IBM BladeCenter HPCWorld Rating
Everything about the IBM BladeCenter H just screams IBM, from the mainframe-like aesthetics to the spartan management interface -- even the "H" model name. Severe-looking matte-black chassis notwithstanding, the BladeCenter H matches most of the major features of the Dell and HP blade systems but at a higher price. The BladeCenter also comes in smaller portions: IBM's slightly shorter chassis holds fewer blades than the Dell and HP enclosures.
With features like the Virtual Fabric virtualized network I/O, the HS22V blades specifically designed for virtualization, and reasonably complete embedded management tools, the BladeCenter H is a solid solution. But the higher price, lower density, and lack of advanced embedded management capabilities are drawbacks.
[ InfoWorld compares the leading blade server solutions: "Blade server shoot-out: Dell, HP, and IBM battle for the virtual data center." ]
Chassis and blades
The IBM BladeCenter H is a 9U chassis that holds 14 blades -- again, 2 fewer than the HP and Dell blade units. Each blade is surprisingly slim, but they're otherwise nearly identical to the others' blade offerings. There's no heads-up LCD, and several of the status lights are small enough to be difficult to see at a glance. The H chassis is also unique in having a vertical DVD-ROM drive set into the bezel on the right-hand side; each blade has a Media Tray button that maps the contents of that drive (or an image via the management console) to that particular blade.
In the rear, there are eight I/O slots, split into two groups of four. The four traditional modules support lower-bandwidth I/O, including gigabit Ethernet and 1x InfiniBand or Myrinet fabrics. The other four modules are tasked for high-bandwidth I/O, including 10G Ethernet and 4x InfiniBand. As with the others, they're mapped 1-to-1 to the blade I/O ports, so you'll need a matched set of I/O modules in the rear to fully support the internal I/O in each blade.
The internal I/O of each blade is more like Dell's than HP's: The gigabit Ethernet interfaces are embedded on each blade, and 10G interfaces are supplied on a mezzanine card. However, like HP, IBM includes a virtual interconnect: IBM's Virtual Fabric for 10G interfaces represents a 10G pipe to a blade as four Ethernet ports that can each be tuned for specific functions or bandwidth caps. Like the HP equivalent, configuring the Virtual Fabric modules is nothing like configuring normal switches, and the IBM techs didn't really know how it worked. We couldn't even get into the configuration for those modules to poke around.
There are two main blade models: the HS22 and the HS22V. The HS22 is built on the model of a traditional blade, with two 2.5-inch SAS hot-swap drive bays, two CPU sockets, and 12 DIMM slots. The HS22V is a different unit altogether, adding 6 more DIMM slots and deleting the 2.5-inch SAS drive bays in favor of two 1.8-inch SSD non-hot-swap bays internal to the blade. This is a blade designed for virtualization with the expectation that a hypervisor will be loaded on the mirrored SSDs and all storage will come from an attached SAN. The HS22V is a great way to expand the RAM footprint of each blade without sacrificing space or additional power required to run SAS drives. HP offers a similar blade, the HP BL490c, but did not bring it to the test.
It's also notable that IBM had the only redundant internal power and I/O connectors on each blade. The midplane is not redundant, but the connections to each blade are, which offers some protection against connector and interface failure at the blade level.
IBM's Web-based management console, called Advanced Management Module (AMM), is clean and sparse. There's no graphical representation of the chassis; in fact, there aren't any images at all. The left-hand frame offers a hierarchical view of the configuration, status, and management options, and the main frame holds the focus of the current selection. While the GUI is very plain, it's also reasonably simple to find what you're looking for. As a stand-alone management console, the AMM is sufficient but not particularly impressive.
Like the Dell manager, the AMM provides links to the dedicated management tools for the I/O modules. However, we were not able to log into anything but the Fibre Channel switch, as there was a problem gaining access to the Virtual Fabric switches directly. As with HP's Virtual Connect, dealing with the Virtual Fabric configuration is not nearly as simple and straightforward as configuring standard Ethernet switches.
There were some other oddities in the AMM, such as the fact that when you bring up a listing of the I/O modules, the table view doesn't list the IP addresses of the modules. Instead, it shows a View link, requiring you to click the link to pop up a window that displays the IP address. Such extra steps quickly become bothersome.
IBM's remote KVM tool is functional, but somewhat difficult to use for certain tasks. For instance, mounting CD/DVD and floppy images is odd in that you can select an image for mounting, and then move it between blades, but you can't easily mount different images to different blades in the same session. However, it is relatively simple to move between blade consoles within one remote KVM console window.
What is rather more impressive is the marriage of the BladeCenter H and IBM Director, IBM's global mastermind tool. Director is a dedicated server monitoring and management tool that provides problem alerting and handy global management capabilities such as automated actions when failures occur. It's free with any IBM hardware, though some advanced capabilities are provided through for-fee plug-ins. Global management tools are available in various forms from Dell and HP as well. They are largely outside the scope of this review, but they warrant a mention.
Unlike the Dell and HP chassis, the IBM BladeCenter H lacks dynamic power and cooling capabilities. Instead of a bevy of smaller fans, the chassis has two large fans that ramp up and down as load dictates. The power supplies are also not as adaptable as those in the Dell and HP chassis. This is a box meant to work long and hard, at the expense of some power savings. There are facilities to implement power capping, as in the HP c7000, though this capability comes at an extra cost.
Without the dynamic power and cooling features found on the other chassis, the IBM solution drew more power during our power measurement tests, topping off at around 1.4kW with two blades pushing a full load and maintaining a 1kW average at idle with two blades inserted and running without load.
[ Return to "Blade server shoot-out: Dell, HP, and IBM battle for the virtual data center" | Read the review of the Dell PowerEdge M1000e, HP BladeSystem c7000, IBM BladeCenter H, or Supermicro SuperBlade. ]
- Tsunamis and falling crates: Behind the blade server review
- InfoWorld review: Intel's Westmere struts its stuff
- Modern multi-core and the next generation of IT
- Intel's Nehalem simply sizzles
- InfoWorld review: Dell, HP, and Lenovo rack servers
- InfoWorld review: Dell's virtualization servers surge ahead
- Nehalem tower servers: Dell, Fujitsu, HP square off
- Last of the red hot Sun servers
Read more about computer hardware in InfoWorld's Hardware Channel.
Paul Venezia is senior contributing editor of the InfoWorld Test Center and writes The Deep End blog.
IBM BladeCenter HPCWorld Rating