Close

Request a demo

CliQr's groundbreaking technology makes it fast and efficient for businesses to move, manage and secure applications onto any private, public or hybrid cloud environment.

CliQr is now part of Cisco Learn More About Cisco
Request a Demo

Blog

#AWS instance types for #HPC workloads

When choosing an instance size for your high CPU workload, you have an important decision to make.  Do you want speed or low price or a combination of the two?  How do you find all that out so you can make an informed decision?  Benchmarking.

Benchmarking 16 Task Blender on AWS

As an example, we decided to test out AWS high CPU workloads using Blender, the popular open source high resolution graphics and animation engine.  Specifically, we used a test workload with 16 different .blend files containing data used for the 3D rendering of animated houses.  We deployed this on a single node and spun up 16 tasks per node in an attempt to take better advantage of multiple processors available on a given VM.

Here’s how those results played out, both in graphical and tabular form with on demand pricing used:

 

16TaskBlenderAWS

Instance Name

vCPU

RAM

Time (hrs)

Cost Per Job

m3.large

2

7.5

3.17

$0.56

m3.xlarge

4

15

1.58

$0.56

m3.2xlarge

8

30

0.83

$0.56

c3.4xlarge

16

30

0.42

$0.84

 

Analyzing Results

What’s really interesting about these results is the cost per job remaining fixed for the first three instance sizes, indicating that (unsurprisingly) AWS knows exactly what it’s doing when pricing their instance types.  Since our workload broke up its tasks into 16 different processes, the performance across all 4 instances halves each time you double the number of processors.  For the first three instance sizes, the price doubles as well which translates into your job cost being the same no matter which of those three instances you choose.

There is a price premium, however, for the C3.4xlarge in our sample.  With its 16 CPUs, it finished the job in roughly half the time as its 8 CPU m3.2xlarge sibling, but is priced more than double its predecessor.  Depending upon your use case, for example if you need your job run as quickly as possible no matter what the cost, this might not be a big deal to you.

Regardless, this is exactly the type of raw data an organization needs to make instance type decisions when running HPC workloads on public clouds.  Only through benchmarking that considers both price AND performance can you get what you need to make an informed decision.  At that point, you can take the specifics of your situation into account and decide for yourself.

Leave a Reply

Your email address will not be published. Required fields are marked *


4 + = five

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>

Stay Informed

Subscribe to our email list for company updates.

CliQr Technologies, CliQr CloudCenter and CloudBlades are trademarks of CliQr Technologies, Inc. All other registered or unregistered trademarks are the sole property of their respective owners.