What is "Enterprise RAID?"
By Thomas Lahive
The RAID market can be divided into three categories: internal commodity RAID, where the controller is typically PCI-based; external commodity RAID subsystems, which scale up to 15 drives and support only one server; and enterprise RAID, which offers virtually unlimited scalability and heterogeneous server support.
The commodity segment--both internal and external--averaged $0.30 per MB during 1997, whereas the enterprise class segment was at $0.80 per MB. All storage vendors want to be considered enterprise RAID suppliers in order to receive healthy margins. However, the barriers to enter the enterprise class segment are intense.
What features are required to be classified in the enterprise RAID segment? Dataquest believes that there are four factors: level of fault tolerance, server support capabilities, extensive software functionality, and service capabilities.
The RAID Advisory Board (RAB) has defined three major Extended Data Availability and Protection (EDAP) classifications (although seven derivatives exist with 21 EDAP criteria. For details, see InfoStor, January, page 20: "The RAB Guide to Nonstop Data Access," or visit www.raid-advisory.com). The RAB EDAP definitions are a good place to start in determining the enterprise-level fault-tolerant capabilities of RAID systems.
Most RAID subsystem suppliers have a process to consolidate servers, although many of these configurations have different levels of reliability and data partitioning capabilities. For example, it is possible to support ten servers within one cabinet by daisy-chaining together multiple controllers and host adapters. Another alternative is to externalize host server support by connecting the subsystem to a switch, hub, or storage "director." Similar to using a switch, mainframe-class disk arrays sometimes have a PCI backbone that can support many host adapters in order to easily connect multiple servers. The PCI approach requires a higher-performance backbone so that the I/O rate will not degrade as more host adapters are added. As a result, the PCI approach is more expensive than daisy-chaining.
Another dimension in defining enterprise storage subsystems is the level of software management. The table differentiates the various types of management capabilities.
Every storage supplier offers some level of service capabilities, but enterprise-class systems provide additional levels of serviceability. The requisite base (e.g., non-enterprise) service capabilities include:
- Hot-swappable components
- Customer-replaceable components
- RS-232 port or network connection for remote management
- Software that monitors erratic components and determines potential failure
- Services to notify IT managers of failed components (can be provided by storage vendor or third party)
Enterprise-level services include all of the capabilities listed above, plus:
- 2 4x7 remote monitoring
- Online microcode upgrades
- Service engineers that can tune subsystems to optimize performance
- Remote monitoring service that can also change hardware configurations remotely in order to keep systems online
To be considered truly enterprise class, storage subsystems must offer the highest reliability level, a comprehensive architecture that enables easy and virtually unlimited server support, multiple levels of software classifications, and robust service capabilities.
It is impossible for one RAID supplier to develop in-house all of the features to meet this classification. Historically, disk drives were the only components with multiple sources. Now, however, not only are subsystem suppliers able to multi-source RAID controllers, but they can also provide directors, fail-over software, backup utilities, etc. from many different sources. The enterprise storage provider of tomorrow will devote its research and development efforts to integrating all of these components, as opposed to just developing microcode.
At this point, the hardware cost differential between enterprise and commodity storage subsystems is about 80%, but Dataquest expects that to quickly decrease to about a 30% cost differential. As a result, suppliers will have to devote more attention to software development and service offerings in order to maintain margins. RAID continues to be the most costly component in the server environment, and users are beginning to realize the benefits of appropriating more resources to the management and protection of the company`s most valuable asset--its data.
Tom Lahive is a senior industry analyst, specializing in server storage and RAID, at Dataquest, a market research firm, in San Jose, CA. Lahive works in the company`s Westborough, MA, offices. He is a regular contributor to InfoStor.