In x86 circles, there’s a new wrinkle in the old scale-up vs. scale-out debate: VMware vSphere 5 licensing.
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.
Last week, the market-leading virtualization vendor introduced a new licensing scheme to go along with its latest vSphere 5 platform. Instead of simply charging per processor as in the past, a per-processor license also provides pooled virtual RAM (vRAM) entitlements. That is, each vSphere 5 license allows one processor and 24, 36 or 48 GB of allocated RAM across a VMware farm, depending on the license type.
Charging for memory utilization sparked a small firestorm among VMware users, many of whom have been outfitting their servers with far more memory than the new licenses allow for, and who are facing additional license purchases in order to upgrade.
Further out, the new licensing may change how IT architects design and buy servers for their VMware estate.
Scale-up vs. scale-out revisited
Under the new model, the central question is whether it is better to buy a single “scale-up” box with two-processors and 192 GB of RAM, or two “scale-out” dual-processor boxes with 96 GB of RAM. From a VMware licensing perspective, the costs are the same, but many IT pros don’t like the idea of paying for RAM without the benefit of the additional CPUs.
“We’ll probably implement more CPUs within the environment, since we’ll be buying CPUs we’d never be buying in any other case,” said Kent Altena, a technical engineer at a midwestern financial services firm. Under the new licensing scheme, the only way to obtain rights to additional vRAM allotments is to more licenses, which are also tied to the CPU.
Going the scale-out route does have some advantages, Altena said, especially in terms of availability. “When something goes down, it affects fewer VMs,” he said.
But that shift away from scale-up also has its drawbacks. For one, Altena said, more servers translate to more opportunities for error, during configuration time or as part of routine maintenance.
Not only that, but implementing more CPUs as a way to obtain more vRAM has its own financial penalties. When you think in terms of ports, power, rackspace and maintenance, “buying servers costs money,” said Bob Plankers, an IT architect at a large midwestern university. “It’s not free at all.”
Indeed, VMware licensing costs aside, the economic advantages of hardware consolidation via scale-up virtualization are significant, said Gary Chen, research manager for enterprise virtualization software at IDC.
“You may not be saving on licensing as much, but there’s still a lot to be said for hardware consolidation in terms of efficiency and power and cooling,” Chen said.
Users considering a switch to a more scale-out model also need to think about Windows Server operating system licenses, wrote Aaron Delp, an IT consultant and blogger. At $3,000 per processor for a Windows DataCenter Edition license, it is typically more cost-effective to do Windows virtualization on fewer processors rather than more, he argued.
Once the dust settles and the disappointment with the new licensing subsides, Plankers said he believed most of his peers will decide to stick with memory-dense, scale-up boxes.
“Fewer bigger machines are better. It may not be as cost-effective as it was a few weeks ago [before the licensing change] but it’s still more cost-effective.”
The middle path
Others advocate a moderate approach: just-right servers with neither too much RAM or too little.
“I don’t like servers with 32 or 64 GB of RAM, but I also don’t like servers with 256 GB,” said Mark Vaughn, another IT consultant and VMware vExpert. At the high end of the spectrum, big memory DIMMs are not only very expensive, but they are also slow.
“I like servers with 96 GB,” he said. That way, you maximize your vSphere vRAM entitlement, get good availability, take advantage of cheaper DIMMs -- all without sacrificing speed.
In other words, “build the biggest server you can as makes financial sense, and then go horizontal.”