Part II: Blade servers and storage implications

Posted on February 01, 2004

RssImageAltText

The second article in our series examines management issues, the role of blade servers in grid computing, and the impact of blades on storage.

By Alex Gorbansky

Comprehensive management and provisioning software is critical to unlocking the scalability and versatility features of blade servers. Seamless scalability refers to the ability of administrators to quickly add more processing power to an application as its needs grow. Versatility takes that one step further by allowing users to dynamically allocate and re-allocate processing power across different applications characterized by dynamic workloads. The ultimate goal is a complete abstraction of blade servers (and standard servers) into a virtual pool of computing resources that can be provisioned across enterprise applications.

An analogy can be drawn between blade servers and the evolution of the Fibre Channel storage area network (SAN) market. Fibre Channel hardware emerged in the late 1990s promising to rescue users from the limitations of direct-attached storage (DAS) through superior performance, scalability, manageability, and improved overall cost of ownership. Because SAN hardware matured much faster than management software, most early adopters never fully realized the complete benefits of SANs. Even today, many users would argue that they are not fully utilizing the capabilities of networked storage due to immature software. Users considering the acquisition of blade servers should remember these lessons.

Blade management software will be one of the most important factors in driving overall blade-server adoption and determining which hardware vendors will succeed in this emerging market. Expect systems vendors to invest heavily in software capabilities and to use these capabilities as a major differentiator.

Blade management solutions include some or all of the following capabilities:

Server provisioning—Server provisioning remains one of the major challenges plaguing systems administrators today. The hot-pluggability of blades resolves some of the physical deployment challenges. Blade management software will automate many remaining manual tasks such as blade assignment to a VLAN and the loading of application and operating system images. In advanced blade servers, images are automatically loaded onto a blade based on pre-defined policies in a process called "blade profiling."

These server-provisioning capabilities enable users to seamlessly make configuration changes across many blades and help bring just-in-time computing closer to reality. Administrators can readily re-purpose blade resources from one workload or application to another by loading the appropriate software stack onto blades.

Fail-over configuration—Through blade management software, administrators can easily designate idle blades to serve as fail-over blades for one or many blades performing a given function. This functionality enables users to assign different numbers of specific blades to various applications based on a given application's processing, performance, and availability requirements. To avoid performance degradation in the event of a blade failure, vendors recommend that users leave a pool of spare blades that can take over production tasks in place of a failed blade.

Remote console—Today's systems administrators require the ability to remotely access servers anytime and from anywhere. For this reason, blade management software gives users the ability to access server blades, the blade chassis, and chassis internals remotely and through a single console. In most cases, blades will remain remotely accessible through the console even when the operating system is not operational. This is sometimes referred to as "lights-out management."

Monitoring and diagnostics—As with other systems management solutions, blade management applications allow users to aggregate key systems metrics and alerts collected through SNMP traps or integrated systems management processors. Metrics and alerts are correlated to both physical and logical resources—an individual process failure will correspond to both a unique server blade and to the application or workload to which the blade and failed processor were assigned. IT organizations that have deployed enterprise-wide systems management software should look for blade management software that can plug into these frameworks.

Secure role-based administration—A bulletproof access control model is the cornerstone of any systems management software. Blade management software is no different and allows the creation of roles and the association of administrative privileges to specific roles for a given blade or logical grouping of blades. Thus, a particular user may have read-only privileges for one set of blades, but at the same time have read/write privileges for another application.

The overall importance of blade management software is underscored by the acquisition and partnership activity in the server-provisioning space, a technology sector that will be key for blade servers. Examples include Veritas acquiring Jareva, Sun-Terraspring, Sun-CenterRun, IBM-ThinkDynamics, and HP partnering with Opsware.

Aside from one-stop shopping, users now have other reasons to celebrate. The spree of acquisitions and partnerships should accelerate the pace at which blade management and provisioning capabilities are fully integrated with existing systems management frameworks. The level of integration with existing management software will be a key criterion in users' evaluation of blade management capabilities. After all, the last thing that users need today is yet another stand-alone management application.

The spate of acquisitions—particularly Veritas-Jareva—also demonstrates that the boundaries between servers and storage are becoming increasingly blurred. Veritas' entry into the server-provisioning market suggests that server and storage management will become more tightly integrated over time, particularly as server-to-storage I/O is virtualized.

Grid computing and blades

Blade servers represent a key underlying technology within "grid-computing" architectures. While the concept of grids is not new—the virtualization of servers, networking, storage, and data resources for dynamic on-demand provisioning across applications—grid deployments have been primarily limited to the research and scientific communities. The adoption of grid architectures in commercial enterprises could dramatically alter the current IT landscape.

The unique physical architecture of blade servers makes them ideal for grid computing. While vendors have articulated their visions for this next evolution in IT, few products have fulfilled the promise. While enterprise deployments of grid technologies have not begun in any large scale, as grid-enabled applications begin to gain traction in data centers, blade-server adoption across not only edge, but more importantly in middleware and database environments, will grow in parallel. The evolution of grid computing will be a critical determinant of the pace and breadth of blade-server adoption.

Impact on storage

As blades gain adoption in enterprise data centers, they will have an impact on existing storage infrastructure and storage management. The extent of this impact will depend largely on the type of application running on the blade. As users evaluate multiple blade options, they should consider how blade servers will impact their existing data-center hardware infrastructure and management frameworks.

Direct-attached storage—Despite some of its architectural limitations, DAS offers an attractive and cost-effective solution for system boot and application storage, as it is dedicated to each server and delivers adequate performance. Internal SCSI, ATA, or Serial ATA drives are ideal for edge server blades where form factor and low cost are key requirements. Today, most blades ship with internal storage for system boot and application storage. Users should validate RAID support for the internal drives. As blades mature, expect more vendors to support booting from external storage.

Network-attached storage—For external, shared file storage, applications usually rely on NAS. Most blade servers have the option of installing an Ethernet switch into the chassis for network connectivity. The blades would then access a NAS device via the LAN similar to a standard rack server. NAS connectivity is appropriate for some Tier 1 blades as well as for Tier 2 and Tier 3 SMP blades.

Fibre Channel SANs—Fibre Channel SAN technology meets high-performance, block-level storage requirements for Tier 2 and Tier 3 applications. Several blade vendors have already integrated a Fibre Channel switch in the blade chassis. Some OEMs are working with Fibre Channel vendors to embed Fibre Channel controllers on the server blade. The introduction of Fibre Channel on individual blades and within the chassis will increase the cost of a blade solution and potentially lower the overall CPU density of the chassis. Fibre Channel switch and host bus adapter (HBA) vendors competing for a piece of the blade-server market must deliver low-cost, small-footprint embedded solutions without sacrificing performance.

iSCSI—iSCSI may represent a more appropriate protocol for block-level connectivity of blades to external storage. iSCSI would not necessitate the addition of new hardware to the blade and would require only the installation of specialized NICs and drivers. Moreover, iSCSI would not introduce the additional levels of management complexity resulting from the deployment of a Fibre Channel switch within a blade chassis. As iSCSI matures and gains more-widespread adoption, we expect it to be one of the primary methods of connectivity of storage to server blades.

Storage management—Embedding Fibre Channel components into proprietary blade architectures may result in additional SAN management challenges. Existing SAN management software may not be able to correctly discover, visualize, and manage embedded HBAs and switches within the blade chassis. As a result, users may be forced to use a separate management tool to configure and control the Fibre Channel components. Seamless integration with existing management schemes will be a key criterion for users evaluating blades with built-in Fibre Channel capabilities.

In addition, as blades drive the virtualization of physical server resources, and as comprehensive server-provisioning solutions gain adoption, storage-provisioning tools may evolve into modules within server- and application-provisioning frameworks.

Blade servers and InfiniBand

The InfiniBand market has seen a revival as major OEMs promote their respective visions of utility computing and bring to market their next-generation blade offerings. The high-performance, low-latency, and switched characteristics of the InfiniBand protocol make it ideal for blade configurations of hundreds of clustered server nodes and applications with stringent performance requirements. InfiniBand also provides a mechanism for the consolidation and virtualization of all server and application I/O. It is not surprising, then, that several OEMs have already announced plans to integrate InfiniBand into their blade solutions, and most are considering InfiniBand for their next-generation blades.

OEMs are integrating InfiniBand into blades in several different ways. Some OEMs are implementing an InfiniBand backplane into their blades and then providing IB-to-FC and IB-to-Ethernet gateways for I/O aggregation.

Another model involves the development of embedded InfiniBand cards on the blade with connectivity to an InfiniBand switch for server clustering and I/O virtualization. The drawbacks of integrating InfiniBand into blade servers are very similar to those of Fibre Channel: InfiniBand increases the overall price tag while introducing additional management complexities. In developing and positioning InfiniBand-based blade solutions, vendors will need to focus on manageability and cost.

Standards

Much noise has been made about the importance of standards in the evolution of blade technology. Existing blade servers are typically characterized by proprietary hardware and interconnect architectures, making it nearly impossible for users to intermix blades from different vendors. Dissimilar hardware designs are sure to make users reluctant to invest in a technology platform with limited flexibility and a potentially limited lifespan.

A derivative of blade vendors' proprietary hardware design is a proprietary management interface. It is the lack of common management hooks that represents the more important challenge that vendors must overcome. While users may gain incremental value from the ability to intermix blades from multiple vendors, these gains would be minimal, especially when compared to the efficiencies derived from a blade management solution capable of supporting a multi-vendor environment.

Although some early adopters may not consider common management standards key evaluation criteria, most large IT shops that source from multiple vendors will need a common management infrastructure and may thus hold off on purchases until standards have evolved.

For this reason, common management standards will be an important driver in the adoption of blade technology in the data center. Intel's Intelligent Platform Management Interface (IPMI), with support from Intel, Dell, Hewlett-Packard, and NEC, defines a common interface for monitoring and managing blade servers. We expect IPMI to serve as a foundation for the development of comprehensive blade management standards.

Conclusions

Blade servers' unique space- and power-efficient design, coupled with flexible management and provisioning capabilities, makes them a natural fit for dynamic, horizontally scalable applications. Blade adoption in middleware and database environments will accelerate over the next few years.

Several key market drivers will influence the pace at which blades permeate the enterprise data center. To succeed in the data center, blades will have to demonstrate the ability to handle high-performance workloads. The importance of automated management and provisioning capabilities cannot be overemphasized. Users will gain only marginal benefits from blade technology without a comprehensive management framework to exploit the hardware design.

The transition of grid, or utility, computing from vision to real solutions will be a critical factor in driving blade adoption. Until major application vendors re-architect their applications to leverage modular, commodity processing units, deploying blade technology beyond the edge will be challenging. A new generation of virtualization software could change the entire landscape and make blades ubiquitous in enterprise data centers.

Note: For vendor/product profiles, view the entire report (Blade Servers: A Revolution in Server Architectures and Its Impact on the Data Center) at www.tanejagroup.com.

Alex Gorbansky is senior analyst at The Taneja Group consulting firm (www.tanejagroup.com) in Hopkinton, MA.


Blade servers: Advice for end users

√ Assess your current and future application needs and how blades address these needs. We recommend that most users initially evaluate blade solutions for edge applications to minimize up-front risk and investment.
√ Take into account how blades will impact your existing hardware infrastructure, including networking, servers, and storage. What additional complexity and costs will result from the introduction of blades?
√ Evaluate the whole solution—hardware and management software. Be sure to not only understand the "speeds and feeds" but to consider how your new infrastructure will be managed. Find out whether vendors' management capabilities support—or will support—blades from other vendors. Does the vendor's software integrate into your existing management frameworks?
√ Understand the vendor's product road map for next-generation designs. Will new blades be backward-compatible? If they are not yet integrated, understand plans for adding Fibre Channel, iSCSI, and/or InfiniBand.
√ Evaluate system availability and fault-tolerance parameters. Are shared components such as power supplies redundant? Is RAID built-in?
√ Develop explicit ROI metrics to understand at what inflection points blades deliver better investment returns compared to equivalent configurations of rack servers, and assess the savings from reduced space and power consumption, as well as management efficiencies.


Representative vendors

Blade servers
Dell
Engenera
Fujitsu
Hewlett-Packard
IBM
Intel
NEC
RLX Technologies
Sun

Blade mgt. software
BladeLogic
Ejasent
Moonlight
Opsware
Veritas

InfiniBand
InfiniCon Systems
Topspin
Voltaire

Storage
Emulex
QLogic
StorageTek


Comment and Contribute
(Maximum characters: 1200). You have
characters left.

InfoStor Article Categories:

SAN - Storage Area Network   Disk Arrays
NAS - Network Attached Storage   Storage Blogs
Storage Management   Archived Issues
Backup and Recovery   Data Storage Archives