Blade server market leaders IBM and Hewlett-Packard Co. are both pouring resources in to developing their wares....
By submitting your email address, you agree to receive emails regarding relevant topic offers from TechTarget and its partners. You can withdraw your consent at any time. Contact TechTarget at 275 Grove Street, Newton, MA.
Most recently, IBM added 10 GigE to its BladeCenter H platform. But, so far, these efforts haven't been enough to convince Rod Lucero, CTO of the Minnesotan VMware consultancy VMPowered, to recommend blade servers to customers undertaking large-scale virtualization and consolidation projects.
Instead, Lucero and his team prefer the standard issue rack-mounted servers. Over the years, he's recommended IBM and Dell, currently favors HP's DL385 and DL585 line, and sees the tide swinging back to IBM with its new quad-core Xeon.
The big problem with blades, Lucero said, is the limited number of network interface cards (NICs) you can attach to a blade -- "two, and that's it," he said, although some newer blade systems are starting to support more.
In rackmount server configuration, VMPowered recommends servers with large numbers of NICs – "never less than six [ports]," Lucero said, and as many 14.
"You really need a lot of I/O to sustain a virtualization infrastructure," he said.
In blades, Fibre Channel and memory capacity are similarly limited relative to rackmount servers.
Not sold on boot-from-SAN
Lucero's other gripe with blades is that they are often sold diskless, thus requiring the blades to boot from SAN – a less-than-perfect process.
For one thing, VMware ESX is very particular about which logical unit number (LUN) it boots from. Also, if you boot from SAN and your SAN crashes, you're stuck. But if the ESX server boots from local disks, it's a relatively easy process to point it to another source of virtual machine images.
In short, boot from SAN "works, but we've seen it be problematic," Lucero said. "We'd certainly never recommend it, especially not with the cost of internal disks today."
Safety in numbers
If you are going to go with blades, you're best off buying them fully populated, Lucero opined, for better negotiating power. Of course, if you do that, you lose one of blade's main selling points – the ability to easily add in more capacity. At the same time though, Lucero said he's often seen companies invest heavily in a sparsely populated blade chassis, only to find that they don't really need the extra capacity down the road.
"Then you have a huge amount of infrastructure that you're not using," he said.
Lucero concedes that a blade chassis does take less power than a comparable rackmount setup. An IBM BladeCenter H, for example, is rated to about 60 amps, while a typical 42U rack requires about 220. But with the consolidation ratios you can get out of hefty rackmount servers, Lucero's not sure it matters. Putting 15 2U servers in a rack, each with two dual cores, could easily give you a consolidation ratio of 20:1, Lucero said – "that's 300 VMs right there!"
Let us know what you think about the story; e-mail: Alex Barrett, News Director
Dig Deeper on Improving server management with virtualization