CODEWITHSUNDEEP
amazon-s3cassandrabackupstorage
Jay Desai
Jay Desai

Reputation: 11

Ephemeral vs. EBS vs. Amazon S3

I am trying to determine what storage to use for Cassandra and Cassandra backups. EBS is the most expensive of the lot so that one I may have pretty much eliminated. ephemeral storage is cheaper compared to S3 if I am planning to commit to pay upfront for 3 years. Not sure how it compares relative to S3 in terms of performance. Anyone have experience for the two use cases (running a database and storage for backup)?

Upvotes: 1

Views: 1835

Answers (2)

Michael B
Michael B

Reputation: 12228

The idea is that you store your database files on EBS, your cache files on ephemeral storage and your backups are saved to S3.

Ephemeral storage is purely for the purpose of temporary caching. It is deleted whenever the server reboots. If the box goes down you lose everything on it. DO NOT store important data on Ephemeral storage!

EBS is a SAN attached disk, you can stripe data across them to produce the performance you need. It is designed for storing working data on, and is safe from anything that could happen to the VM.

S3 is akin to a remote file share, it doesn't have the performance you need to install working data to, and can't (easily) be mounted, hence is ideal for backing up to. You can then have archiving and long term storage options set.

Edited to be more Cassandra specific

While Cassandra is designed to be fault tolerant, and accept the complete loss of cluster nodes. Unless you are designing it to be heavily distributed across multiple regions you still run the risk of losing all of your data up to the point of your last back up if you don't use EBS volumes. Since you are concerned about cost I am guessing that you aren't looking to be massively distributed, so it would be more cost effective to have all of your databases on EBS volumes that way if a region fails you won't lose data to ephemeral storage.

Upvotes: 2

Chris Lohfink
Chris Lohfink

Reputation: 16410

With Cassandra its recommended to use ephemeral for you data and backups to S3. EBS is generally considered an anti-pattern and not recommended but the ssds can work. Check out http://docs.datastax.com/en/cassandra/2.1/cassandra/planning/architecturePlanningEC2_c.html

Upvotes: 4

Related Questions