1 files changed, 22 insertions, 0 deletions
diff --git a/src/site/03_performance.mkd b/src/site/03_performance.mkd
new file mode 100644
index 0000000..34e545c
--- /dev/null
+++ b/src/site/03_performance.mkd
@@ -0,0 +1,22 @@
+
+## Performance
+
+The information provided here may be based on flawed test procedures.  You have to be the judge of what is performant and non-performant.
+
+### iciql statement generation
+
+Performance of iciql statement generation is not currently benchmarked.
+
+### iciql+database performance comparison
+
+The following data was generated by running the *single-threaded* iciql test suite.  All database connections are pooled and re-used within each execution of the test suite using [Apache Commons DBCP](http://commons.apache.org/dbcp).
+
+Connections are pooled to normalize embedded database performance with out-of-process database performance.  Some of the Java embedded database configurations have a very high startup-time penalty.  Notably, H2 is slow to open a database and its performance is substantially affected if connection pooling is not enabled to keep the embedded database open.
+
+All tables are created as CACHED when the database distinguishes between CACHED and MEMORY tables.
+  
+All performance numbers include the combined overhead of iciql statement generation and JUnit 4 test framework execution so they are not bare-metal database metrics.  
+
+<pre>
+%DBPERFORMANCE%
+</pre>
+\ No newline at end of file