[EMAIL PROTECTED] wrote:
Author: markt
Date: Sat Nov 4 16:33:02 2006
New Revision: 471306
URL: http://svn.apache.org/viewvc?view=rev&rev=471306
Log:
Fix bug 37356. I know concern has been expressed about using syncs here.
I have tested the impact of using syncs with both an artificial test case that
just calls access() followed by endAccess() on the StandardSession object and
with JMeter using different numbers of threads to make a request in the same
session.
For the single thread case:
- JMeter impact was negligible.
- With the wrapper, it was about 150 nanoseconds per request which explains why
the JMeter test didn't shown anything.
For two threads:
- JMeter impact was about 9 microseconds (0.4%) per request
- The wrapper was much higher at 50 milliseconds but this is expected given the
nature of the test is such that far more contention than normal will be seen
For five threads
- JMeter impact was about 75 microseconds (4%) per request
My conclusion is that the uncontended impact is negligible and the contended
impact is sufficiently low for syncs to be acceptable.
Obviously, these figures are for my hardware but it is fairly standard so the
figures should be typical.
I think you should use the strict compliance flag, as only 2 people need
activity checking.
Rémy
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
