Amazon Web Services added a new measure of server performance and capacity to its Elastic Compute Cloud management...
By submitting your personal information, you agree that TechTarget and its partners may contact you regarding relevant content, products and special offers.
console to complement its existing EC2 Compute Units.
While virtual CPUs (vCPUs) are now the unit of measure on Amazon's EC2 instance description page, EC2 Compute Units (ECU) have not disappeared from Amazon Web Services' (AWS) management console. The change to vCPU doesn't impact billing, either.
ECU provides the relative measure of the integer processing power of an AWS EC2 instance, according to an Amazon FAQ. Amazon noted it may add or substitute measures into the definition of an ECU if those metrics give a clearer picture of compute capacity.
Customers will get that clarity with the addition of vCPU -- a more standard measure of shared CPU power used in clouds such as VMware's vCloud.
When a new instance is launched via the AWS Management Console, the cloud operator will see ECU values alongside the new vCPU measurement, along with information about the underlying hardware. For instance, for M3 instances, each vCPU is a hyper-thread from an Intel Xeon E5-2670 processor.
"In general, ECUs were hard and we often found ourselves converting to cores -- closer aligned to vCPUs," said Jim O'Neill, CIO for online marketing software company HubSpot Inc., based in Cambridge, Mass.
While the ultimate impact is unclear, O'Neill said "over the long term, fewer proprietary terminology or frameworks [are] good for the industry at large to help compare apples to apples vs. apples to oranges."
Jim O'NeillCIO, Hubspot Inc.
The change led to some speculation last week that Amazon might offer build-your-own instances with a mix-and-match inventory of CPU, memory, storage and network resources -- as opposed to pre-baked instances, which AWS offers now -- but that appears unlikely, according to Kyle Hilgendorf, analyst with Gartner Inc.
"I don't see any tea leaves yet as to whether [it] would or would not do that," Hilgendorf said. "Keep in mind, full customization of instance sizes and granular pricing could wreak havoc on capacity management and performance."
This is more of a simplicity move, Hilgendorf said.
"Imagine a day a few years down the road with Moore's Law where even a small instance would have to start being described with 10 ECUs," he said. "At some point, that starts to sound silly."