Benchmarking a t2.micro EC2 Instance

I was curious how much performance you can really get out of a free tier EC2 instance, so I decided to run some benchmarks on one of them. The "t" class burstable instances seem to benchmark inconsistently at times, so I ran these benchmarks with multiple attempts.

Benchmark Setup

First, I started with a t2.micro (1 CPU, 1 GB memory, 10 GB disk) EC2 instance in the US-West2 (Oregon) region running RHEL 7.4.

I then installed the sysbench benchmarking tools with the following commands

yum -y update (Because patching your boxes is important, dawg!)

yum install https://dl.fedoraproject.org/pub/epel/epel-release-latest-7.noarch.rpm

yum install sysbench

Basic Benchmarks

After I got sysbench installed, I ran these rather basic CPU benchmarks a few times:

sysbench cpu --cpu-max-prime=6000000 --threads=1 run

sysbench cpu --cpu-max-prime=6000000 --threads=2 run

A t2.micro instance only has a single CPU, so the results for a dual threaded workload were basically the same as the single threaded workload. In both cases, the benchmarks completed in roughly 19 1/2 seconds. 

How fast is that? Not all that fast. When I ran this same benchmark on a Core i3 based desktop, it also completed the benchmark in about 19.5 seconds.

I then followed that up with a basic disk I/O benchmark:

sysbench --num-threads=16 --test=fileio --file-total-size=1G --file-test-mode=rndrw prepare

The creation process takes 15.8 seconds, at a rate of 65 MB/sec

sysbench --num-threads=16 --test=fileio --file-total-size=1G --file-test-mode=rndrw run

Here are the stats from this run:

File operations:

    reads/s:                      1790.95

    writes/s:                     1193.44

    fsyncs/s:                     3813.25

Throughput:

    read, MiB/s:                  27.98

    written, MiB/s:               18.65

General statistics:

    total time:                          10.0346s

    total number of events:              68225

Finally, let's clean up after ourselves:

sysbench --num-threads=16 --test=fileio --file-total-size=1G --file-test-mode=rndrw cleanup

I then ran this benchmark a second time, and got basically the same result.. a 10 second run.

Exploring CPU Throttling 

The thing to remember about "t" class EC2 instances is that the performance on them will drop if you run a constant workload on it. I'm not sure what the "t" really stands for in an AWS EC2 instance parlance, but perhaps it should stand for "throttled". 

To get a good idea how the system will perform under a constant load, we should run these benchmarks for a longer time and see what happens when AWS starts throttling us.

So, let's keep the system busy with a longer CPU benchmark:

sysbench cpu --cpu-max-prime=120000000 --threads=2 run

This helps to keep the CPU busy at 100% utilization for awhile. As a result, we start to burn though our CPU credits as shown here:

I was able to burn though the CPU credits for this instance in about 45 minutes. I then run my original CPU benchmark from before:

sysbench cpu --cpu-max-prime=6000000 --threads=2 run

It now takes 121.25 seconds to run. BIG performance difference. 

When I get some free time, I should run these benchmarks on larger instance types and see if I get similar results. Stay tuned...