W3cubDocs

/SQLite

Internal Versus External BLOBs in SQLite

If you have a database of large BLOBs, do you get better read performance when you store the complete BLOB content directly in the database or is it faster to store each BLOB in a separate file and store just the corresponding filename in the database?

To try to answer this, we ran 49 test cases with various BLOB sizes and SQLite page sizes on a Linux workstation (Ubuntu circa 2011 with the Ext4 filesystem on a fast SATA disk). For each test case, a database was created that contains 100MB of BLOB content. The sizes of the BLOBs ranged from 10KB to 1MB. The number of BLOBs varied in order to keep the total BLOB content at about 100MB. (Hence, 100 BLOBs for the 1MB size and 10000 BLOBs for the 10K size and so forth.) SQLite version 3.7.8 (2011-09-19) was used.

Update: New measurements for SQLite version 3.19.0 (2017-05-22) show that SQLite is about 35% faster than direct disk I/O for both reads and writes of 10KB blobs.

The matrix below shows the time needed to read BLOBs stored in separate files divided by the time needed to read BLOBs stored entirely in the database. Hence, for numbers larger than 1.0, it is faster to store the BLOBs directly in the database. For numbers smaller than 1.0, it is faster to store the BLOBs in separate files.

In every case, the pager cache size was adjusted to keep the amount of cache memory at about 2MB. For example, a 2000 page cache was used for 1024 byte pages and a 31 page cache was used for 65536 byte pages. The BLOB values were read in a random order.

Database Page Size BLOB size
10k 20k 50k 100k 200k 500k 1m
1024 1.535 1.020 0.608 0.456 0.330 0.247 0.233
2048 2.004 1.437 0.870 0.636 0.483 0.372 0.340
4096 2.261 1.886 1.173 0.890 0.701 0.526 0.487
8192 2.240 1.866 1.334 1.035 0.830 0.625 0.720
16384 2.439 1.757 1.292 1.023 0.829 0.820 0.598
32768 1.878 1.843 1.296 0.981 0.976 0.675 0.613
65536 1.256 1.255 1.339 0.983 0.769 0.687 0.609

We deduce the following rules of thumb from the matrix above:

  • A database page size of 8192 or 16384 gives the best performance for large BLOB I/O.

  • For BLOBs smaller than 100KB, reads are faster when the BLOBs are stored directly in the database file. For BLOBs larger than 100KB, reads from a separate file are faster.

Of course, your mileage may vary depending on hardware, filesystem, and operating system. Double-check these figures on target hardware before committing to a particular design.

SQLite is in the Public Domain.
https://sqlite.org/intern-v-extern-blob.html