Hi Andy,
very promising results. Its great to see another store handling the
100M dataset with a good performance.
Andreas will run another experiment with a slightly changed version of
the benchmark on our machine in about 2 weeks. We will include TDB
into this run so that we get directly comparable numbers for TDB.
If you have additional ideas for improving the benchmark before the
second run, please let us know.
We already added two extra queries to the mix: A CONSTRUCT query and a
query with unbound predicates.
Cheers
Chris
This report is for TDB running the Berlin SPARQL Benchmark [1]
TDB is a persistent subsystem for Jena.
Thanks to Andreas Schultz who provided the configuration files and
updated tools necessary to run the benchmark on the downloaded
datasets.
Date: 2008-08
== Results
= Load
Using TDB's bulk loader: "tdbloader"
Cold load into an empty database.
50K 250K 1M 5M 25M 100M
Time/s 5.56 15.25 50.46 256.65 1258.07 5370.09
TPS 9,020 16,427 19,824 19,483 19,872 18,621
= Query
50K 250K 1M 5M 25M
100M
Query 1 0.007711 0.036216 0.066769 0.137431 0.442776
1.463148
Query 2 0.032370 0.046859 0.047317 0.047442 0.046882
0.056338
Query 3 0.008156 0.021385 0.058452 0.141895 0.410103
1.429356
Query 4 0.011558 0.019491 0.067215 0.220680 0.654449
2.571348
Query 5 0.069205 0.357140 0.973735 3.624503 21.729141
84.927111
Query 6 0.005759 0.043215 0.049023 0.062252 0.524726
11.309403
Query 7 0.037831 0.045588 0.197200 0.436487 0.711938
0.771399
Query 8 0.027343 0.060075 0.235520 0.480224 0.798117
0.834888
Query 9 0.006449 0.006994 0.009983 0.021610 0.032256
0.041381
Query 10 0.004890 0.014506 0.060580 0.103583 0.187552
0.197450
== Software
Jena 2.5.6 <http://jena.sf.net/>
ARQ 2.4 <http://jena.sf.net/ARQ>
TDB 0.5 <http://jena.sf.net/TDB>
Joseki 3.2 <http://www.joseki.org>
Java 1.6.0_02
Sun Java HotSpot(TM) 64-Bit Server VM (build
1.6.0_02-b05, mixed mode)
OS: RHEL5
TDB configuration:
None (TDB has no configuration options)
== Hardware
CPU: 4 CPU: AMD Opteron(tm) Processor 280 @1.8GHz
Memory: 10Gbytes
Note: this used non-local disk, using a shared disk array in a data
center. This provides a more realistic set up of enterprise use. It
does make it difficult to draw deep conclusions when comparing with
different setups using local disk.
Disk: HP EVA8000 running firmware xcsp-6100.
The LUNs are configured as VRAID5, with read cache enabled and write
cache in write-back mode. Two Qlogic fibre-channel ports (2GB
speed) connected to an MDS 9140 edge switch each, then an MDS9509
core switch and finally the EVA.
Andy
[1] http://www4.wiwiss.fu-berlin.de/bizer/BerlinSPARQLBenchmark/
--------------------------------------------
Hewlett-Packard Limited
Registered Office: Cain Road, Bracknell, Berks RG12 1HN
Registered No: 690597 England
