On Monday 29 January 2001 10:15, Seb Bacon wrote:
> I've just been experimenting with the RAM Cache manager, and FWIW I thought
> I'd share my findings:
> I ran some simple tests using ab against a single, dynamic page which has
> some display logic both in the filesystem product and dtml.  I cached the
> index_html using the default RAM Cache Manager settings.
> Without caching, response time appeared to increase in direct proportion to
> the number of simultaneous requests up to 10.  Above 10 connections,
> response time worsened dramatically.  When there were 15 simultaneous
> connections, the response time averaged at approximately 10 seconds.  The
> relationship looked logarithmic, but I didn't really take enough samples to
> be sure.  Plus, the system I tested it on only has 96Mb RAM, and it was
> swapping a bit.
> In all cases the cache improved performance.  The average response time was
> improved by up to 70% where more than 10 concurrent users were simulated.
> There was an improvement of 20% - 25% between 5 and 10 concurrent requests.
> There was no significant improvement where there were no concurrent
> requests.  The relationship between the number of simultaneous requests and
> the response time remained linear across all measured levels of
> concurrency.

very cool. although i find the lack of improvement on a non concurrent 
requests a bit strange if its a the dtml is doing real computation vs. 
serving a string from memory.

> Not exactly a real-life usage scenario, but it sounds impressive :)  Not
> sure I understand what the cache keys are for though... is there any more
> documentation forthcoming on this?

so you can cache requests based on dynamic input. 

useful for authentication to present different cache pages to different 
users, and also to present common request dependent information from cache.

as for real documentation, use the source luke


Zope maillist  -  [EMAIL PROTECTED]
**   No cross posts or HTML encoding!  **
(Related lists - 
 http://lists.zope.org/mailman/listinfo/zope-dev )

Reply via email to