Return on Investment

Because they rely on proprietary chassis equipment and management console designs, blade products are still more costly than conventional servers in initial outlay.

But in terms of total cost of ownership, vendors and analysts claim blades win over conventional systems once a critical number of blades have been installed.

Blades offer several operational and price/performance advantages, halving costs.

Real estate: Dense blade arrays occupy less space, reducing rental and construction costs.

Power: Blade systems, particularly those based on low-power, laptop-class chips, can consume less than 50% of that used by conventional systems.

Management: Increasingly sophisticated consoles allow administrators to ‘ignite’ and ‘extinguish’ individual blades in line with changing processing loads. Faulty blades can be ‘ripped and replaced’ in minutes, with minimal disruption.

Problems

Too hot to handle

Vendors agree that blades will become smaller and more powerful, and blade systems will become increasingly dense – but only if the industry can solve its growing problem with heat dissipation.

Today, a fully loaded 42 unit blade chassis produces roughly 12 Kilowatts of heat, about 50% more than a typical household oven. Actual deployments of such fully loaded blade chassis are still rare, but server hosting services companies have already noticed their exceptional thermal impact on nearby systems and are now charging premium rates for hosting blades in response.

Experts estimate that air-cooling systems must move 140 cubic feet of air for every 1 Kilowatt generated, to keep temperatures below the 75º Fahrenheit point, at which system performance dips and component life-span is reduced.

The risks are clear: data centres may need to use the extra free space created by installing blades to accommodate airflow instead.

Suppliers

The Blade Runners

RLX Technologies launched the first commercial blade servers in 2001. RLX established the blade architecture, and made innovative use of Transmeta’s Crusoe chip, exploiting its low-power characteristics to limit the heat produced by densely packed processors. IDC estimates RLX’s market share at below 5% of the overall market, but the company continues to innovate.

The world’s largest systems companies have now come to dominate the blade server market with the balance of power largely reflecting the existing overall server market status quo. IDC credits IBM with 47.1% of the Western European market for Intel-based blades; Hewlett-Packard has 37.2%, Fujitsu Siemens 8% and Dell 5.2%.

Dell’s low position – relative to its large share of the overall server market – reflects a half-hearted approach to what has proved so far to be a low-volume market. However, the company is expected to refresh its PowerEdge range in Q4 2004, when it says it will launch a one-unit blade system with 50% greater density than its rivals.

Sun is also a factor, but does not appear in IDC’s ranking, as it has only recently added AMD Athlon and Intel Xeon processors to its blade range. Nevertheless, Sun and its technology partner, Fujitsu Siemens, are the first blade vendors to support Intel and Sparc blades in the same chassis. This allows compute-intensive elements of large applications to be deployed against 64-bit Sparc blades, while at the same time, front-end elements, such as web serving, are configured against dense Intel arrays. With its FlexFrame management system, Fujitsu Siemens has tuned this capability to the particular needs of SAP’s MySAP – an application-specific trend that is likely to be mirrored by other vendors.

Hewlett-Packard, IBM, Sun and Fujitsu all see blades as playing an important role in their larger grid and utility computing strategies (see below). So far, IBM is the only vendor to ad-dress the supercomputer space with a blade product. Its eServer Cluster 1350 topped the independent benchmarking ratings when it was released last year; IBM claims it significantly lowers the price and management costs for truly high-performance systems.

Trends

Blades buying patterns

The collapse in server sales following the dot-com bust stifled early blade server sales. But as businesses begin to reinvest in IT, and start to replace servers acquired in the run up to Y2K, blade sales are expected help lead overall server market recovery. In 2003, blades accounted for 27% of total server shipments, a share IDC says it expects to increase this year, and to reach 40% by 2008.

As the market heats up, the type and scale of blade systems are also expected to change. Most early sales featured relatively small arrays of simple single-processor blades, often chosen by telcos and ISPs. During 2004, as vendors bring more dual- and quad-processor blades to market, investment banks, biotech and oil companies are expected to drive sales of compute-intensive systems, and of high-end, blade- based Linux clusters.

As market volumes pick up, the rate of product innovation should also accelerate. Dell has flagged its intention to attack the ultra-dense one unit blade arena, and aggressively reduce the price/performance benchmark in this volume market. Meanwhile, Sun, IBM and Hewlett-Packard are likely to focus on management software, tightening the fit between proprietary blade chassis and their grid and utility systems strategies.

Architecture

On demand computing

Blade servers look set to play a central role in the creation of highly efficient virtualised data centres, where the total contents are treated as a single, virtual resource.

Data centre virtualisation will mean that applications are no longer confined to a predetermined set of servers and other physical resources. Instead, their share of the virtual resource can be allocated to them dynamically, according to their own fluctuating requirements, reflecting business priorities.

The modular nature of blades means they are ideally suited for virtualisation. Blade management products, such as IBM’s Administrator and RLX’s Control Tower, are designed to make it easier for administrators to remotely provision, tune and reconfigure blades, and to give them access to other resources, such as network equipment and storage.

At its utility computing research centre in Bristol, for instance, HP uses a template-based management console that allows virtual resources to be provisioned and redeployed rapidly. Ultimately, applications are expected to take ‘responsibility’ for dynamically discovering and provisioning their own physical resources.