Local First Database Comparison

In this project I have implemented the exact same chat application with different local first database technologies. You can use it to compare metrics and learn about the differences. The chat app is a web based angular application, with functionality similar to Whatsapp Web.

Implemented Databases:

Metrics

All metrics are measured automatically via code in a browser tests (chrome:headless). The results heavily depend on the developers device. You should compare the values relative to another and not as absolute values. Also you might want to create new metrics that better represent how you would use the respective database.

You can reproduce these values by running sh measure-metrics.sh in the root folder.

| Metric \ Project | rxdb-localstorage | rxdb-dexie | pouchdb | watermelondb | aws | firebase | pglite | sqlite-wasm | | --------------------------------------- | ----------------- | ---------- | ------- | ------------ | ------ | -------- | ------- | ----------- | | Bundle size, plain JavaScript | 988kb | 1088kb | 911kb | 1055kb | 1954kb | 1083kb | 1393kb | 969kb | | Bundle size (js+wasm), minified+gzip | 247kb | 280kb | 223kb | 245kb | 455kb | 272kb | 6044kb | 641kb | | First angular component render | 136ms | 129ms | 131ms | 138ms | 129ms | 162ms | 135ms | 131ms | | Page load time | 195ms | 171ms | 166ms | 176ms | 164ms | 203ms | 173ms | 164ms | | First full render | 179ms | 301ms | 597ms | 192ms | 257ms | 595ms | 1559ms | 268ms | | Insert one message | 6ms | 16ms | 16ms | 4ms | 17ms | 229ms | 136ms | 12ms | | Inserting 20 messages one after another | 128ms | 156ms | 113ms | 94ms | 307ms | 3949ms | 4986ms | 161ms | | Inserting 20 messages in parallel | 27ms | 130ms | 46ms | 82ms | 65ms | 2520ms | 2389ms | 53ms | | Message insert to message list change | 4ms | 15ms | 104ms | 3ms | 33ms | 19ms | 286ms | 14ms | | Message search query time | 7ms | 30ms | 56ms | 6ms | 326ms | 170ms | 147ms | 5ms | | First full render with many messages | 185ms | 393ms | 980ms | 166ms | 278ms | 690ms | 1577ms | 213ms | | Storage usage | 727kb | 1344kb | 1569kb | 1256kb | 105kb | 884kb | 15226kb | 5286kb |

Metrics Explanation

• Bundle size, plain JavaScript: The full JavaScript bundle size (including WebWorker code), without minification or gzip.
• Bundle size (js+wasm), minified+gzip: The full JavaScript bundle size (including WebWorker code) plus the size of the WebAssembly files and PGlite's data bundle, after minification and gzip compression.
• Page load time: How long does it take to download and parse the JavaScript bundle.
• First angular component render: How long does it take for the first angular component to be rendered.
• First full render: How long does it take until all relevant data is displayed for the first time.
• Insert one message: How long does it take to insert a single message.
• Inserting 20 messages one after another: How long does it take to insert 20 messages in serial.
• Inserting 20 messages in parallel: How long does it take to insert 20 messages in parallel.
• Message insert to message list change: How long does it take until a new message is rendered to the dom.
• User change to message list change: How long does it take from changing the user to the displaying of the new messages list.
• Message search query time: How long does it take to search for a given message by regex/like-operator.
• First full render with many messages: Time to first full render when many messages exist.
• Storage usage: Size of the stored data (IndexedDB, OPFS and localStorage) after inserting the full test dataset.

Investigations

Why is LokiJS so much faster?

WatermelonDB uses the LokiJS database as storage, which is an in memory database that regularly persists the data to IndexedDB either on interval, or when the browser tab is closed. By doing so, less slow IndexedDB transaction are used. Keeping and processing the data in memory has the benefit of being much faster, but it also has its downsides:

• Initial page load is much slower when much data is already stored in the database, because all data must be loaded before any database operation can be done.
• All data must fit into memory.
• Data can be lost when the JavaScript process is killed ungracefully like when the browser crashes or the power of the PC is terminated.
• There is no multi-tab-support with plain LokiJS. The data is not shared between multiple browser tabs of the same origin.

Why is Firebase so slow on first render?

On the first page load, Firebase ensures that the local data is equal to the server side state. This means that the client has to be online at application startup which is the reason why Firebase is not completely offline first. To ensure the equalness of client side data, Firebase has to perform several requests to the backend, before the database will respond to queries. This makes the inital page load slow, and it becomes even more slower, the more data exists and has to be validated.

Why is PouchDB so slow?

For the PouchDB and RxDB (with PouchDB storage) I used the old Indexeddb adapter. It is much less optimized than the new adapter, but the new one made problems with returning the correct query results. Theses problems have been fixed on the PouchDB master branch, but I have to wait for the next PouchDB release. I will update the repo when this change can be done.

Why does AWS Datastore need so much less storage space?

AWS Datastore does not save any metadata together with the documents. Instead only the plain documents are stored in IndexedDB. They can do this because they only allow simple queries and do not keep a local version history.

Why is the PGlite bundle so big?

PGlite is not a JavaScript database, it is the real PostgreSQL server compiled to WebAssembly. Before the first query can run, the browser has to download and compile the whole Postgres runtime:

• pglite.wasm (~10MB, ~3.4MB gzipped): the PostgreSQL server itself.
• pglite.data (~6.3MB, ~1.9MB gzipped): a packed file system that Postgres needs at runtime. It contains the loadable modules (mostly character set converters and the snowball full-text-search stemmer, ~3.1MB), the system catalog bootstrap file postgres.bki (~950kb), ICU collation data (~920kb) and the SQL setup scripts, timezone and full-text-search files.
• initdb.wasm (~400kb): used to create a new database on the first start.

Feature Map

| Feature \ Project | aws | firebase | pouchdb | rxdb-localstorage | rxdb-dexie | watermelondb | pglite | sqlite-wasm | | ------------------------ | ------------------ | -------------------------------------------- | ------- | ----------------- | ---------- | -------------------------------------- | ------------------------------------------ | ----------- | | Offline First | No, login required | Partially, must be online on first page load | Yes | Yes | Yes | Yes | Yes | Yes | | Realtime Replication | Yes | Yes | Yes | Yes | Yes | Partially, must be implemented by hand | Partially, read-path sync with ElectricSQL | No | | Multi Tab Support | Yes | Yes | No | Yes | Yes | Partially, relies on online sync | Partially, requires PGliteWorker | No | | Observable Queries | No | Yes | No | Yes | Yes | Yes | Yes | No | | Complex Queries | No | Yes | Yes | Yes | Yes | Yes | Yes | Yes | | Client Side Encryption | No | No | Yes | Yes | Yes | No | No | No | | Schema Support | Yes | No | No | Yes | Yes | Yes | Yes | Yes | | Custom Backend | No | No | No | Yes | Yes | Yes | No | No | | Custom Conflict Handling | Yes | No | Yes | Yes | Yes | No | No | No |

Starting the projects

All sub-projects use the same port and so can not be started in parallel.

Installation

• You must have installed Node.js
• Clone this project
• In the root folder, run npm install to install the dependencies.
• In the root folder, run npm run build to build all projects.

Firebase Firestore

• Run npm run start:firebase to start the mock server and the production build frontend.
• Or run npm run dev:firebase to start the mock server and the development frontend server.

• Open http://localhost:3000/ to browse the frontend.

AWS Amplify & Datastore

The official AWS mock does not allow a live replication at this point. So you first have to setup an amplify project in the ./projects/aws folder by using this tutorial

• Run npm run start:aws to start the mock server and the production build frontend.
• Or run npm run dev:aws to start the mock server and the development frontend server.

• Open http://localhost:3000/ to browse the frontend.

PouchDB

• Run npm run start:pouchdb to start the mock server and the production build frontend.
• Or run npm run dev:pouchdb to start the mock server and the development frontend server.

• Open http://localhost:3000/ to browse the frontend.

RxDB LocalStorage

• Run npm run start:rxdb-localstorage to start the mock server and the production build frontend.
• Or run npm run dev:rxdb-localstorage to start the mock server and the development frontend server.

• Open http://localhost:3000/ to browse the frontend.

RxDB Dexie.js

• Run npm run start:rxdb-dexie to start the mock server and the production build frontend.
• Or run npm run dev:rxdb-dexie to start the mock server and the development frontend server.

• Open http://localhost:3000/ to browse the frontend.

WatermelonDB

• Run npm run start:watermelondb to start the mock server and the production build frontend.
• Or run npm run dev:watermelondb to start the mock server and the development frontend server.

• Open http://localhost:3000/ to browse the frontend.

PGlite

• Run npm run start:pglite to start the production build frontend.
• Or run npm run dev:pglite to start the development frontend server.

• Open http://localhost:3000/ to browse the frontend.

SQLite Wasm

• Run npm run start:sqlite-wasm to start the production build frontend.
• Or run npm run dev:sqlite-wasm to start the development frontend server.

• Open http://localhost:3000/ to browse the frontend.