This article provides a detailed overview of best-practices for AWS S3 on how to improve performance, also highlights few of the use-cases in general.

AWS S3 Design Patterns for Optimizing Performance

AWS S3 is widely adopted by majority of the users whose day to day interactions revolve around storage, cloud operations & deployments. AWS S3 offers high scalability. Uploading & downloading objects is easy as S3 supports a very high transaction rate (few hundreds to thousand transactions/second). As there are no limits on how to use S3, It’s better to understand the nature of S3 service and make your applications suitable to achieve great performance.

Let us go through the guidelines for optimizing performance and best-practices for designing your applications that are using S3.

Horizontal  Storage Connections
  • AWS S3 is not a typical storage server, it identifies itself as massive distributed system. AWS suggests to leverage multiple concurrent requests to Amazon S3 while it can be uploading (or) retrieving etc.
  • It’s best recommended to spread all the requests over different connections to maximize the available bandwidth from AWS S3.
  • There are practically no limits for the number of connections made to your S3 bucket.

Byte-Range Fetches
  • Use Range HTTP header in GET request, this allows user to fetch a specific byte-range from any object & transmitting only the required portion.
  • Concurrent connections can be used to fetch different byte ranges from within the same object from Amazon S3. This improves the overall throughput when compared with single whole object request.
  • 8-MB (or) 16-MB can be best fit sizes for byte-range requests.

Leverage S3 Transfer Acceleration
  • It’s advisable to use S3 Transfer Acceleration(S3-TA) to transfer files securely over long geographic distances between the client & an S3 bucket.
  • S3-TA uses AWS CloudFront global edge locations. When data hits an edge-location, AWS will use an optimized n/w path for routing to its respective source.
  • When transferring GBs/TBs of data globally, it’s recommended to leverage S3-TA.
  • A separate S3-TA can also be used depending on your data origin. Other conditions like n/w configs, routing etc. will be varied from different time zones & locations.
  • The users are only charged for data-transfers if S3-TA improved their upload performance.

Latest Version of the AWS SDKs
  • It’s recommended to choose the updated/latest SDKs which can take advantage of latest best practices imbedded in them.
  • For Ex, most of the active SDKs include default-logic to automatically retry requests on 5XX errors & are actively contributing code to support & respond to slow (or) delayed connections.
  • Leverage HTTP REST API requests to optimize performance by following multiple connections to allow fetching of object data in parallel, retires on slow requests etc.

Using Caching for Frequently Accessed Content
  • S3 is actively used as a centralized datastore where a “working set” of data is repeatedly requested by users. Other AWS services like CloudFront, ElastiCache & Elemental-MediaStore can assist users in optimizing performance by caching data based on requests.
  • Adopting & implementing caching practices in architecture can reduce latency and improve data-transfer rates. Apps that are built using cache mechanisms will send only few requests when necessary and greatly reduces costs as well.
  • AWS Cloudfront caches data over at edge-locations all over the world. It will be a great advantage while using S3 with Cloudfront.
  • AWS ElastiCache is a managed, in-memory cache. This provisions EC2 instances in the specified region that will cache objects in its memory. This improves latency with GET requests & improves overall throughput.
  • AWS Elemental MediaStore is similar to Cloudfront in few ways but solely focused to assist video & media driven apps to imporve their workflows using S3.

Timeouts and Retries for Latency-Sensitive Applications
  • S3 is designed to be  intelligent enough to scale itself when necessary. But there are few scenarios user receives temporary HTTP 5XX request responses. The possible cause here is , S3 is working internally to optimize for a different request rate.
  • S3 advises tracking and aggressively retrying slower operations for latency-sensitive applications.
  • Stable & consistent response times are typically provided for every fixed-size requests. I you feel something’s wrong,
  • S3 provides more consistent response times for each of the fixed-size requests. If seeing abnormal behavior, find the slowest 1% of requests and then retry them. Sometimes, even a single retry is greatly efficient at reducing latency.
  • Generally when using a massive requests, AWS suggests to check how much throughput is being achieved & retry the slowest requests. Try a retry after couple of seconds & other after few more seconds.

Final Thoughts

Previously, random prefix names are used to improve S3 performance and it was suggested by AWS itself. Now, its no longer necessary and can use a simple date-based naming patterns.  These recommendations changes with time as AWS team is working on improving S3 along the way. But, As of now, the above patterns discussed can be a great start for optimizing S3 to improve performance.

Here are few awesome resources on AWS Services:
AWS S3 Bucket Details
AWS S3 Bucket Versioning
AWS S3 Life-Cycle Management
AWS S3 File Explorer
Create AWS S3 Access Keys
AWS S3 Bucket Costs
AWS S3 Custom Key Store

  • CloudySave is an all-round one stop-shop for your organization & teams to reduce your AWS Cloud Costs by more than 55%.
  • Cloudysave’s goal is to provide clear visibility about the spending and usage patterns to your Engineers and Ops teams.
  • Have a quick look at CloudySave’s Cost Caluculator to estimate real-time AWS costs.
  • Sign up Now and uncover instant savings opportunities.


Nikhil Araga is working as a DevOps Engineer who works with Cloud Migrations & DevOps team. He is proud to be part of a team who religiously follow Agile and has a SHIFT-LEFT mindset. He is actively working to create multiple CI/CD pipelines & stable deployment patterns. His key areas of interests include, Cloud Security, FinOps and DevOps practices.