Kurt wong, director of storage services and Chuck Cancilla, storage engineer, have the challenging task of managing over 250TB of storage at Inovant, the wholly owned service arm of Visa International in San Francisco, CA. One way they're meeting that challenge is by organizing their storage into service tiers.
The goal of the service tiers is to move toward a shared network environment balancing business needs, performance and cost while not being tied to a specific technology. "We are not building around a technology, but around tiers that satisfy business services," says Wong.
Inovant's storage tiers consist of EMC Symmetrix on the high end, and midrange IBM Shark and Network Appliance F800 filers. Using a planned combination of high-end and midrange storage is gaining popularity among storage managers. Successful implementation depends on understanding what you're getting with each tier, and what your overall challenges are.
Compare and contrast
To explain high-end vs. midtier, architecture and service levels must correlate to centralized and distributed computing models. High end: Centralized computing is usually synonymous with high-end storage. This is storage located within the data center and scales from tens to hundreds of terabytes to service more than one application and host platform. Not only do high-end systems protect data from loss due to hardware or software errors, but data is kept available through mirroring of cache and disk. Maintenance must be performed while the storage subsystem is running a normal production workload. While there are high-end distributed systems deployed in production, it's not a widespread practice due to cost.
Some examples of high-end storage systems are EMC DMX or Hitachi Data Systems (HDS) Lightning V series, along with a few new technologies such as utility-based storage from 3PAR and network unified storage from LeftHand Networks.
|What's in a name?|
Another set of terms being used today to describe midtier and high-end storage is monolithic and modular. The implication is that modular storage is part of a building block that can grow as you expand, and monolithic is a large storage repository that's filled up as needed. The confusion comes when modular is assumed to be midtier and monolithic is high end. However, there are cases where a high-end modular system such as EMC's DMX800 is a great fit for an evolving infrastructure. Whereas an IBM ESS--mostly viewed as monolithic--is really a midtier solution because of its less-than-robust availability features.
Chuck Hollis, VP of markets and products for EMC, says the most important distinction between the two classes of storage should be based on how customers are using storage and that will determine their need for a monolithic or modular solution. "To me [modular vs. monolithic], is about buying trends and packaging," he says.
Midtier: Aimed at a distributed enterprise and designed with features such as RAID protection and replication, it's meant to service discrete applications and accept some data unavailability. Because midtier storage is usually placed in remote locations, management is important. While performance is a consideration with a midtier system, and many of the newest offerings offer incredible throughput (the HDS 9500 V yields 4GB/s internal throughput), it's not the primary concern. Price, flexibility and manageability rule the day when it comes to the midtier.
Midrange users seem to know more about what they don't want than what they do. According to Phil Townsend, HDS senior director of worldwide marketing, customers "don't want islands of information" or to "compromise on storage management." Plus, the market is crowded with HP, IBM, and Network Appliance, as well as HDS, EMC and many other companies. Some examples of midtier storage subsystems include the HDS 9500 V series and the EMC CX600 (also sold by Dell).
Evolution of midtier
Midtier storage is now a significant part of nearly every enterprise storage strategy. Most midrange storage systems are a dual-controller system with availability options to help minimize loss of data access and adequate performance. HDS, for example, has enhanced almost every area of its midtier platform (see "The evolving HDS midtier disk systems," this page).
Most midtier storage systems have the following characteristics:
Total Storage: Disks continue to spin faster with denser capacities, yielding better performance. Today 146GB drives are available in the latest midtier systems such as the HDS 9500 V.
Internal Bandwidth: Faster cache, bus and interfaces have reshaped the architecture of the midtier. HDS doubled the 9200's internal throughput as it evolved into the 9500V.
Maximum Disks: With denser packaging, the maximum number of disks has increased. This--combined with larger disk capacity--has pushed midtier storage to capacities rivaling the largest high-end systems of only a few years back.
However, this doesn't mean the 9500 V will replace 9900 class systems in the enterprise. The midtier strategy for HDS is clear: It has a midtier product for discrete applications, while their high-end solution is designed for consolidating applications.
The evolution of midrange storage can make it confusing when deciding whether you need midrange or high-end systems, particularly if you just focus on specs. For example, several HDS 9500 V midrange systems with 4GB/s throughput could be placed in the data center to handle the I/O workload of 16 discrete applications, or a single high-end EMC DMX800 could handle the same workload. There's little difference in performance throughput, but the high-end system has at least two advantages--and you pay for this.
A quick look at the relative cost for 1TB of storage illustrates how much is being paid for the high-end advanced features and availability. For 1TB of midtier storage, the average cost across the vendors discussed here is approximately $50,000, while the entry point for high-end storage is a little less than $200,000. For this additional money, you gain added availability and functionality.
For example, at the midtier, you can still perform replication within the subsystem. But at the midtier level, you'll experience degraded performance and potential unavailability of applications. High-end storage subsystems with shared applications don't have these problems.
One disadvantage is resources consumed by each of the 16 applications ebb and flow as applications change and evolve. When one of these applications isn't running a cache-friendly access pattern, other applications use the latent resources.
Another difference is performance has more to do with sustained performance. For example, when ShadowImage software is running on a Sun high-end StorEdge 9980V, there's no noticeable decrease in performance on the subsystem. As a general rule of thumb, a high-end storage subsystem has from six to eight times the performance of a midtier solution on a terabyte-for-terabyte basis.
Another common misconception is that you can buy midtier storage and use host-based software products to deliver the functionality in a high-end array. Application program interfaces (APIs) for the midtier storage platforms aren't as feature-rich as those for high-end systems, thus storage managers aren't able to manage a lower-end solution as well as the high with the storage resource management platforms available today. There's been progress made in the common information model (CIM) standard, but more time is needed.
While there may be financial benefits to an n-tier environment, management and ease of deployment outweigh this consideration for most users.
Even at Inovant, manageability is a significant challenge in the service tier environment. "Our ideal solution would be some type of enterprise-level switched network-attached storage," says Cancilla, which he termed SNAS. This allows Inovant the ease of NAS management and the performance of a switched Fibre topology.
Both high-end and midrange storage are moving targets and continue to evolve. Some of what is expected on future products will change the debate about tiered storage.
Vendors and end users agree that the following is high up on their must-do and wish lists:
- Manageability: increasing the number of gigabytes managed per person.
- Visualization: the ability to have a single view of the SAN and associated components
- File sharing: virtualization at the file level and user view
- Service: providing storage as a service
Manageability continues to expand in breadth and depth through various software offerings and some management features are being built directly into the storage subsystems. The same management functions of high-end systems can be duplicated at the midtier, but at the cost of performance and availability. When determining the performance requirements of a storage system, take into consideration the subsystem operating at peak workload with advanced functions all running.
SAN visualization from a common set of tools is also emerging from both hardware and software vendors. In addition to visualizing the storage network from a hardware view, it's also important to increase usability of the end-user storage experience. This is becoming more important, since the amount of information a user is accessing daily is increasing. Hiding the complexity of the underlying information infrastructure will continue to simplify storage operations. One implementation of the virtualization concept is the development of a common file system that allows easier file duplication and sharing across mixed high-end and midtier environments independent of the attached host systems.
Storage as a service is a business trend that continues to formalize as storage becomes more of a commodity and business units push back on complex storage acquisition models. Both the midtier and high-end continue to evolve on fast and distinct paths in the storage community. Look for these tiers to coexist for quite some time with added manageability, visualization, virtualization and service utilization models.
Dig deeper on Tiered storage