openRDF.org Welcome Guest   | Login   
  Search  
  Index  | Recent Threads  | Who's Online  | User List  | Search  | Help  | RSS feeds

Forum has been closed down
This forum has been closed down due to extensive spamming activities. Please use the mailing list instead.


Quick Go »
Thread Status: Normal
Total posts in this thread: 1
[Add To My Favorites] [Watch this Thread]
Author
Previous Thread This topic has been viewed 18319 times and has 0 replies Next Thread
Feb 15, 2006 12:45:20 PM

arjohn
OpenRDF project lead
Member's Avatar

The Netherlands
Joined: Jan 23, 2004
Posts: 1289
Status: Offline
NativeStore LUBM benchmark results

(follow up from this posting)

LUBM is a benchmark for RDF/OWL repositories that has been developed by the Lehigh University. In response to a question whether the native RDF store in Sesame 2 scales and performs better than the one in Sesame 1.x, we started benchmarking Sesame 2's NativeStore using LUBM. Results can be found below and in the attached PDF. A comparative test against Sesame 1's native store is still on the TODO list.

Tests were performed on "desktop" hardware: Intel Pentium 4 2.80GHz; 1024 MB DDR333 RAM; Seagate ST380011A harddisk running "Linux version 2.6.13-15-default (geeko@buildhost) (gcc version 4.0.2 20050901 (prerelease) (SUSE Linux)) #1 Tue Sep 13 14:56:15 UTC 2005" (cat /proc/version).

LUBM's data generator was used to generated test sets of various size, scaling from LUBM-1 with ~100,000 statements upto LUBM-500 with ~70,000,000 statements. Two sets of queries were run: the queries in serql-queries.conf that are bundled with the LUBM benchmark, and the queries in optimized-serql-queries.conf that have been optimized a little by hand (mainly reordering of triple patterns).

The NativeStore has been configured to create and use two triple indexes: spoc and posc. The NativeStore is a pure triple store, as no inferencer for this store has been implemented yet. This affects some of the query results in that they don't return complete answers, or don't return an answer at all. These particular queries tend to have very low query evaluation times.

Attached is a PDF version of the OpenDocument spreadsheet with the benchmark results. The query evaluation time figures show two lines: the red line represents the original query, the yellow line represents the optimized query.

A word of caution on the use of benchmarks and statistics: "Then there is the man who drowned crossing a stream with an average depth of six inches." -- W.I.E. Gates smile
----------------------------------------
Attachment LUBM-NativeStore-monza.pdf (49438 bytes) (Download Count: 1924)

----------------------------------------
Arjohn Kampman, OpenRDF project lead, Aduna
Show Printable Version of Post        Hidden to Guest [Link] Report threatening or abusive post: please login first  Go to top 
[Show Printable Version of Thread]