financial exchanges didn’t need IOMeter to complete trades.  Telecom companies didn’t need it to bill cellular customers.  Who could possibly need IOMeter for their business?  Imagine my glee when host bus adapter manufacturers and switch manufacturers started caring about IOPS as a marketing tool.  Finally, my dream customer had arrived.  My apologies to storage industry exhibit hall wanderers; I too have tired of seeing IOMeter.

This brings us back, admittedly after a brief tangent, to million IOPS claims.  I hesitate to examine the number of vendors that are persistently and proudly proclaiming profound performance.  One million IOPS.  Yawn!  Is that all you’ve got!  In fact, I would argue if the extent of your marketing message is your IOPS you don’t have enough… marketing talent.

Using a solid state storage device is about a lot of things: application acceleration, lower power consumption, enabling business growth and solving mission critical problems.  Tell us the customer stories.  How many 1,000,000 IOPS customer stories have you read?  Hmmm.  Fair enough, 1 million IOPS sounds like so many that people will stop worrying about whether the storage device can meet their production performance requirements.  But can we stop at 1 million? Perhaps we should shoot for 2 million.  No.  It is time for the SSD IOPS marketing proliferation to be stopped while the customers still care.   Start designing systems that satisfy the range of customer buying requirements:  low latency (the number one reason most customers benefit from SSD), good-enough IOPS, bandwidth suitable to the application’s goal, five 9’s reliability, low mean time to repair, low power consumption, interoperability and low total cost of ownership.