Joined: 6 Jun 2011 15:51:29
I've recently made a change to the number displayed in the CPU load chart to fix the problem with instances with large EC2 instances reporting incorrect high load percentages in the load chart and load bar. The CPU load chart and cluster table now displays the actual 15min load average as read from /proc/loadavg. (http://en.wikipedia.org/wiki/Load_%28computing%29)
The load % number is usually calculated by dividing the load average, as read from /proc/loadavg, by the number of CPUs. The problem with this calculation used in PPCD instances on EC2, is that in order to correctly calculate the load for an EC2 instance we need to divide this number by the number of EC2 Compute Units an instance is allocated. This caused users with very large instances with many CPUs to get a very high numbers in the load chart, which was deceiving, since the number actually still needed to be divided by # of EC2 Compute Units. For instance, in the case of a m2.2xlarge the load % would have to be divided by 13 to get the actual load %. (http://aws.amazon.com/ec2/instance-types/)
Since there does not appear to be an API method to retrieve the number of EC2 Compute Units associated with an instance, I think displaying the actual /proc/loadavg number is the best thing we can do for now.