✅ Managing Terabytes of Data with Amazon S3.pdf
0 likes44 views
The document discusses the management of large data using Amazon S3, highlighting its capabilities as a managed object storage service since 2006. It details various storage tiers, optimization strategies, and a specific application use case involving 150TB of data, where transitioning between storage classes significantly reduces costs. Key learnings emphasize the importance of periodic analysis and the effectiveness of Amazon S3 in providing reliable data storage solutions.
✅ Managing Terabytes of Data with Amazon S3.pdf
- 1. Managing Terabytes of Data with Amazon S3 Dhaval Nagar AWS Hero, AWS SME TRACK 01 - MODERN APPS - SERVERLESS
- 2. ● Founder @ APPGAMBIT, AWS Consulting Partner ● 12x AWS Certified ● AWS Hero (since 2020) ● AWS Certification SME ● AWS Surat User Group Lead ● Practicing Barista Introduction Dhaval Nagar
- 3. Agenda ● Amazon S3 - Storage Powerhouse ● Use Case ● Storage Optimisation ● Storage Cost Breakdown ● Key Learnings
- 4. Amazon S3 - Storage Powerhouse ● One of the earliest and oldest AWS services, released in 2006 ● Managed Object Storage Service ● Designed to be used over HTTPS APIs to do simple file operations (Simple Storage Service) ● Works as a backbone for many AWS Services ● Dozens of features from Different Storage Tiers, Versioning, Web Hosting, In-Place Data Queries, Event Notifications, etc ● Manages over 100 trillion objects and manages millions of requests per second ● Famously known for 11 9’s of Durability
- 5. Use Case ● Application captures and processes Large 3D Video Files ● Requires Multi-tenant Storage with current data of over 150TB ● Amazon S3 is used to save the raw data files before processing ● Raw files are fewer in quantity but quite large in size ● Raw files are then processed to generate output files ● Output files are smaller in size but millions in quantity ● Key Objectives: ○ Achieve optimal storage organization based on the processing workflow ○ Generate Per-Tenant Storage cost breakdown
- 6. Job Scheduled (Auto / Manual) Video Files Uploaded Processed Files Validated (Auto / Manual) Original Video is ready for Archival Use Case - Status
- 7. ● Amazon S3 offers many storage classes like Standard, Standard Infrequent Access, One-Zoned and different Cold Storage classes in Glacier ● Each Tier emulates a real-world use case to keep data in specific storage type ● Standard is the most common and the most expensive tier ● The starting point is to enable the Intelligent Tier!! However, that is not the optimal choice in all conditions. Storage Optimisation for Raw Files S3 Standaard S3 Intelligent-Tiering S3 Standard-IA S3 One Zone-IA S3 Glacier Instant, Flexible, Retrieval and Deep Archive
- 8. Intelligent Tier ● Intelligent Tier is perfect for unpredictable usage patterns ● Internally using multiple tiers to automatically move objects based on the age of the object
- 9. Intelligent Tier Internal Transitions Frequent Access Same as Standard Tier This is default tier when object storage class is set to Intelligent Tier Infrequent Access Same as Standard Infrequent Access Tier IA is applied when an object is not accessed for 30 consecutive days Archive Instant Access Same as Glacier Instant Object is archived if not access for 90 consecutive days Archive Access Same as Glacier Flexible This is optional transition Deep Archive Same as Glacier Deep This is optional transition
- 10. ● With Intelligent Tier, will need to wait for next 30 days to switch the tier ● With Manual Storage Class update, the Infrequent Access can be switched directly and avoid the 30 days of standard cost - which is 50% ● With average 10 TB of monthly uploads, the Standard tier will cost around $230 for new data Reconsider Intelligent Tier for Predictable Flow File Uploaded File Processed File Unused Day 1 Day N+30 Day 3-5 Move to IA File Uploaded File Processed Day 1 Day 3-5 Move to IA
- 11. ● Output files are configured to use the Standard Intelligent Tier ● The consumption of the processed images are infrequent and unpredictable ● Intelligent Tier gives the best of both the worlds
- 12. Storage Tier Decision Tree (CloudHealth by VMware)
- 13. ● Standard Infrequent Access is another possible storage option ● It is 50% cheaper compared to Standard Frequent Access tier ● However, AWS charges prorated monthly cost for the first 30 days if the objects in the Infrequent Access tier are moved to another tier or deleted. ● Again for us the access pattern was not entirely, Infrequent.
- 14. ● By applying multiple transition and calculation strategies, we were able to reduce around 40% of storage cost per Month ● Current strategies are aligned with the business and operation flow, but they can change from time to time ● Schedule Periodic Optimisation Exercises - Not Too Early, Not Too Late Current State
- 15. ● One bucket Per Customer is not a scalable solution (not a practical solution for SaaS company) ● S3 has hard limit of 1000 buckets Per Account Per-Tenant Storage Breakdown Storage Bucket /tenant1/** /tenant2/** /tenantN/**
- 16. ● AWS provides tags to organize account resources, and cost allocation tags to track your AWS costs on a detailed level ● S3 allows to set Tags for Buckets as well as individual objects inside the bucket ● Cost Allocation Tags is the most efficient practice to create logical partitioning within the resources ● However, Tags has additional COST Cost Analysis for Cost Analysis
- 17. ● S3 allows to set Tags for individual objects ● For millions of objects settings multiple tags was not a desired solution ● While the cost is not huge, we needed to run additional automation on objects ● So we decided to build and store Object Metadata into CSV files and DynamoDB Cost Analysis for Cost Analysis $0.01 for 10,000 Tags And we have Millions of Objects
- 18. ● As large number of objects are located in the Intelligent Tier, using the LIST API is very cheap ● Metadata storage helps to build a Cost Breakdown Per Tenant with Per Storage Category ○ Count of Raw Files, Size and Cost ○ Count of Output Files, Size and Cost ● Helped to create granular cost reports like Per Month or Per Project Cost Analysis for Cost Analysis $0.01 for 10,000 Tags Per Month $0.0055 for 1000 LIST Object API Each API call can return Max 1000 Objects Entries
- 19. ● Listing Objects is not a frequent operation, only required when we need to re-build the Metadata ● This is a very tiny optimisation, but the output of Metadata helps in different analysis easily ● This only includes the Storage cost and there is no Data Transfer cost calculated
- 20. ● Understand the flow and state transition of your data ● Do the periodic cost and operational analysis ● At scale, small optimisations results in big savings ● There will always be an opportunity to optimise, in cost or operations ● S3 is by far one of the most reliable AWS services - It was designed for the ultimate Performance, Scale and Reliability. Key Learnings
- 21. Thank You Request to share feedback and join AWS User Groups