I'm using the external storage for storing events in a database while they are waiting to be sent to the server.
I'm seeing really bad performance when inserting records. I know the external memory can be slow but I wanted to see some number so I wrote a small app which tests it.
Here is the code:
public static final int INSERTS = 100;
File dbFile = new File(Environment.getExternalStorageDirectory(), "test.sqlite3");
// File dbFile = new File(getFilesDir(), "test.sqlite3");
dbFile.delete();
SQLiteDatabase db = SQLiteDatabase.openOrCreateDatabase(dbFile, null);
db.execSQL("CREATE TABLE events (_id integer primary key autoincrement, event_type TEXT NOT NULL, timestamp BIGINT, data TEXT);");
db.execSQL("CREATE INDEX mainIndex ON events (event_type, timestamp ASC);");
InsertHelper helper = new InsertHelper(db, "events");
final int eventTypeCol = helper.getColumnIndex("event_type");
final int timestampCol = helper.getColumnIndex("timestamp");
final int dataCol = helper.getColumnIndex("data");
long start = System.currentTimeMillis();
String eventType = "foo", data = "bar";
long timestamp = 4711;
for(int i = 0; i < INSERTS; ++i) {
helper.prepareForInsert();
helper.bind(eventTypeCol, eventType);
helper.bind(timestampCol, timestamp);
helper.bind(dataCol, data);
helper.execute();
}
long end = System.currentTimeMillis();
Log.i("Test", String.format("InsertHelper, Speed: %d ms, Records per second: %.2f", (int)(end-start), 1000*(double)INSERTS/(double)(end-start)));
db.close();
dbFile.delete();
db = SQLiteDatabase.openOrCreateDatabase(dbFile, null);
db.execSQL("CREATE TABLE events (_id integer primary key autoincrement, event_type TEXT NOT NULL, timestamp BIGINT, data TEXT);");
db.execSQL("CREATE INDEX mainIndex ON events (event_type, timestamp ASC);");
start = System.currentTimeMillis();
ContentValues cv = new ContentValues();
for(int i = 0; i < INSERTS; ++i) {
cv.put("event_type", eventType);
cv.put("timestamp", timestamp);
cv.put("data", data);
db.insert("events", null, cv);
}
end = System.currentTimeMillis();
Log.i("Test", String.format("Normal, Speed: %d ms, Records per second: %.2f", end-start, 1000*(double)INSERTS/(double)(end-start)));
db.close();
dbFile.delete();
The database is exactly as the one my real app is using, I tried removing the index but it made no difference.
Here are the results:
Nexus One, Internal memory Method | Records | Time (ms) | Records per second -------------+---------+-----------+-------------------- Normal | 100 | 2072 | 48.26 InsertHelper | 100 | 1662 | 60.17 Nexus One, External memory: Method | Records | Time (ms) | Records per second -------------+---------+-----------+-------------------- Normal | 100 | 7390 | 13.53 InsertHelper | 100 | 7152 | 13.98 Emulator, Internal memory: Method | Records | Time (ms) | Records per second -------------+---------+-----------+-------------------- Normal | 100 | 1803 | 55.46 InsertHelper | 100 | 3075 | 32.52 Emulator, External memory: Method | Records | Time (ms) | Records per second -------------+---------+-----------+-------------------- Normal | 100 | 5742 | 17.42 InsertHelper | 100 | 7164 | 13.96
As you can see the emulator cannot be trusted, InsertHelper
should be faster if anything.
This is, of course, to be expected, the test was mostly done out of curiosity.
What have me concerned however is the bad performance on my phone when using external memory, have I missed some crucial aspect of SQLiteDatabase
or is it simply so that the SD card will be slow?
I can add that in my real app I've disabled locking and it makes little difference.
CommonsWare is correct in his comment. Something that makes a big difference for db performance is using transactions. Wrap your insert loop in a transaction. I'm not 100% sure if it would work with the InsertHelper but you can try replacing your for loop with this:
db.beginTransaction();
try {
for(int i = 0; i < INSERTS; ++i) {
helper.prepareForInsert();
helper.bind(eventTypeCol, eventType);
helper.bind(timestampCol, timestamp);
helper.bind(dataCol, data);
helper.execute();
}
db.setTransactionSuccessful();
} finally {
db.endTransaction();
}
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With