It's getting harder to distinguish between a low-end enterprise array and a high-end midrange array--but there are still some important differences.
High-end midrange arrays and low-end enterprise arrays look awfully similar these days. During the last few years, features like clustering, mirrored cache, replication and snapshots have trickled down from enterprise arrays, while low-priced SATA drives have moved up to enterprise arrays to further confuse the already somewhat arbitrary array classifications.
|How we define high-end midrange arrays|
Identifying which high-end midrange arrays to include in this article wasn't easy. Depending on how they're defined, there are from 40 to almost 100 different midrange array models. This article focused on models that users would most likely consider an upgrade to their current midrange model or as a replacement for an enterprise-class storage array.
We used the following four general characteristics as the criteria to define high-end midrange arrays:
The midrange array category is getting bigger, a catchall of low-end commodity systems and high-end arrays with specialized features (see "How we define high-end midrange arrays," at right). But there are distinct differences in array architectures in the midrange category that vendors don't advertise. These include:
- Low-end midrange arrays are typified by high-capacity SATA drives often with minimal or no storage management features in the storage array controllers such as replication or high availability. Low-end arrays may come with only one controller or two controllers configured as active-standby.
- High-end midrange arrays contain active-active or dual-active controllers, and generally support a mix of disk drive options such as Fibre Channel (FC), SAS, Fibre Attached Technology Adapted (FATA) or SATA. Some also offer advanced storage management features like asynchronous replication, snapshots, thin provisioning and automated tiered storage that ship as standard or optional features.
- Low-end enterprise arrays closely resemble high-end midrange arrays but have controllers in active-active configurations; may use proprietary ASICs for faster processing; and deliver higher levels of availability, reliability and performance for applications that have little tolerance for service disruptions.
The real differences
Vendor marketing rhetoric surrounding the numerous midrange array features obfuscate the real differences that exist among storage arrays and which features truly matter. Ease of use, five 9s, redundant hardware, scalability and tiered storage all show up as talking points on product data sheets; unfortunately, vendors rarely provide details as to how their array delivers on these features, how these features will significantly benefit the user or how the product is different from competing arrays.
Ambiguous terms like "ease of use" inspired Deanne Brueggeman, support services manager at St. Louis-based Arch Coal Inc., to list the different vendor's high-end midrange array features as starting points when evaluating them. Once the major categories were listed, Brueggeman created specific sub-points for each feature to ensure the storage array delivers specific functionality in the manner she expects. For instance, when verifying a vendor's claims about the availability of its midrange arrays, she looks at how the vendor defines terms like meantime between failures, as well as how storage array volumes are managed.
Five 9s is used increasingly by vendors to describe the availability of their arrays, but vendor definitions of five 9s vary. When applied to storage arrays, five 9s is almost always understood to mean fewer than six minutes of downtime per year--planned or unplanned--but that's not how all vendors see it. EMC Corp., for example, excludes planned downtime from the definition when applying it to its Clariion CX3 model 80 midrange array. Other vendors either refuse to define it or are vague in terms of how they use it or in what context this term applies to their midrange arrays.
|High-end midrange arrays|
|Click here for High-end midrange arrays: Key features and considerations (PDF).|
Users also need to quantify what a vendor means when it says its array is "available." Most high-end midrange storage arrays contain only two controllers; but during periods of firmware upgrades or in the event of a failure of one of the controllers, there will be a 50% drop in performance. This, of course, will be an issue with applications that require high performance 24x7x365.
Active-active describes a configuration where two controllers have the ability to provide high availability and uninterrupted access to their back-end disk drives. However, EMC (and other storage vendors selling midrange arrays) uses the term "dual active" in its data sheets to describe its Clariion CX3 UltraScale Series architecture: "The CX3 UltraScale architecture utilizes 'dual active' storage processors that are each able to access every drive in the array," according to the firm.
However, dual active isn't the same as active-active. A midrange array operating in dual-active mode lets both controllers accept I/O all of the time, although only one controller at a time can control and access specific LUNs assigned to that controller.
Most low-end enterprise arrays, such as Hitachi Data Systems' TagmaStore Network Storage Controller model NSC55, operate in an active-active configuration. This means that either controller can accept I/O from any of a server's host bus adapters (HBAs) and load balance the server's I/Os across the HBAs to the same back-end array LUN. With dual-active configurations, servers may only access the LUN through one controller. An active-active configuration also improves performance during firmware upgrades or when a failed part needs to be replaced.
Justin Cirelli, senior Unix administrator at Fiberlink Communications Corp., Blue Bell, PA, uses the high end of HDS' midrange class, the HDS TagmaStore Adaptable Modular Storage model AMS1000, which operates in dual-active mode. He finds this configuration adds administrative time and complexity to his environment because it forces him to manually balance the LUNs between the AMS1000's two controllers.
On the host side, Cirelli must ensure that the LUNs used by the server app are evenly balanced across the two controllers. It's common for an admin to assign odd-numbered LUNs to one controller and even-numbered LUNs to the second controller before presenting them to the server. However, this type of LUN assignment adds an additional administrative burden because Cirelli must keep track of how many odd- and even-numbered LUNs are assigned to a specific app. If an app is assigned too many odd- or even-numbered LUNs to store its data, one storage array controller may end up handling the majority of that app's I/O and performance will suffer.
To prevent this, Cirelli's first choice was the HDS NSC55--an enterprise-class array targeted at midsized companies--because of its active-active configuration. The NSC55's controllers offer equal, uninhibited access to all back-end LUNs, while the server's multipath software balances I/O across the server's HBAs regardless of which LUN is assigned to the app. But Cirelli ended up buying the AMS1000 instead of the NSC55 for financial, not technical, reasons: "A year ago the AMS1000 cost 35% to 40% less than the NSC55," he says.
|Alternative midrange array architectures|
Can't find a midrange array that quite meets your needs? Here are some new midrange storage arrays that may better match your specific requirements.
FUJITSU ETERNUS2000. The Eternus2000 allows administrators to enable or disable 128-bit encryption for each logical volume, which prevents another storage array from reading the data on the disk if the disk is ever removed. Its massive array of idle disks technology allows administrators to set policies on a specific disk or groups of disks so they're powered down when not in use. Fujitsu claims users can see up to a 20% reduction in power usage vs. other arrays.
LEFTHAND NETWORKS INC. SAN/iQ SOFTWARE. As companies consolidate Windows and Linux servers using VMware, they may end up with excess servers with no express purpose. LeftHand Networks' SAN/iQ software allows companies to repurpose those servers into clustered storage nodes accessible over Ethernet using iSCSI. There's no theoretical limit to the number of servers in a cluster; LeftHand Networks has successfully tested up to 40 nodes. You may also purchase Hewlett-Packard Co. and IBM Corp. servers preconfigured with the SAN/iQ software. The main downside of this configuration is if a server goes offline, all data behind that server becomes inaccessible; users should therefore mirror data across multiple servers to prevent unexpected outages.
OVERLAND STORAGE INC. ULTAMUS RAID 1200. This unit is designed for users who need high availability and high capacity at a lower price. Suggested retail pricing for the Ultamus RAID 1200 starts at approximately $1,000/TB for SATA drives and $5,000/TB for SAS drives; the unit scales to 45TB using SATA drives.
NEC CORP. D-SERIES NEC's D-Series storage arrays allow users to scale the same controller unit from a 2U, four Fibre Channel (FC) Port D1-10 with 96 SAS or SATA drives to a D8-1040 with 64 FC Ports with 1,152 SAS or SATA drives. Most of the upgrades occur offline except with the highest-end models.
Same name, different price
In addition to controller architectures, replication software, volume snapshots and tiered storage management can vary significantly among arrays, including how vendors charge for these important functions. About a year ago, Arch Coal's Brueggeman replaced a small Hewlett-Packard (HP) Co. FC SAN with an EqualLogic Inc. iSCSI PS200E array SAN because "Arch Coal was being priced to death for each software component," she says. EqualLogic licensed its array management and replication software as part of the price of the array. (The PS200E has since been discontinued, but EqualLogic still supports the model.)
Fiberlink Communications' Cirelli, who's satisfied with the overall performance and support he receives on his HDS AMS1000, indicates that if asynchronous replication had been available at a lower cost or free from another vendor, his purchase decision might have been different. He has two AMS1000s in different sites and considered using HDS' TrueCopy Remote Replication software to replicate data between them. However, it would cost more than $300,000 to license the software, and he would need to dedicate an FC port on each array for replication and purchase an FC-to-IP router for both sites. Because of this, Cirelli decided to continue his use of host-based replication software, which is considerably less expensive than HDS' array-based option.
Service and support
Price, performance and replication software are the major factors users consider when choosing among different midrange arrays; all things being about equal, a vendor's service/support offerings and price usually tip the scale on a buy or no-buy decision.
Cirelli realizes he paid a premium for HDS' storage arrays, but wants field engineers who understand his shop's configuration. When he made his decision to stay with HDS, Cirelli looked at a competing storage array, but didn't feel confident that the vendor understood his situation or what he was trying to set up. "If I ever need to recover from a huge failure, it is likely going to result from a human error," says Cirelli. "I need to have confidence that the support engineer understands my environment and can help me rebuild it."
Louis Skelton, VP of technical services at Scientific Games Racing in New York City, chose HP's StorageWorks EVA5000 and EVA8000 midrange arrays mainly because of HP's service organization. Scientific Games Racing processes approximately $11 billion annually in pari-mutuel wagering for events such as horse and dog racing and jai alai in North America, with additional revenue generated by overseas events like Korean bullfights. Because races or fights occur only once and at any time, a service interruption or failure is catastrophic.
Scientific Games Racing's operations need to operate 24x7x365, and Skelton must ensure their system is architected to handle peak loads at different hours. Although admittedly an HP shop, this combination of difficult-to-predict variables prompted Skelton to place a higher priority on infrastructure stability rather than on midrange array feature functionality. "HP's ability to design an end-to-end solution for our environment was more important than the price or even the feature set on other midrange arrays," he says.
Yet there are cases where specific midrange array features close the sale. Todd Rayl, VP, managed security services at Business Vitals in Columbia, SC, was initially skeptical about the quality of service (QoS) feature of Pillar Data Systems' Axiom 500 midrange array, which places data requiring higher performance toward the outer bands of a disk drive and stores data needing lower performance on the inner bands. He was also concerned about Pillar's relative newcomer status as a storage array provider, but was more concerned about how EMC and Network Appliance (NetApp) Inc. priced their software licensing upgrades. For Rayl, these were unnecessary costs he couldn't justify paying. Exponential growth rate in the services business could quickly take him to the capacity limits of a NetApp or EMC product line, and then force him to upgrade the SAN hardware and purchase all new licensing. "Pillar's simplified software licensing model was a key factor in our decision-making process," says Rayl.
Because Pillar required only one license for its Axiom 500 midrange array, there was no additional expense for software features that EMC and NetApp categorized as optional. And after using the QoS feature, Rayl found the Axiom 500 outperformed his existing EMC Symmetrix 8430.
The good news is that there are a growing number of good midrange storage arrays to choose from and it's a buyers' market. Some buyers are finding it hard to ignore the savings that midrange arrays from new companies offer. In many cases, these new-company arrays perform as well as better-known products, and offer features and functions that cost substantially less.