How “autonomic” do you want tiered storage to become?

Read More >>

Most storage folks refer to the growth of data as being the today’s primary storage challenge. However, other than CapEx and Opex, it is the byproducts of this growth, management and accessibility, which are the defined, operational, data center pain points.

When optimizing storage resources the objective has to be the continual matching of the value and requirements of the data with the value and performance of the storage. Since the early days of HSM, resolving this challenge has been the motivation behind storage tiering. There is no contention among storage professionals about this point; they all desire and most strive to put the right data on the right storage.

Tiering technology has come up against barriers in adoption mainly due to administrative complexity and overhead. Elimination of these barriers is the catalyst driving advances in automation with the benefits of autonomic architectures being the ultimate end goal.

An autonomic architecture is a very lofty goal but it is achievable and is, I suggest, the way of the future. The few solutions that are currently in play that implement true autonomic thinking illustrate the unquestionable value such architecture can be to the user community.

Defining Performance Starved Applications?

Read More >>

In a couple of my earlier posts I have referred to performance starved applications. To give some insight into what I am talking about I thought a brief post on the subject was warranted.

Performance starved applications are those applications where performance is negatively impacted because it is not receiving sufficient IO to deliver optimal performance. Examples would be transactional heavy databases, eCommerce applications, analytics (financial, business, seismic), exchange, BI applications, virtual environments etc.

Some of the more common symptoms of a performance starved application being;
1. Applications hit a performance ceiling
2. Heavy peak loads that negatively affect response times and QoS
3. Unpredictable workloads that negatively affect response times and QoS
4. IT resources deployed to simply sustain acceptable performance levels.

So what can cause normally high performance applications to become performance starved? As with many issues there are a number of potential causes. They can be either;

a. CPU I/O bound or
b. Disk storage I/O bound or
c. Or some other reason such as an infrastructure bottleneck or an application that is handicapped because it is single threaded.(more common than you may think)

When checking on application performance issues the first check should be the CPU performance, particularly its utilization. If the CPU is running with an average utilization greater than 50% the problem is most likely that the application is CPU bound. However, if the average CPU utilization rate is below 50%, the CPU is probably in a wait state created by something other than the CPU; the lower the utilization rate the more likely this is to be true. If the CPU has been eliminated as the culprit the next area to focus on would be the storage.

With storage first look at queue depth which is the number of I/O request waiting to be serviced by the storage system. A key variable that influences queue depth is the number of drives or spindles that are available. The more spindles the greater is the ability of the storage subsystem to respond to I/O request. This is why highly dense packaging such as the Atrato V1000 delivers a high access density by packing 160, 2½” drives into a 3U, rack mountable envelope. This is a configuration that can deliver a very high IOP performance increasingly demanded by performance hungry applications such as found in virtualized environments.

However adding more drives to reduce queue depth can be problematic. Simply adding more spindles to get I/O performance can be expensive and if the application does not need the capacity -- a waste. Short stroking expensive FC drives to improve IO performance a great example of what I call an unnatural and wasteful act.

However if the queue depth is zero but the bottleneck is still storage what is the issue? The issue must be latency and the obvious solution is a faster drive. But what if the fastest drives available are already installed? This is where SSD’s become attractive.

Will adding SSD’s help administrators eliminate storage IO bottlenecks?

Read More >>

Will adding SSD’s help administrators eliminate storage IO bottlenecks? The answer is probably yes if a true hybrid solution, with intelligent tiered management, is implemented but unlikely if the tiering management is a simple retread of traditional tiered management thinking.

So, it depends. First consider the data landscape the company has, which is highly dependent upon the applications in play. While SSD’s are a key element of performance tiering solutions it is the tiered management software that is critical element to an effective integration of SSD and HDD. When done correctly Hybrid solutions support mixed media vLUN’s that seamlessly, transparently and efficiently place data on the most effective media in a timely manner. This approach is especially effective for companies experiencing data access profiles that create what I call, performance starved applications.

Performance degraded applications that are caused by storage IO bottlenecks can include transaction heavy db applications, data mining, financial and business analytics, business decision systems, virtualization implementations etc. So in short, absolutely, SSD’s can provide a solution but with the caveat that it is the autonomic, intelligent tiered storage management software, not the physical media, that is the essence of a scalable, pragmatic, high performance solution.

It is the advanced tiered storage management tools that integrate the performance attributes of SSD or/and RAM and the capacity attributes of HDD. This managed integration of technologies delivers an interesting and effective blending of complementary storage attributes to address unique performance challenges.

Beware: Not all SSD solutions deliver equal performance, simply adding SSD into the storage hierarchy without the support of meaningful management software will not meet expectations.

Comparing Performance Tiering to Archival Tiered Solutions

Read More >>

Tiered storage is a technology that enables the storage administrator to match the value and requirements of the data with the performance and cost of the storage. However not all tiering solutions deliver the same value proposition, despite the marketing buzz.

Storage tiering has evolved over the past 12 to 18 months into what can be called performance tiering and archival tiering.

Performance tiering is where the data placement decision is performance driven and focus on optimizing the use of tier 0 and tier 1 storage to optimize application performance . It is a technology that can fully integrate and exploit the advantages of solid state disk.

Archival tiering is the more traditional and pragmatic view of storage tiering, think HSM. It’s focus is to free-up expensive Tier 1 storage, optimizing the usage of storage assets by simply demoting aged and inactive data to lower and less expensive tiers of storage with the lowest tier being tape offload.

What separates performance tiering from the more lethargic traditional implementations is granularity and responsiveness. Performance tiering continuously monitors data access profiles at a small block size (4k to 64K) and responds near instantly to changes in workload data access patterns. This requires performance tiering solutions to be highly automated and in at least one implementation reach the level best described as autonomic.

In comparison, traditional approaches look at much larger data chunks, or even at the file level, and use a batch approach to move data between tiers in time intervals to be determined by the administrator. In short there is a delay in the detection of data access profile changes and the movement of the data to a more appropriate storage. Ok for traditional tiering applications but not for performance tiering.

The primary objective of performance tiering is to deliver greater performance and eliminate storage IO bottlenecks whereas the primary objective of traditional tiering solutions is to more effectively manage storage assets. There are many solutions currently available but beware of tradition solutions that have been repurposed to include SSD technology but do not have the characteristics that can truly manage and exploit the significant performance gains that solid state brings to the storage hierarchy.

Tiered Storage Management: An Evolving Technology

Read More >>

Using mixed media to match the value of data to the cost of storage is not a new concept. Storage tiering is a constantly evolving architecture that has undergone drastic changes since its beginnings. Think HSM of the 80’s. However tiered storage management has bifurcated into two major, but dissimilar and complementary classes of tiered storage management.

1. Tiering for Archive: Increases efficiencies by improving the archiving process.
2. Tiering for Performance: Accelerates performance by promoting highly accessed data to a high speed media tier.

The traditional objective of tiering was to worry the basic concern of capacity management. The main goal was to preserve the highly expensive Tier 1 storage for the significant data that had value that matched the cost of the storage and migrating the less used or less valuable data to lesser expensive spinning storage or even tape while maintaining its “online” available to the end-user.

Times however have changed. Although capacity management is still a concern, arguably, the more pressing issue is improving application performance. Tiering to obtain the data access rate demanded by I/O hungry applications is at the forefront of the current tiering movement. An interesting anecdote was related to me recently. A CIO of a well known company in Denver countered our recent green presentation by bluntly stating the if he had to choose between using dirty coal fired generated power or sacrifice his SLA obligations, his unequivocal choice would be to build the biggest, dirtiest power station. Such is life’s priorities.

Tiering for Archive:
Migrating aging data to archive and freeing up expensive storage media was the original motivation for tiered storage management. This was a policy-based approach that identified data that has become dormant, lost its value and was ready to be archived. Hierarchical Storage Management (HSM) is the first technology to come to mind in this category and depending on criteria used to define the operational policies such as age, last date of use, file type, user, application etc, data can be tiered down to inexpensive and consequently slower tiers of storage. Such tiering however can be bidirectional with a policy based decision process determining what data should be once again promoted to a higher performing media. End of quarter or end of year financials is a good example of such variable value, mobile data. However this pre-fetching approach is not comparable to what has become known as performance tiering.

Tiering for Performance Acceleration:
Tiering, with the primary objective of accelerating performance is the latest evolution of storage tiering. This innovation has evolved over the past year or two, enabled by the evolution of SSD as a primary storage option. What has become known as Tier 0 has been with us for many years in the form of RAM based storage but the volatility and expense of that technology restricted its use to caching functions and eliminated any consideration of it as a viable option for primary storage.

So, performance based tiering consists of two tiers. Tier 1, the spinning disk tier and a high performance Tier 0, solid state disk (SSD). Despite my earlier comment RAM can contribute to a Tier 0 layer but generally when partnered with SSD otherwise it is restricted to traditional caching functionality. Unlike tiering for archive, tiering for performance (access speed) promotes active data to a faster tier instead of pushing dormant data to a lower tier. This effectively makes highly active data available in the highest performing media, solid state.

Eliminating the Management Complexity of Tiered Storage Management:
The challenge however that has plagued the reality of tiered storage management has been its management complexity. This is the Achilles heel of this innovative technology and has in many cases has compromised the effective integration of SSD resulting in the self defensive reaction of over provisioning.

One company that does have an innovative approach to solving this problem is Atrato's Inc. Unique to their approach is their Access Profiler, a function of their ApplicationSmart™ software suite. This feature uses advanced algorithms from a branch of mathematics known as Feature Vector Analysis, normally associated with applications such as facial recognition, to track access patterns for a given data set. By continuously monitoring the data stream, in real time, the Access Profiler is able to take the tedious task of data classification out of the hands of the IT administrator. In addition, movement of the data to Tier 0 SSD will happen whenever there is a perceptible change in the access pattern as deemed by the Access Profiler.

Autonomic Data Movement:
After the access profile of the data has been determined the Access Profiler will communicate the information to the Tiered Management Engine (TME). The TME will then make the appropriate adjustments to the solid state tier by promoting data from the spinning disk pool and writing over data in the SSD tier that no longer meets the access rate criteria. This autonomic process takes place without any user intervention. Should the user wish for greater control of data classification and placement, settings, multiple levels of control are available with the ability to turn ApplicationSmart™ functionality on or off at an application or LUN specific basis.

In contrast to archive tiering, performance tiering has to deal with both writes and reads. When migrating data from Tier 1 to Tier 2 or lower, there is no need to deal with write operations as the data in question already resides in a spinning Tier and has outlived its time in the high access category. However in the case of performance tiering, solutions that deal predominantly with performance acceleration, writes are most certainly a concern. To address this issue Atrato has developed what they call an Ingest Accelerator (IA). This feature creates an IO ingest log on SSD rather than directly onto spinning disk. This enables a rapid acknowledgment to and release of the client initiator, significantly increasing the IO rate. A second unique capability of the IA is advanced I/O reforming that optimizes the write time from the ingest log to Tier 1.

For more details check out the Atrato white paper “Autonomic Tiered Storage”

Not-so-Autonomic Storage Tiering

Read More >>

Autonomic is a word that is increasing being used to describe the ultimate in ease of use of data storage equipment with “autonomic tiering” having recently come into vogue. However, despite there being a plethora of tiering solutions available not all of them are as autonomic as they claim.

My first recollection of the use of the word in an IT context was IBM’s Self-Managing Autonomic Technology which they describe as enabling a fluid response to change like having the ability to manage themselves and dynamically adapt to change such as identifying and correcting problems often before they are noticed by IT personnel. In the context of a tiered storage solution this means the transparent movement of data between tiers with little or no human intervention.

To the purist (read non-marketer) many of the big name solutions lack the ability to perform truly autonomic data movement. Though they have the ability to move data programmatically or via a policy based process according to a data classification which is certainly a large step forward in the quest to optimize a storage environment, they cannot transparently (autonomically) move that data to the appropriate tier. In short they have not eliminated much of the management complexity that commonly accompanies tiered storage management.

These not-so-autonomic solutions require the administrator to define a policy as to when data should be moved (batch processing). If this policy based approach sounds eerily similar to the traditional HSM's of the past you are in good company. The value of an autonomic solution lies in its ability to be proactive instead of reactive. Without the power to perform data movement in conjunction with real time access patterns, a solution becomes purely reactive.

When an administrator is forced to make a best guess as to the most advantageous time to move data from Tier 1 to Tier 0, the placement of aged data is inevitable. This aged data may not be relevant to current access demands leading to cache misses, and latency issues. It is true that data with a predictable access pattern can be pre-loaded and available on SSD when required but this is far from autonomic.

These less than autonomic companies will tell customer that by requiring policy based batch processing, they are really giving administrators much needed control over their environment. The opinion of at least one company Atrato Inc, is that administrators should have the choice as to whether data should be manually or autonomically managed on an application/LUN specific basis. With data centers expanding into the petabytes, the storage model simply cannot support the do-it-yourself approach. Intelligent, transparent movement is, after all, the whole point of implementing a so called "autonomic solution".

The following list is of companies that offer a flavor of tiered storage management. It is not intended to be an exclusive list by any definition simply a sample. Atrato Inc, HP, EMC, Compellent, 3PAR, HDS, BluArc, and Avere.

Don’t let Virtual Desktop (VDI) “Storage Tax” take you by surprise!

Read More >>

My favorite phrase these days is that “server virtualization breaks traditional storage” and the more I talk with folks who are at the sharp end of a virtualization implementation the more I am convinced of its truth.

There are several reasons we can point to that are responsible for this dilemma. Robin Harris of StorageMojo coined a very descriptive phrase, the “IO Blender” to describe one of the issues, the mixing of multiple virtual IO streams resulting in the delivery of a single stream of random IO to the storage. This characteristic of virtualization neutralizes much of the IO management smarts that architects of traditional storage have developed over the years. A topic I am reserving for a future blog.

However what about implementing a Virtual Desktop Infrastructure and what are the strains that VDI places on the storage infrastructure?

Data growth is the most obvious impact. As each local desktop is replaced with a virtual desktop the local storage burden now transfers to the centralized storage pool. Note, local storage tends to be cheap consumer grade storage, data center storage is much more expensive. So, assuming that there is an average local demand of 10GB, multiply that by 1000 virtual user’s, that translates to an additional 10TB of storage that the data center storage administrator has to find. With the aggressive growth in unstructured enterprise data an average user requirement of 10GB may be an understatement. True there are many techniques such as tiering, thin provisioning, data deduplication and compression to help to keep data growth in check but it remains a challenge.

However, without demoting the significance of data growth perhaps the greater problem is data access or more accurately access density.

Access density is a measure of performance; according to Fred Moore “access density is the ratio of disk drive performance measured in IO/sec to the capacity of the drive”. In the case of a storage system it would be the total IOPS the system can deliver to the total capacity. Another factor that influences IO performance is drive utilization. As drives begin to fill with data, latency increases and eventually the system ability to deliver its maximum IOPS is impaired. As a side thought avoiding this problem is what short stroking is all about. Short stroking limits the amount of data stored on the drive to the outer bands of the drive platters. While this maintains maximum IO performance it results in only 20% or so of the disk capacity being used. Effective, but very expensive.

So what has this to do with VDI?

First let us assume that a desktop has an average IOP requirement of 20 IOPS. However when an application such as Visio is accessed the IO demand goes through the roof. Considering that laptops are designed to exploit the maximum IO its hard drive can deliver, my 20 IOPS assumption may be conservative. In an environment that has 1000 VDI clients, that is an additional 20,000 IOP of “average” demand placed on the storage. However, a major characteristic of VDI is that at start of day the majority of the clients boot there virtual system at close to the same time which creates a peak IO demand on the storage well above the average state. A great example of an enterprises requirement for high velocity data.

So the question to be determined is what Peak IO performance should the storage be able to deliver? The answer lies in the QoS expectations but for illustration let us assume a peak of 50,000 IOPS is the requirement. Based on a FC drive with an approximate 200 IOP capability this translates to an additional 250 spindles and if SATA is planned this number doubles. A lot of drives and this does not allow for the impact of the additional capacity demand.

This additional storage requirement driven by a need for increased access performance as well as increased data center capacity is what I refer to as virtualizations “Storage Tax”. It can be a rather unpleasant surprise to the unsuspecting.

So what are the potential solutions?

First step understand performance expectations and plan early for additional resources, chances are you will need them. Possible options;

1. Deploy a large number of traditional storage spindles. Concurrent access to these spindles will deliver high IOPS. Short stroking is a technique that could be applied. This is an expensive option in both capital as well as operational expense. Maximum tax if you will.
2. A solution that delivers high IOPS through dense spindle packaging enabling concurrent access to a large number of spindles. These architectures deliver an economical option without compromising performance.
3. SSD is an option and will most certainly meet the access performance requirements. However SSD is still expensive and depending on the capacity requirements may eliminate pure SSD as a practical option.
4. The most effective solution for the larger implementation is probably a tiered solution that combines SSD and spinning disk. While most storage vendors now offering tiered solutions in their portfolio it is not the storage (SSD or HDD) that is the key differentiator but the quality of their tiered storage management. Remember tiered solutions should eliminate problems and complexities not create them.