EC2: Pricing Options and Networking

In part three we launched a server and paid the regular price. But you can pay far less for the same server, depending on how you use it. In this part we'll also solve a problem left hanging: the IP address that changes every time you stop the server.

Pricing options #

Spot: the cheapest, with a catch #

Spot servers are very cheap, but AWS can take them back whenever it needs the capacity, with only two minutes' notice. That makes them excellent for work that can stop and resume: data processing, video encoding, CI pipelines and containers.

They're wrong for a database or anything that can't survive a sudden stop. You can check current prices yourself:

aws ec2 describe-spot-price-history \
  --instance-types t3.micro \
  --product-descriptions "Linux/UNIX" \
  --start-time "$(date -u +%Y-%m-%dT%H:%M:%S)" \
  --query "SpotPriceHistory[].[AvailabilityZone,SpotPrice]" \
  --output table

How to choose #

The practical rule for any real project is a mix:

1. The minimum that always runs: cover it with Savings Plans.
2. Extra capacity at peak times: On-Demand.
3. Work that tolerates interruption: Spot.

Don't commit to a year before you know your real usage. Run on On-Demand for a few months first, and look at the numbers in Cost Explorer.

Addresses: private, public and Elastic IP #

Every server has a private address for talking inside your network, and may have a public address reachable from the internet.

An Elastic IP is a fixed public address you own in your account and attach to any server. You can move it to another server if the first one fails:

# Reserve a fixed address
aws ec2 allocate-address

# Attach it to the server (replace the IDs with your own)
aws ec2 associate-address \
  --instance-id i-0123456789abcdef0 \
  --allocation-id eipalloc-0123456789abcdef0

# When you're done: detach and release it so you stop paying
aws ec2 disassociate-address --association-id eipassoc-0123456789abcdef0
aws ec2 release-address --allocation-id eipalloc-0123456789abcdef0

Most of the time you don't need an Elastic IP at all. The cleaner solution is to point a domain name at a load balancer, which we'll cover in the high availability section.

Placement groups #

Normally AWS decides where your server lives. To control that, you use a placement group, which has three strategies:

Network interfaces (ENI) #

An ENI (Elastic Network Interface) is a virtual network card that carries the private address and security groups. Its best feature: you can detach it from one server and attach it to another in the same Availability Zone, and the address moves with it. That makes a quick failover to a standby server easy.

Hibernation #

When you stop a regular server, everything in memory (RAM) is lost, and it boots from scratch when it starts again. Hibernation saves the memory's contents to the system disk, so the server resumes where it left off. That helps applications that take a long time to start or warm up a cache.

The conditions: the system disk must be encrypted EBS with enough room for the memory, you enable it when you create the server, and a server can't stay hibernated for more than 60 days.

Summary #

• On-Demand for flexibility, Savings Plans for steady load, Spot for work that tolerates interruption. The smart choice is a mix.
• A public address changes when you stop the server, an Elastic IP stays fixed, and all public IPv4 addresses cost money.
• Cluster for speed, Spread for isolation, Partition for huge distributed systems.
• An ENI is a network card that can move between servers, and hibernation keeps memory and speeds up startup.

Test yourself #

Before you move on, try these questions to check you've got this part. Each one is a practical scenario, and the explanation appears once you answer.

In part five we finish the servers section with storage: disks, system images and shared file systems.