From this page I see that t2 CPU can turbo boost up to 3.3Ghz: https://aws.amazon.com/ec2/instance-types/. enter image description here

And this AWS official video said t2 CPU could be 60% faster than m4:https://youtu.be/FkMslBsVYFU?t=447 enter image description here

Then I made a testing:
Launched a t2.micro instances(ubuntu 16.04), use sysbench to make the CPU keep running at 100%. In the meantime I opened another terminal tab and repeated command grep MHz /proc/cpuinfo, but the result was always 2.4 (I expected 3.3). I also made the same CPU benchmark on a m4 instance and the running time was same with t2.

So, how to I make t2 CPU run 60% faster than m4 CPU?


First, you will NEVER see real MHz on cloud/virtual servers due to host OS which is designed to emulate real hardware instead of be the hardware.

Second, the charts you posted metered in ECU, not MHz.

I've made additional research, the results are below. As you probably know, t2 instances have burst capability and during those bursts they run at more than 100% of CPU, however it can be measured in real tasks, not in MHz.

I run two instances: t2.medium and c4.large, as on the first screenshot of your video. All instances were created using the same Amazon Linux 2 LTS Candidate AMI 2017.12.0 (HVM), SSD Volume Type - ami-db1688a2

To create load I used https://github.com/phoronix-test-suite/phoronix-test-suite/ :

yum install -y php php-cli php-common php-gd php-pdo php-process php-xml
mkdir tmp; cd tmp
wget http://phoronix-test-suite.com/releases/phoronix-test-suite-7.6.0.tar.gz
tar -xvf phoronix-test-suite-7.6.0.tar.gz
cd phoronix-test-suite
./phoronix-test-suite install pts/build-apache
./phoronix-test-suite batch-setup # (answers are no and no)

then run each of those tests three times

./phoronix-test-suite batch-run pts/build-apache

and recorded average value for each test, like here:

enter image description here

The results are in the chart, data are seconds, less means better:

enter image description here

The gain is ~20% which is les than on your screenshot but we don't know what tests were used by AWS (I asked but they didn't answer) nevertheless as you can see t2.medium was faster than c4.large.

p.s. I also run bz2 decompress test: t2.medium was slower there perhaps due to more intensive disk usage (I used the cheapest disk option).

| improve this answer | |
  • "Inter Turbo" is listed in the table. The "server burst" should be the "burst above t2 baseline(5%,10%,etc..)" that spend CPU credit, that's not what I was talking about. I have updated my question please take a look. – Elect2 Jan 22 '18 at 22:41
  • @StanHou updated. – Putnik Jan 23 '18 at 9:09
  • ,you said "measure performance in ECUs". Did you mean that ECU is not a measure of CPU speed? – Elect2 Jan 23 '18 at 10:04
  • @StanHou Imagine you are riding a bike inside flying plane. Speedometer of the bike will not show you your speed in relation to ground. The example is not strictly correct but you have the idea. According to the screenshot, ECU measures performance, but "Effective ECU" != MHz and speed != performance. – Putnik Jan 23 '18 at 10:11
  • So in your thought, what is exactly "Effective ECU", and why you think the benchmark can not measure Effective ECU? – Elect2 Jan 23 '18 at 11:21

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.