Maaz Zindani

Loading the index…

↑↓ move↵ openesc close4072 pages

Cloud and platforms

DynamoDB vs Cosmos DB vs Firestore: How Each Meters Reads and Writes

In short

The three meter different things. DynamoDB meters item size, in 4 KB units for reads and 1 KB units for writes. Cosmos DB meters work in request units, where size, indexing and some consistency levels change the charge. Firestore Standard counts documents and index entries read, and its pricing page names no size tranche.

What counts as one read on each service?

DynamoDB counts in size tranches. A strongly consistent read of an item up to 4 KB takes one read unit, an eventually consistent read takes half a unit, and a transactional read takes two. A 10 KB item is rounded up to 12 KB before the units are worked out.

Where the rounding happens depends on the call. A Query adds up every returned item and rounds once, so ten items totalling 40.8 KB count as 44 KB. BatchGetItem rounds each item first, so items of 1.5 KB and 6.5 KB count as 12 KB, not 8 KB. A Scan is metered on the items it evaluates, not the items it returns, and reading an item that does not exist still consumes throughput.

Cosmos DB counts in request units, a blended measure of CPU, IOPS and memory. A point read of a 1 KB item costs 1 RU and a 100 KB item costs 10 RU. A query has no fixed price. Predicates, results and source data size all feed in, although the same query on the same data always costs the same.

Firestore Standard counts documents. Each document a query returns is one read, a query costs at least one read even when it returns nothing, and an offset bills every skipped document. Index entries are billed as well, at one read for each batch of up to 1,000 entries, except that queries with up to one range field are not charged for them.

Checked against: source 1, source 5, source 6, source 10

How does each service meter a write?

In DynamoDB one write unit covers an item up to 1 KB, rounded up, so a 1.6 KB item counts as 2 KB. UpdateItem is charged on the larger of the item sizes before and after, even if you change one attribute. A conditional write whose condition fails still consumes write capacity. A transactional write takes two units per tranche because DynamoDB does a prepare write and a commit write, and a cancelled transaction still consumes that capacity.

Cosmos DB bases a write on item size and on how many properties the indexing policy covers. The optimize-cost page says inserting a 1 KB item without indexing costs around 5.5 RU and that replacing an item costs twice the insert. A 1 KB read is 1 RU, so the read and write scales are not the same size. I would treat 5.5 as a documented estimate and read the request charge on a real response.

Firestore Standard charges for each document write and delete, and a set or an update counts as one write. The Enterprise edition of the same product meters differently: write units in 1 KiB tranches, with an update charged on the larger of the before and after sizes plus the index entries it touches.

Checked against: source 1, source 6, source 10, source 11

How do size, consistency and indexes change the amount?

The table lists only what each vendor page says. Where a cell says the pages are silent, that is not a claim that the effect is zero.

DynamoDB indexes deserve a second look, which the table has no room for. When a write changes an indexed key attribute, the index needs two writes, one to delete the old entry and one to put the new one. On a provisioned table, an index without enough write capacity for the table writes throttles those writes, and AWS advises giving the index at least the write capacity of the base table.

FactorDynamoDBCosmos DBFirestore Standard
Item sizeRounded up per item to 4 KB (read) or 1 KB (write)RU rises with size: 1 KB read is 1 RU, 100 KB read is 10 RUCounted per document, no size tranche on the pricing page; Enterprise edition uses 4 KiB and 1 KiB tranches
ConsistencyEventual halves a read, transactional doubles itStrong and bounded staleness roughly double read costNot a billing factor on the pages I read
Indexes on writeEach affected global secondary index consumes its own write unitsEvery property is indexed by default; indexing fewer lowers write RUNot a write meter; index entries are billed when read and when stored
Hard ceiling per documentNot on the pages I readNot on the pages I read1 MiB document, 40,000 index entries

Checked against: source 1, source 4, source 5, source 6, source 10, source 11, source 12

How do provisioned and on-demand modes differ?

DynamoDB sets the mode per table. On-demand bills per read and write request unit and charges nothing for throughput at zero traffic. Provisioned bills hourly for the capacity you set, whether or not you use it, and auto scaling can adjust that figure. You can move from provisioned to on-demand up to four times in a rolling 24 hours, and back at any time.

On-demand has limits of its own. It instantly serves up to double the previous peak, and a jump beyond that inside 30 minutes can throttle. The default table-level throughput quota is 40,000 units.

Cosmos DB has three modes. Manual provisioned throughput is set in steps of 100 RU/s and billed hourly on what you provision. Autoscale takes a maximum, scales between 10 percent and 100 percent of it, and bills each hour for the highest RU/s reached in that hour. Serverless bills the RUs consumed. Serverless is an account type, runs in one region, and refuses any throughput setting.

The Firestore Standard pages I read describe per-operation billing only, with a daily free quota of 50,000 reads, 20,000 writes and 20,000 deletes. I found no provisioned-throughput setting there.

Checked against: source 2, source 3, source 7, source 8, source 9, source 10

What should you check before comparing two bills?

Work through these in order, because each follows from a difference above.

  • Typical item size against the tranche: 4 KB and 1 KB on DynamoDB, a 1 KB reference on Cosmos DB, none on Firestore Standard.
  • Read path: point read, query or scan, because each service meters them differently.
  • Index count: global secondary indexes, indexed properties or composite indexes.
  • Consistency: eventual or strong on DynamoDB, the account level on Cosmos DB.
  • Listeners on Firestore: each added or updated document in the result set is billed as a read.
  • Mode or edition: on-demand or provisioned, serverless or autoscale, Standard or Enterprise.

Checked against: source 1, source 5, source 10

What else do people ask?

Is a DynamoDB read unit the same as a Cosmos DB request unit?

No. A DynamoDB read unit counts 4 KB tranches of one item. A Cosmos DB request unit blends CPU, IOPS and memory, with a 1 KB point read as the reference. Converting between them needs item size, indexing and the read path.

Does Firestore have capacity units?

Standard edition counts document reads, writes and deletes. Enterprise edition uses read units in 4 KiB tranches and write units in 1 KiB tranches. It is one product with two meters.

Sources

How was this page made?

I drafted this page with AI assistance from the official documents listed under Sources. This page compares vendor documents that I fetched on 19 Nov 2026. I did not run DynamoDB, Cosmos DB or Firestore workloads for it. It does not cover prices, storage and backup meters, multi-region or global tables, or Cosmos DB APIs other than NoSQL. Found a mistake? Tell me.

Keep reading

Next step

Have a project in mind?

Tell me about it. A discovery call within 48 hours.