top of page
  • JP

How to save costs on S3 running Data Lake

Cloud services provides useful resources to scale your business faster but not always we can measure cloud costs when we’re starting a business from the scratch or even being a solid business, costs always makes part of the strategy for any company who want to provide a better service.

Me and my teammates have worked in a Data platform based on events enable to process 350 million events every day. We provide data to the client applications and to the businesses teams to make decisions and it always a challenge do deal with the massive data traffic and how we can maintain these data and saving money with storage at the same time.

Storage is too expensive and there are some strategies to save money. For this post I’ll describe some strategies that we’ve adopted to save costs on S3 (Simple Storage Service) and I hope we can help it.


Strategy #1

Amazon S3 storage classes

Amazon S3 provides a way to manage files through life cycle settings, out there you can set ways to move files to different storage classes depending on the file’s age and access frequency. This strategy can save a lot of money to your company.

Working with storage class enable us saving costs. By default, data are stored on S3 Standard storage class. This storage type has some benefits of storage and data access but we realized that after data transformed in the Silver layer, data in the Bronze layer it wasn’t accessed very often and it was totally possible to move them to a cheaper storage class.

We decided to move it using life cycle settings to S3 Intelligent Tiering storage class. This storage class it was a perfect fit to our context because we could save costs with storage and even in case to access these files for a reason we could keeping a fair cost. We’re working on for a better scenario which we could set it a life cycle in the Silver layer to move files that hasn’t been accessed for a period to a cheaper storage class but at the moment we need to access historical files with high frequency.

If you check AWS documentation you’ll note that there’s some cheapest storage classes but you and your team should to analyse each case because how cheapest is to store data more expensive will be to access them. So, be careful, try to understand the patterns about storage and data access in your Data Lake architecture before choosing a storage class that could fit better to your business.

Strategy #2

Partitioning Data

Apache Spark is the most famous framework to process a large amount of data and has been adopted by data teams around the world. During the data transformation using Spark you can set it a Dataframe to partition data through a specific column. This approach is too useful to perform SQL queries better.

Note that partitioning approach has no relation to S3 directly but the usage avoids full scans in S3 objects. Full scans means that after SQL queries, the SQL engine can load gigabytes even terabytes of data. This could be very expensive to your company, because you can be charged easily depending on amount of loaded data. So, partitioning data has an important role when we need to save costs.

Strategy #3

Delta Lake vacuum

Delta Lake has an interesting feature called vacuum that’s a mechanism to remove files from the disk with no usage. Usually teams adopt this strategy after restoring versions that some files will be remain and they won’t be managed by Delta Lake.

For example, in the image below we have 5 versions of Delta tables and their partitions.

Suppose that we need to restore to version because we found some inconsistent data after version 1.

After this command, Delta will point his management to version 1 as the current version but the parquet files related to others version will be there with no usage.

We can remove these parquets running vacuum command as shown below.

Note that parquets related to versions after 1 were removed releasing space in the storage.

For more details I strongly recommend seeing Delta Lake documentation.

Well that’s it, I hope you enjoyed it!

Posts recentes

Ver tudo
bottom of page