[ndnSIM] The weird results of replacement policy

[Alex Afanasyev]

Hi Pan,

You probably want to try to play with simulation parameters.  First, run simulation for way more than just 60 seconds, since in 60 seconds you most likely does not get enough interest packets that form Zipf distribution.

Other parameters to play with: try to increase content store size gradually and see what will change.  Another thing is to see how ConsumerZipfMandelbrot parameters affect the result.   

There is a small chance that there are "too many" popular items and Lru fails to keep all of them in cache, however I would mostly suspect short simulation time.

---
Alex

On Jul 2, 2013, at 6:43 PM, Pan <ihenrypan at 163.com> wrote:

> Hi All,
> 
> I did some experiments on replacement policy and the results were weird.
> 
> The scenario just based on the ndn-simple.cc. The configuration information as follows:
> The network topology:
>     +----------+     1Mbps      +--------+     1Mbps      +----------+
>   | consumer | <------------> | router | <------------> | producer |
>     +----------+         10ms    +--------+          10ms   +----------+
>  The consumer is ConsumerZipfMandelbrot model. NumberOfContents set to 1000.
> ContentStore size is 100.
> Simulation time is 60s.
> 
> Now, I test exsiting replacement policy(Lru, Fifo and Random) by using the ndn::CsTracer to statis the cache hit rate.
> The cache hit rate of policy are:
> Lru: 26.7%
> Fifo: 23.8%
> Random: 33.2%
> Random policy's hit rate is greatest in every experiments. Random policy's hit rate should be least in the theory. It's weird.
> 
> Is the scenario has problems?
> Could you explain the results for me?
> Thank you in advance.
> 
> ---
> Pan

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.lists.cs.ucla.edu/pipermail/ndnsim/attachments/20130703/eb87d876/attachment.html>