InfinityDB

Compare Java Edition with InfinityDB

The trial download now includes testers that enable direct comparisons between two embeddable database engines: Sleepy Cat Java Edition and InfinityDB. Using compressible data, the tests show that InfinityDB is 13 to 22 times faster and from 37 to 100 times more space efficient because of its unique dynamic data compression. Our clients typically subject the Infinity Database Engine B*Tree to performance, stress, and scalability tests and prove for themselves that it outperforms the Java edition. They choose InfinityDB for performance, incorruptibility, compression, and simplicity, whether it is to be embedded in server applications; throughout a distributed system for storage of configuration data; or within a PDA application.

We appreciate our customers who do this work for themselves, but now we offer our own testers, to speed the initial evaluation. These tests support our claims, while revealing how to code with our API. Access the download link from our home page.

InfinityDB is cost effective whether you are comparing it with the Java Edition or to rolling your own in-house data store. Its simple API and single file architecture simplifies the design and implementation phase; its incorruptibility eliminates support issues; its high performance enables the most powerful server applications; its extreme compressibility saves hardware; and its small footprint makes it embeddable in PDA's.

We are happy to recommend the best strategy for employing our flexible data model for your particular circumstances -- we even provide up to three free hours of support.

Check out our white papers and the other blog entries.

We look forward to hearing from you.
Jennifer Douglas
Licensing and Support Director

Performance with Compressible Data Better than Expected

We recently added a performance tester to the examples directory of the trial and deployment downloads: InfinityDBPerformanceTest.java. The results were pleasantly surprising. When tested against a common form of compressible data, speed and disk space efficiency for all operations improved dramatically. We have been quoting 30K inserts/sec/GHz, but with this rather typical compressible data,we are getting 120K insertions/sec/GHz, in a warm cache (with no disk I/O, i.e. all data in the cache). The cold cache performance (which requires allocating cache blocks during warmup) is only slightly slower.

The performance for pure random data is roughly what we expected, with next() being a bit slower than expected, and insert()/delete() somewhat faster.

Data compression was also dramatic: 20MB of compressible data was stored in 1.4MB. Data compression for purely random data is only 28% worse than the theoretical minimum, with the 20MB stored in 25637K bytes. A recent third-party potential client found that the database generated by the SleepyCat (Berkely DB) B-Tree took 140MB for similar purely random data.

This compressible data uses repeating 10-byte prefixes and repeating 10-byte suffixes. The former is common in many uses of the engine (see ItemSpaceDataStructures.htm). The Latter simulates attribute values with small numbers of distinct values. Also, the ZLib compressor got going with its Huffman coding because we used only 40 values per byte, simulating common English text.

Please examine the tester itself and try changing various parameters. Access the download link from our home page. Here is the actual output:

Performance test of InfinityDB B-Tree database engine.
File=c:\temp\InfinityDBPerformanceTest.idb
Entries=1000000 bytes/entry=20

RANDOM COMPRESSIBLE DATA

insert with cold cache, compressible data time=3.6259999999999994 iterations/sec=275785.9900717044
compressible db size=1420075
next with compressible data time=3.5159999999999996 iterations/sec=284414.1069397042
delete with compressible data time=3.6409999999999996 iterations/sec=274649.8214776161
insert warm cache, compressible data time=3.3129999999999997 iterations/sec=301841.2315122246

RANDOM UNCOMPRESSIBLE DATA

insert cold cache, uncompressible data time=14.219 iterations/sec=70328.43378577959
uncompressible db size=25637958
next with uncompressible data time=7.561999999999999 iterations/sec=132240.1481089659
delete with uncompressible data time=9.577 iterations/sec=104416.83199331732
insert warm cache, uncompressible data time=9.39 iterations/sec=106496.27263045793

Performance test done.