Ask HN: How will using containers save me money?

1 points by sly010 10y ago ↗ HN

I am in a situation, where I am running multiple small independent apps in the cloud. They are each very low traffic with occasional traffic spikes, so they require very little resources on a day-to-day basis.

In the ancient world they would be running on a large colocated server under different users, perhaps using something like chroot, etc. The resources would be very well utilized but it would be a security and operations nightmare.

In the IAAS world I can rent instances of various sizes, one per my small application. This would improve on security and operations, but the resources are horribly allocated. I need every server to be sized so it can carry it's app at peak traffic, but otherwise they are all idling most of the time.

In the container (CAAS?) world security and operations are basically solved, but I am still not convinced about resource utilization.

Both AWS and GCE promise to deploy containers on top of their "compute platform" and charge me for used instances + some more for the management of the container platform.

So it seems by using containers to deploy my application I am still paying for the same underutilized resources, while giving the platform provider the possibility to optimize things behind my back. Basically I am saving THEM money.

I know could just pay for one big box and run all my applications on it, running docker and deploying containers by hand, but then I would be back managing things myself.

So how can I use containers to better utilize resources and save ME money?

4 comments

[ 4.6 ms ] story [ 20.5 ms ] thread

tmaly 10y ago ↗

I am still running micro-services outside containers. The main problem for me is upgrading the various programs nginx, postgresql etc. I have to recompile code by hand in many cases instead of having it automated in containers.

This is where I think containers would save me the most time and money.

moondev 10y ago ↗

Not all application resource demands fit nicely inside instance types.

Once you determine your application needs, you can set your cpu/memory resource needs per container/service. Then you can run these containers inside your fleet more smartly.

This really comes into play for memory and cpu ratio. Look at the price difference between:

one m4.2xlarge = 32gb mem, 8 cores = $350 a month

eight t2 medium = 32gb memory, 8 cores = $300 a month

You get twice as much compute for less money with the ability to mix and match service needs instead of being constrained to individual instances.

sly010 10y ago ↗

The key thing I missed is the fact that I am not limited to running a single container per instance. I never actually tried to run anything on either container engine and from their high level marketing blabla this was not immediately obvious.

Yep, you can run as many containers as resources available.

You set the cpu limit per container based on cpu units. There are 1024 available per core.