I have performed a simple experiment executing the sysbench CPU test to compute large primes on two micro instances. Each instance is located in a separate availability zone in US-East Virgina 1a and 1c. Both instances are running on two different physical hosts. The instance in 1a is running on a Dual-Core AMD Opteron(tm) Processor 2218 HE. The instance in 1c is running on a Intel Xeon 5430 2.66Ghz.

The cpu task involves computing large primes, involving large integer divisions which chews up a large amount of the CPU. As the micros can burst up to 2 ECUs temporarily, a decent performance is observed initially but then the hypervisor throttles usage on both instances and performance drops.

I have found that the instance running in 1c (Xeon processor) to be roughly about 3 times faster at computing the exact same script!

```
Amd Opteron (1a) Micro | Intel Xeon (1c) Micro
```

total number of events: 618 -- 10:45:38 | total number of events: 1992 -- 10:45:52

total number of events: 595 -- 10:45:40 | total number of events: 2039 -- 10:45:54

total number of events: 558 -- 10:45:42 | total number of events: 2079 -- 10:45:56

total number of events: 611 -- 10:45:44 | total number of events: 2075 -- 10:45:58

total number of events: 432 -- 10:45:47 | total number of events: 179 -- 10:46:01

total number of events: 14 -- 10:45:50 | total number of events: 49 -- 10:46:04

total number of events: 14 -- 10:45:53 | total number of events: 52 -- 10:46:07

total number of events: 15 -- 10:45:55 total number of events: 52 -- 10:46:09

total number of events: 23 -- 10:45:59 | total number of events: 50 -- 10:46:12

total number of events: 14 -- 10:46:02 | total number of events: 53 -- 10:46:15

This is a quick output from the sample script I ran with sysbench.

Is this as a direct result due to the processor type on the underlying physical host or is my 1a instance located in a so called noisy neighbourhood. Even after throttling its clear that the Intel based micro instance outperforms the AMD Opteron.

Thanks in advance