Home > Focus
 Print Friendly Page ||  Email this story

Focus: Blade Servers
The next cut

Second-generation blade servers provide the right mix of high rack density and computing power. Here's a look at the guises of blades to come. by Ong Boon Kiat

If first-generation blade servers herald density above performance, second-generation ones—currently shipped by major vendors—signal a more mainstream approach. While high rack densities are still important, vendors are clearly less willing to forgo power to get there.

Consider Compaq's first generation blade, the Proliant BL e-Class blade servers, which squeezed 20 servers—each fitted with a 700 MHz Ultra-low voltage Intel Pentium III chip—into a 3U rack. In stark contrast, Hewlett-Packard's latest Proliant BL p-Class (essentially a Compaq-inherited product) now packs only eight servers in 6U, but comes with regular 1.4 GHz Pentium III processors and dual-processor capabilities.

Following HP in using mainstream (as opposed to low-voltage) processors is IBM. It recently launched the Blade Center, a dual-Xeon capable blade server. Compared to the HP p-Class blade, its obvious direct rival, the Blade Center packs a higher server count: Up to 14 servers in a 7U rack. Dell's upcoming and first blade product, the PowerEdge 1655MC (targeted for launch next January), will house six blades in 3U, and packs dual-1.3 GHz Intel Pentium III processors.

Blade servers like the Dell 1655MC, because they use full-fledged processors and chipsets which need space to breathe, are less dense. And when vendors suit up their blades with heat-spewing quad-processor architecture next year, expect even fatter blade servers to emerge. So for MIS whose appetite is whetted by the promise of reclaimed server room space, expectations may have to be pegged back by next year. So what else can MIS expect to see from this relatively infant technology next year?

Recapping blades

A blade is simply a chassis-less server—make that many chassis-less servers bandied in a specially-designed rack connected via a custom-made backplane.

The blade itself is the motherboard, which plugs into the backplane. There are many standards for connecting blade servers, like compact PCI (cPCI), PCI-X or Fast Ethernet. Most are customized, like HP's p-Class blades, which use 2 separate backplanes for mating.

Blades are largely identical, but can be configured with more disk or processors to serve specific functions, like a NAS or firewall.

Then there are the common resource components for the blades, including cabling, power supplies and cooling fans, which are shared by all blade servers in the enclosure.

Stripping down and pooling hardware this way saves money, although startup costs, which includes the purchase of the rack, is higher than rack-mount servers up to a "break-even" point, usually five servers.

But the main enticement of blades is in its ease of setup and aggregation compared to rack-mount servers.

In particular, a powerful blade feature is its ability to provision a fresh blade without an administrator physically touching the server. The result: True computing-on-demand, where blades can be quickly transformed into Web, DNS or file servers as required.

Niche no more?
After flirting with low-voltage chips, blade vendors are now retooling for conventional server processors. They are also hoping to extend blade servers beyond its initial edge-computing paradigm.

"Blades grew out of the need for densities in data centers and this is not going away. But we now like to take a more holistic approach to building blade servers," said Tony Parkinson, Manager-Product Marketing, Industry Standard Servers, Enterprise Systems Group, Asia-Pacific, Hewlett-Packard. This means building them for more conventional server applications. He describes the p-Class servers as suitable for ASP hosting, terminal server applications and media streaming—all heavy-duty chores.

Shashi b Mal, Brand Executive, eServer xSeries and Intellistation Server Group, IBM ASEAN/SA believes that taking on the role as real enterprise servers will be critical for the viability of blade servers. He believes that blades will find a wider niche because many enterprise applications, such as Microsoft Exchange and Web servers like Apache have "scale-out" requirements, which suits blade servers to a tee.

Some application, though, will be ill-suited for blades. For these, conventional servers or mainframes will be better options, said Mal. One common example is Microsoft's SQL Server, which can use up to 32 processors. Applications like this have "scale-up" requirements and are built to use multiple processor hardware and operating systems.

More power
Applications will also dictate whether blades can take on a heavier load. This will likely take a while since blade-optimized applications, especially those of the distributive processing and deep server clustering genre are not here yet. The signal, according to some market watchers, will be when a huge enterprise presence like Oracle becomes optimized on the multiple nodes of blade servers rather than just one huge node of the mainframe (or a large Unix server).

In the near term, most vendors are now looking at quad-processor blades next year, but not much beyond that. As for Itanium blades, most aren't even thinking about it yet, partly because of the lack of applications. "There is no value proposition for Itanium blades because there is relatively little demand for IA-64 at the moment," said Damian Crotty, Director, Enterprise Servers, Server CoC, APAC, Dell Computer Asia.

Dell, however, is prepping for a "super" blade architecture that is slated for launch next year, coined "brick". This will be a modular server architecture that shares a blade-like backplane, but instead of identical servers will feature specific modules customized for processing, I/O, disk, etc.

"Bricks will be application powerhouses," said Crotty, adding that they will be targeted at mission-critical applications like Oracle and SAP, and vertical industries like the financial sector.

Sharpening storage blades
With storage being acknowledged as a business do-or-die, it is not surprising that blade vendors are working hard to fold advanced storage techniques into their products. HP will be delivering direct SAN attach ports (via fiber channel, or FC) early next year on its blade backplane. Expect also other SAN interconnect standards like Infiniband and iSCSI to appear late next year.

According to HP's Parkinson, true failover clustering from server to disk will become possible with when FC SAN connectivity arrives in its p-Class servers. And working in tandem with server clustering over, say, a Beowulf solution such as Scyld, this means that blade servers can acquire robustness that will probably surpass rack-mount servers.
In fact, Parkinson believes that Infiniband, with a switch-based architecture well suited a variety of hardware standards, can drastically enhance the blade paradigm. "With Infiniband, you get redundancy throughout," he said. This translates to a more robust server and storage cluster with better load balancing capabilities. And might a SAN itself be subsumed in a blade rack? Already, vendors are offering blade modules that act as dedicated storage devices. Dell's upcoming brick server will take this further by separating processor from disk. And with FC now being built into backplanes, the storage potential for blades looks immense.

Better control
Besides the speed, form factor and processing power components, intangible features like manageability and server aggregation will also become better. In fact, these "intangibles" are probably more important for enterprises in the log run.

For many vendors, these are also the intangibles that define their strategies. For instance, Sun's blade strategy calls for a complete framework integration that encompasses processor optimization (with SPARC microprocessors), OS enhancements (with Solaris) and middleware availability like Java, Grid Engine and SunONE. The idea is to have integration beyond the racks of blades.

"The main aim is to achieve high efficiency," said Chong Soon Cheong, Senior Manager, Product Management, Hardware, Asia South, Sun Microsystems. "We want to create a multi-layered system management that ranges from Blade, Shelf, Rack to Enterprise level will be the basis for integration into the overall architectural design." Good obviously, if you are using Sun hardware or related applications.

HP touts Compaq's Insight Manager management software as its crown jewel in its blade product. With it, MIS can hot-swap disks (useful when retiring or reconfiguring servers), populate applications using pre-configured disk images and manage both HP blade and rack servers using a common console.

Similarly, Dell points to manageability as the key selling point for its blades. Dell's Crotty pointed to the 1655MC's integrated management card and embedded KVM switch as key highlights. The integrated management card provides chassis and blade monitoring and remote power control, whereas the OS independent keyboard, video, and mouse (KVM) embedded switches enables remote console access.

And improving access—from networks, other blades and hardware—is probably blade vendors' biggest focus in the near future. Remember that blade is essentially a distributed computing paradigm. Fostering an infrastructure with lightning-fast access to devices will be key.

This article first appeared in Network Computing - Asian Edition.