With greenDAO 3 released, it’s time again to look at the Android ORM performance landscape and do some benchmarks. This time, we also tested newer ORMs for SQLite like DbFlow, requery, SQLDelight, and SquiDB. Also, we had an extensive look at benchmarks done by others. Let’s get started with our results:
At Droidcon Berlin, we noticed a lot of questions around databases and that many people weren’t aware of SQLite alternatives and Object-Relational Mappers (ORMs). Therefore, we follow up with an overview of the mobile database landscape.
What is a mobile database?
While Wikipedia defines a mobile database as “either a stationary database that can be connected to by a mobile computing device […] over a mobile network, or a database which is actually stored by the mobile device,” we solely refer to databases that run on the mobile device itself.
Why use a mobile database?
There are some advantages associated with using a mobile database:
I don’t believe in any statistics I didn’t fake myself (German saying)
Doing good performance benchmarks is hard and benchmarking on Android is no exception. Here is a checklist on how to do fair and robust benchmarks:
We are happy to announce greenDAO Version 3 today. Since its initial release 5 years ago, greenDAO has always been the fastest ORM for Android. It was also the first ORM to apply code generation for maximum performance. With greenDAO 3, we made code generation much more convenient: by adopting entity annotations you can drop the generator project. In its place comes the all new Gradle plugin, which automatically triggers code generation at build time. Like that, you can simply use greenDAO 3 annotations on your entities:
That's really good news. Just by using the new version 3.5.0 of SQLCipher, you can reduce your APK size (and build/deployment time). Not just the binary .so files got smaller, also the big "icudt46l.dat" file was removed in 3.5.0.
To see the effect in an app, we took the greenDAO example app and compared the resulting sizes. Of course, the version using built-in SQLite is by far the smallest, but your APK may lose 4.4 MB if you were using a previous version of SQLCipher:
We noticed one incompatibility with SQLCipher 3.5.x however. Android's SQLiteDatabase defines two additional collations:
In addition to SQLite's default BINARY collator, Android supplies two more, LOCALIZED, which changes with the system's current locale, and UNICODE, which is the Unicode Collation Algorithm and not tailored to the current locale.
Running SQL ORDER commands with "COLLATE LOCALIZED" does not work anymore starting with SQLCipher 3.5.0. That is the downside of the ICU (and ASOP code) removal.
So, what about greenDAO support for SQLCipher 3.5.x? greenDAO used "COLLATE LOCALIZED" in the QueryBuilder when specifying orders using string properties. To ensure compatibility with SQLCipher 3.5, we just released greendao-encryption V2.2.2 without the LOCALIZED collation.
We just released greenDAO 3 beta: it makes the generator project optional and moves to Java annotations. Before the final release however, we want to get your feedback. In particular, we experimented with alternative annotation processing that gives greenDAO more power while it avoids byte code manipulation.Our goal with greenDAO3 is to put the developer in control over entity classes while augmenting the entity code with a little bit of generated code.
At yesterday’s Droidcon Berlin barcamp, we compared different database approaches (SQLite, ORMs, NoSQL). In our interactive session we asked about 80 developers how they use databases on Android and what advantages and disadvantages they see in each approach. Here are the gathered results:
~40% of the participants have used SQLite without additional tools.
Encryption is the central feature of today’s greenDAO 2.2 release. Actually, greenDAO seems to be the first Android ORM to officially support SQLCipher (non-beta). Since it’s first (source-only) release in August 2015, greenDAO’s encryption support has been successfully tested in a complex setup for months. So, if you want to store sensitive data in your database, give it a try!
There is an obscure scenario causing some older Android versions to throw java.lang.NoClassDefFoundError when trying to register subscribers in EventBus. It was reported often in connection to the class PersistableBundle, which was introduced in Android API level 21. It seems like an Android bug with reflection, but of course you don’t want your app to crash. Thus, we just added an FAQ entry with covering some background and a couple of solutions. One preferred solution is to update to EventBus 3 along with the subscriber index. Because the index is created during build time, it avoids problematic (and slow) reflection altogether!
Today, we are pleased to release EventBus 3. Previous versions of EventBus rejected to use annotations for performance reasons. Now, version 3 embraces annotations to improve performance. How does that make sense? Unfortunately, Android’s performance did not get any better. Reflection on annotations is still very bad compared to Desktop Java, even on Android 6 and ART. EventBus 3 resolves this problem by introducing a new annotation processor, which will index all annotations at build time. It generates a class containing all the data that would be expensive to get at run time. Build-time indexing makes EventBus 3 the fastest EventBus ever.