ld-profiler

A dirt-simple Linked Data profiler. For now, a simple CLI that takes a Linked Data interface description (LDID) in VoID format and performs two profiling tasks:

1. For all example resources given in the LDID a look-up is performed and the response times are displayed
2. For SPARQL end points announced in the LDID an array of generic queries is executed and the response times are displayed

How does it work?

Invoke syntax is:

python ldpro.py ( -h | -p (LDID) | -c (LDID) )  [r]

where:

-h ... prints help
-p ... loads LDID (the path to a local file), runs profiler and reports results
-c ... loads LDID (the path to a local file) and reports on its content (use to verify LDID)
 r ... optionally, specifies the number of runs (defaults to 3)

For example, to perform 5 profile runs for each of the profile targets described in a LDID that is provided in the test directory, use the following command:

python ldpro.py -p test/ldid-dbpedia.ttl 5

When the above command is issued, the result should look something like this:

================================================================================

I. Results for example resources:

 Dereferencing http://dbpedia.org/data/Digital_Enterprise_Research_Institute.rdf

  Run 0: 0.121638059616ms
  Run 1: 0.0866210460663ms
  Run 2: 0.0787258148193ms
  Run 3: 0.0795500278473ms
  Run 4: 0.079206943512ms
  -------------------------
  Average: 0.0891483783722ms

 Dereferencing http://dbpedia.org/data/Galway.rdf

  Run 0: 0.367477178574ms
  Run 1: 0.36781001091ms
  Run 2: 0.366918802261ms
  Run 3: 0.396972179413ms
  Run 4: 0.366279840469ms
  -------------------------
  Average: 0.373091602325ms

II. Results for SPARQL end point:

 Executing queries againt http://dbpedia.org/sparql

 Query: SELECT ?s WHERE { ?s ?p ?o } LIMIT 1

  Run 0: 0.0598089694977ms
  Run 1: 0.0606231689453ms
  Run 2: 0.0609130859375ms
  Run 3: 0.0849769115448ms
  Run 4: 0.0592401027679ms
  -------------------------
  Average: 0.0651124477386ms

 Query: SELECT ?s WHERE { ?s a ?o } LIMIT 1

  Run 0: 0.062292098999ms
  Run 1: 0.0598909854889ms
  Run 2: 0.0597748756409ms
  Run 3: 0.0594980716705ms
  Run 4: 0.0634031295776ms
  -------------------------
  Average: 0.0609718322754ms

================================================================================

Who is behind this?

When Michael read Cary Millsap's excellent ACM Queue article Thinking clearly about performance he was wondering if profiling in the Linked Data world has already been established. He looked around and didn't find much convincing stuff so he decided to write something himself.

License

This software is in the Public Domain.

Ideas

• SPARQL runs: externalise input so that new queries can be added without changing the script
• JSON output
• GAE app with URI
  • Graphical representation of the results (bar chart, avg.)
  • Web service for interactive access
• Microdata support
• Measure throughput (?)