This chapter covers Amazon S3 Glacier retrieval options and their associated costs, a critical topic for the SAA-C03 exam's Cost Optimized domain. Understanding the trade-offs between retrieval speed and cost is essential for designing cost-effective archival storage solutions. Approximately 10-15% of exam questions touch on S3 storage classes and lifecycle policies, with a subset specifically testing Glacier retrieval options and cost implications. You will learn the exact retrieval times, pricing models, and use cases for Expedited, Standard, and Bulk retrievals, as well as how to optimize costs using provisioned capacity and free tier allowances.
Jump to a section
Imagine a massive library archive where books are stored in a deep underground vault. The library offers three retrieval options for archived books. First, Expedited Retrieval: you pay a premium fee, and a librarian immediately goes to the vault, retrieves the book, and brings it to the front desk within 1-5 minutes. This is for urgent needs. Second, Standard Retrieval: you submit a request, and the librarian processes it in a batch with other requests, retrieving your book within 3-5 hours. This is the default, cost-effective option. Third, Bulk Retrieval: you submit many requests at once, and the librarian schedules a single trip to the vault to retrieve all books together, taking 5-12 hours. This is the cheapest per book but slowest. Additionally, the library has a special 'free retrieval' offer: if you store over 10 petabytes of books, you get free bulk retrievals. The library also charges a 'retrieval fee' per book retrieved, with higher fees for faster options. If you retrieve too many books in a short time, the library may throttle your requests to manage workload. This mirrors Amazon S3 Glacier's retrieval options: Expedited (1-5 minutes), Standard (3-5 hours), and Bulk (5-12 hours), each with different costs and throughput limits. 'Free retrieval' corresponds to the 'free tier' for bulk retrievals when using Glacier Deep Archive with large volumes. Retrieval fees vary by speed, and throttling occurs via provisioned capacity units for expedited retrievals.
Amazon S3 Glacier is a storage class within Amazon S3 designed for long-term archival of data that is infrequently accessed and can tolerate retrieval delays. It offers extremely low storage costs (as low as $0.00099 per GB/month for Glacier Deep Archive) compared to S3 Standard (around $0.023 per GB/month). However, the trade-off is that data is not immediately accessible; you must initiate a retrieval request and wait for the data to be restored to a temporary staging area before you can access it. The retrieval time depends on the option you choose, and each option has a different cost structure.
Glacier is ideal for compliance archives, digital media preservation, scientific research data, and backup retention where retrieval is rare but must be possible when needed. The SAA-C03 exam tests your ability to select the right retrieval option based on business requirements (e.g., recovery time objective, cost constraints).
How Glacier Retrieval Works Internally
When you upload an object to S3 Glacier (or transition it via lifecycle policy), the object is stored in archival storage. The object is not directly accessible via standard S3 GET requests. To access the data, you must perform a RestoreObject operation (using the S3 API, CLI, or Console). This initiates a retrieval job that moves the data from the archival tier to a temporary staging location within S3 Glacier (not S3 Standard). Once the restore is complete, you can access the data using a GET request to the original object key, but the data is only available for a limited time (the restore duration you specify, typically 1-30 days). After that, the temporary copy is deleted, and the object returns to the archived state.
The retrieval process involves the following steps: 1. You submit a restore request specifying the retrieval tier (Expedited, Standard, or Bulk) and the number of days the restored copy should be available. 2. S3 Glacier queues the request based on the tier. Expedited requests are prioritized, Standard requests are processed in the order received, and Bulk requests are processed in batches. 3. The data is retrieved from archival storage and copied to the temporary staging area. The time to complete depends on the tier and the size of the object (for Expedited and Standard, up to 250 MB per request; for Bulk, up to 1 GB per request). 4. Once the restore is complete, the object is accessible for the specified duration. You can download it via S3 GET requests. 5. After the duration expires, the temporary copy is automatically deleted. You can also delete it manually by aborting the restore.
Key point: You cannot directly read from Glacier; you must always restore first. The restored object is billed for storage at the S3 Glacier storage class rate (not S3 Standard) during the restore duration.
Key Components, Values, Defaults, and Timers
Retrieval Options and Timings: - Expedited: 1-5 minutes for most objects (typically under 250 MB). For larger objects, retrieval time may increase. This is the fastest but most expensive option. - Standard: 3-5 hours for most objects. This is the default option and balances cost and speed. - Bulk: 5-12 hours for most objects. This is the cheapest option and is ideal for large-scale retrievals where time is not critical.
Costs: - Retrieval Fees (per GB retrieved): - Expedited: $0.03 per GB - Standard: $0.01 per GB - Bulk: $0.0025 per GB (for Glacier) or $0.00099 per GB (for Glacier Deep Archive) - Request Fees (per 1000 requests): - Expedited: $10 per 1000 requests - Standard: $0.05 per 1000 requests - Bulk: $0.025 per 1000 requests - Data Transfer Out: Standard S3 data transfer pricing applies (e.g., first 1 GB/month free, then tiered). - Storage Fees: During the restore period, you pay for the temporary copy at the Glacier storage rate (not S3 Standard).
Provisioned Capacity: - For Expedited retrievals, you can purchase provisioned capacity to ensure that your retrieval requests are processed without throttling. This is useful for business-critical data where you need guaranteed fast access. - Cost: $100 per provisioned capacity unit (PCU) per month. Each PCU ensures that your Expedited retrievals are processed within 1-5 minutes, even during peak demand. - Without provisioned capacity, Expedited retrievals are subject to available capacity and may be throttled.
Free Tier for Bulk Retrievals: - For Glacier Deep Archive, the first 10 PB of Bulk retrievals per month are free. This is a significant cost-saving feature for large-scale archives. - For Glacier, there is no free tier for Bulk retrievals (though costs are low).
Restore Duration: - You must specify the number of days the restored object should be accessible (1-30 days). The default is 1 day. - You can extend the restore duration by initiating a new restore request before the current one expires.
Throttling and Limits: - Expedited retrievals without provisioned capacity are subject to throttling. AWS may limit the number of concurrent Expedited retrievals based on overall system load. - Standard and Bulk retrievals have no throttling but are processed in a first-in-first-out (FIFO) order for Standard, and batch order for Bulk. - Maximum retrieval rate: For Expedited, up to 250 MB per request; for Standard, up to 250 MB; for Bulk, up to 1 GB per request. Larger objects are retrieved in parts.
Configuration and Verification Commands
Using AWS CLI to initiate a restore:
aws s3api restore-object --bucket my-bucket --key my-archive.zip --restore-request '{"Days":7,"GlacierJobParameters":{"Tier":"Expedited"}}'--restore-request is a JSON structure. Days specifies how long the restored copy is available. Tier can be Expedited, Standard, or Bulk.
Checking restore status:
aws s3api head-object --bucket my-bucket --key my-archive.zipLook for the Restore field in the output. It will show ongoing-request="true" while the restore is in progress, or ongoing-request="false" with expiry date when complete.
Using S3 Console: - Navigate to the object, select 'Initiate restore', choose the tier, and set the restore duration.
Using S3 Lifecycle Policies: - You can automate transitions to Glacier and set expiration actions. Lifecycle policies do not trigger retrievals; they only manage storage class transitions and deletions.
How Glacier Interacts with Related Technologies
S3 Lifecycle Policies: You can configure rules to automatically transition objects from S3 Standard to S3 Glacier (or Glacier Deep Archive) after a specified number of days. For example, transition to Glacier after 30 days, then to Glacier Deep Archive after 365 days. This reduces storage costs for aging data.
S3 Object Lock: Can be used with Glacier to enforce retention periods (WORM). Retrieval is still required to access locked objects.
AWS Backup: Can use Glacier as a backup destination. Restores from AWS Backup use the same retrieval options.
S3 Glacier Select: Allows you to run SQL queries on archived data without restoring the entire object. This can reduce retrieval costs by only retrieving the data you need. However, it is only available for Glacier (not Glacier Deep Archive) and supports CSV, JSON, and Parquet formats.
S3 Batch Operations: Can be used to perform bulk restores of many objects. You can create a batch job that initiates restores for a list of objects, using a single retrieval tier.
Exam Tips
Memorize the retrieval times: Expedited (1-5 min), Standard (3-5 hours), Bulk (5-12 hours).
Know the costs per GB: Expedited $0.03, Standard $0.01, Bulk $0.0025 (Glacier) or $0.00099 (Glacier Deep Archive).
Understand that Bulk retrievals from Glacier Deep Archive have a free tier of 10 PB per month.
Provisioned capacity is only for Expedited retrievals and costs $100 per unit per month.
Retrieval requests are subject to throttling for Expedited without provisioned capacity; Standard and Bulk are not throttled but are queued.
The restored object is billed at Glacier storage rates during the restore period, not S3 Standard.
You cannot directly read from Glacier; you must always restore first.
Initiate Restore Request
You submit a restore request via the S3 API, CLI, or Console. Specify the retrieval tier (Expedited, Standard, or Bulk) and the number of days the restored object should be accessible (1-30). The request is sent to the S3 Glacier service, which acknowledges it and returns a restore ID. At this point, the object remains in the archived state, and you cannot access it yet. The request is queued based on the tier: Expedited requests are placed in a high-priority queue, Standard in a normal queue, and Bulk in a batch queue. If you have provisioned capacity for Expedited, the request is guaranteed to be processed within 1-5 minutes; otherwise, it may be throttled if system load is high.
Queuing and Prioritization
S3 Glacier internally manages multiple queues for each tier. Expedited requests are processed as soon as possible, but without provisioned capacity, AWS may limit the number of concurrent Expedited retrievals to ensure fair usage. Standard requests are processed in FIFO order, meaning if you submit multiple Standard requests, they are handled in the order received. Bulk requests are collected and processed in batches; AWS may wait until a sufficient number of Bulk requests accumulate or a time threshold is reached before processing the batch. The exact batching algorithm is not documented, but typical completion times are 5-12 hours. During this step, no data movement occurs yet.
Data Retrieval from Archival Storage
When the request reaches the front of the queue, S3 Glacier begins reading the data from the archival storage media (which may be tape or disk-based, depending on the storage class). The data is read into memory and then written to a temporary staging area within the Glacier infrastructure. For Expedited and Standard, the retrieval is done in chunks of up to 250 MB per request; for Bulk, up to 1 GB per request. If the object is larger than these limits, it is retrieved in multiple parts, which may increase total retrieval time. The staging area is a temporary S3 bucket-like location that holds the restored data for the duration you specified. During this step, you cannot access the data yet.
Restore Completion Notification
Once the data has been fully copied to the staging area, S3 Glacier updates the object's metadata to indicate that the restore is complete. The `Restore` field in the object's head changes from `ongoing-request="true"` to `ongoing-request="false"` and includes an expiry date. You can now access the object using standard S3 GET requests. The data is available for the number of days you specified (e.g., 7 days). During this time, you can download the object as many times as you want. However, you are billed for the storage of the restored copy at the Glacier storage class rate (not S3 Standard). If you delete the object before the expiry, the temporary copy is removed immediately.
Expiration and Cleanup
After the restore duration expires, S3 Glacier automatically deletes the temporary copy from the staging area. The object reverts to its archived state, and you can no longer access it without initiating a new restore. The object's metadata returns to showing no active restore. Note that you can extend the restore duration by initiating a new restore request before the current one expires; this resets the expiry timer. However, you cannot permanently keep the object in a restored state; you must accept the cost of repeated restores if you need continuous access. If you want to avoid repeated restores, consider moving the object to a different storage class (e.g., S3 Standard) by copying it, but this incurs retrieval and storage costs.
Enterprise Scenario 1: Financial Institution Compliance Archive
A large bank must retain customer transaction records for 7 years to comply with regulatory requirements. The data is accessed only for audits, which occur about once per year. The bank chooses S3 Glacier Deep Archive for storage due to its low cost ($0.00099/GB/month). For annual audits, they need to retrieve approximately 500 TB of data. They use Bulk retrieval to minimize costs, as the audit can wait a day. The total retrieval cost for 500 TB is about $500 (at $0.00099/GB). They also benefit from the free tier of 10 PB per month for Bulk retrievals from Glacier Deep Archive, so the actual cost is zero. However, they must ensure that the retrieval completes within 12 hours to meet audit deadlines. In production, they automate the restore process using AWS Batch and S3 Batch Operations to handle the large number of objects. A common misconfiguration is forgetting to set the restore duration long enough; they set it to 30 days to allow ample time for the auditors to access the data. If they had chosen Standard retrieval, the cost would be $5,000, and Expedited would be $15,000, which would be unnecessarily expensive.
Enterprise Scenario 2: Media Company Digital Asset Archive
A media company stores raw video footage from past productions in S3 Glacier. Occasionally, an editor needs to access a specific clip to create a retrospective. The retrieval must happen quickly to meet tight production deadlines. The company uses Expedited retrieval with provisioned capacity to guarantee 1-5 minute retrieval times. They purchase 2 provisioned capacity units ($200/month) to handle up to 2 concurrent Expedited retrievals. Each retrieval of a 10 GB video costs $0.30 in retrieval fees ($0.03/GB). Without provisioned capacity, they risk throttling during peak hours, which could delay production. In production, they monitor retrieval times using CloudWatch metrics and adjust the number of provisioned capacity units based on usage patterns. A common pitfall is using Expedited without provisioned capacity for critical retrievals; during high demand, retrievals may take much longer than 5 minutes, causing missed deadlines.
Enterprise Scenario 3: Research Institution Scientific Data
A research institution stores petabytes of climate simulation data in S3 Glacier. Researchers frequently need to retrieve subsets of data for analysis. They use S3 Glacier Select to run SQL queries on the archived data without restoring entire objects, reducing retrieval costs. For example, they can query for temperature readings in a specific region from a large CSV file. This avoids the cost of restoring the entire object (which could be hundreds of GB) and only retrieves the matching rows. However, Glacier Select is not available for Glacier Deep Archive, so they store frequently queried data in standard Glacier. They also use Standard retrieval for larger restores when they need the full dataset. A common mistake is assuming Glacier Select works on all Glacier storage classes; it only works on standard Glacier, not Glacier Deep Archive. They also need to ensure the data is in a supported format (CSV, JSON, or Parquet).
SAA-C03 Exam Focus on Glacier Retrieval Options and Costs
The SAA-C03 exam tests this topic primarily under Objective 4.2: Determine cost-optimized storage solutions. You will be presented with scenarios that require selecting the appropriate storage class and retrieval option based on cost and access requirements. Key objective codes: COST 4.2.1 (Identify appropriate storage class for archival data) and COST 4.2.2 (Evaluate retrieval options and costs).
Common Wrong Answers and Why Candidates Choose Them
Choosing Expedited retrieval for all archival data – Candidates often assume that faster is always better, ignoring the significant cost. The exam expects you to match retrieval speed to business needs. If the scenario says 'data is accessed rarely and can be retrieved within 12 hours,' Bulk is the correct choice. Expedited would be an expensive overkill.
Selecting Standard retrieval for infrequent access with no time constraint – Standard retrieval costs $0.01/GB, while Bulk costs $0.0025/GB. Many candidates default to Standard because it's the default option, but Bulk is cheaper for large volumes. The exam will test if you know the exact cost difference.
Assuming provisioned capacity is needed for all Expedited retrievals – Provisioned capacity is only necessary if you need guaranteed 1-5 minute retrieval times. If the scenario doesn't mention a strict SLA, Expedited without provisioned capacity may suffice, though it could be throttled. Candidates often over-provision.
Forgetting that Bulk retrievals from Glacier Deep Archive have a free tier – The free tier of 10 PB per month is a major cost-saver. Many candidates miss this and calculate higher costs than necessary.
Specific Numbers and Values That Appear on the Exam
Retrieval times: Expedited (1-5 min), Standard (3-5 hours), Bulk (5-12 hours).
Retrieval costs per GB: Expedited $0.03, Standard $0.01, Bulk $0.0025 (Glacier) or $0.00099 (Glacier Deep Archive).
Request costs per 1000 requests: Expedited $10, Standard $0.05, Bulk $0.025.
Provisioned capacity: $100 per unit per month (only for Expedited).
Free tier: 10 PB per month of Bulk retrievals from Glacier Deep Archive.
Restore duration: 1-30 days (default 1 day).
Edge Cases and Exceptions
Expedited retrieval for objects >250 MB – May take longer than 5 minutes because the object is retrieved in parts. The exam may test that Expedited is best for small objects.
Glacier Select – Only works on standard Glacier, not Glacier Deep Archive. Also, it supports only CSV, JSON, and Parquet.
Throttling of Expedited without provisioned capacity – AWS may throttle Expedited requests if there is high demand. The exam might present a scenario where the company needs fast retrieval but has no provisioned capacity, leading to delays.
Restore duration billing – You pay storage costs for the restored copy at the Glacier storage class rate, not S3 Standard. Some candidates mistakenly think the restored copy is free or charged at S3 Standard rates.
How to Eliminate Wrong Answers
If the scenario mentions 'retrieval within minutes' and 'cost is not a concern,' choose Expedited. If 'cost is a concern,' consider Standard or Bulk.
If the scenario mentions 'retrieval within 12 hours' and 'large volume,' choose Bulk (especially for Glacier Deep Archive with free tier).
If the scenario mentions 'guaranteed retrieval time of 1-5 minutes,' you must select provisioned capacity with Expedited.
If the scenario involves querying data without restoring the entire object, consider Glacier Select (but only for Glacier, not Glacier Deep Archive).
Always check if the data is in Glacier or Glacier Deep Archive, as costs and free tier differ.
Expedited retrieval: 1-5 minutes, $0.03/GB, requires provisioned capacity for guaranteed speed.
Standard retrieval: 3-5 hours, $0.01/GB, default option.
Bulk retrieval: 5-12 hours, $0.0025/GB (Glacier) or $0.00099/GB (Glacier Deep Archive), free tier of 10 PB/month for Glacier Deep Archive.
Provisioned capacity: $100 per unit per month, only for Expedited retrievals.
Glacier Deep Archive does not support Expedited retrieval.
You must restore objects before accessing them; restored copy is billed at Glacier storage rates during the restore period.
Glacier Select allows querying archived data without full restore (only for standard Glacier, not Deep Archive).
Throttling can occur for Expedited without provisioned capacity; Standard and Bulk are not throttled but queued.
These come up on the exam all the time. Here's how to tell them apart.
Expedited Retrieval
Retrieval time: 1-5 minutes
Cost: $0.03 per GB retrieved
Request cost: $10 per 1000 requests
Requires provisioned capacity for guaranteed speed
Not available for Glacier Deep Archive
Standard Retrieval
Retrieval time: 3-5 hours
Cost: $0.01 per GB retrieved
Request cost: $0.05 per 1000 requests
No provisioned capacity needed; FIFO queue
Available for both Glacier and Glacier Deep Archive
Standard Retrieval
Retrieval time: 3-5 hours
Cost: $0.01 per GB retrieved
Request cost: $0.05 per 1000 requests
FIFO processing order
No free tier
Bulk Retrieval
Retrieval time: 5-12 hours
Cost: $0.0025 per GB (Glacier) or $0.00099 per GB (Glacier Deep Archive)
Request cost: $0.025 per 1000 requests
Batch processing; may be delayed
Free tier: 10 PB per month for Glacier Deep Archive
Mistake
You can directly read data from S3 Glacier using GET requests without restoring.
Correct
No. You must initiate a RestoreObject request first. Once the restore is complete, you can use GET requests to access the temporary copy. The object remains in the archived state until restored.
Mistake
Standard retrieval always completes in 3-5 hours for any size object.
Correct
The 3-5 hour estimate applies to most objects, but larger objects may take longer because they are retrieved in multiple parts. For very large objects (e.g., >10 GB), retrieval time can exceed 5 hours.
Mistake
Provisioned capacity guarantees Expedited retrieval for all objects.
Correct
Provisioned capacity ensures that your Expedited requests are processed within 1-5 minutes, but it does not guarantee that the retrieval itself completes within that time for very large objects. Also, it only covers Expedited retrievals, not Standard or Bulk.
Mistake
Bulk retrieval is always the cheapest option.
Correct
Bulk retrieval has the lowest per-GB cost, but it also has a per-request cost. For very small objects (e.g., a few KB), the per-request cost may dominate, making Standard retrieval cheaper. However, for most cases, Bulk is the cheapest.
Mistake
Glacier Deep Archive supports Expedited retrieval.
Correct
No. Glacier Deep Archive only supports Standard (3-5 hours) and Bulk (5-12 hours) retrievals. Expedited is not available for Glacier Deep Archive. This is a common exam trap.
Reveal each answer, then mark whether you got it right. Score 60%+ to unlock the next chapter.
There are three retrieval options: Expedited (1-5 minutes), Standard (3-5 hours), and Bulk (5-12 hours). Each has different costs and use cases. Expedited is the fastest and most expensive, Bulk is the slowest and cheapest. Glacier Deep Archive only supports Standard and Bulk.
Costs include retrieval fees per GB and per request fees. For Expedited: $0.03/GB + $10 per 1000 requests. Standard: $0.01/GB + $0.05 per 1000 requests. Bulk: $0.0025/GB (Glacier) or $0.00099/GB (Glacier Deep Archive) + $0.025 per 1000 requests. Data transfer out also applies.
Provisioned capacity ensures that your Expedited retrieval requests are processed within 1-5 minutes, even during high demand. Each unit costs $100 per month. Without it, Expedited requests may be throttled. It is only available for Expedited retrievals.
No. Glacier Select is only available for standard Glacier, not Glacier Deep Archive. It allows you to query data using SQL without restoring the entire object, but only for CSV, JSON, and Parquet formats.
You can specify a restore duration between 1 and 30 days. After that, the temporary copy is automatically deleted. You can extend the duration by initiating a new restore request before the current one expires.
The first 10 PB of Bulk retrievals from Glacier Deep Archive per month are free. This applies to the retrieval fee per GB, but request fees and data transfer out still apply.
Use the command: aws s3api restore-object --bucket my-bucket --key my-object --restore-request '{"Days":7,"GlacierJobParameters":{"Tier":"Standard"}}'. Replace the tier and days as needed.
You've just covered Glacier Retrieval Options and Costs — now see how well it sticks with free SAA-C03 practice questions. Full explanations included, no account needed.
Done with this chapter?