Using  the  Working-­‐Set  Theory  to  Model  CCN  Cache  Performance                    Florin  Coras  1,  Damien  Saucez...
Upcoming SlideShare
Loading in …5
×

CCNxCon2012: Poster Session: Adapting Working-set Theory for Optimal Cache Eviction

746 views

Published on

Adapting Working-set Theory for Optimal Cache Eviction
Florin Coras (Technical University of Catalonia), Damien Saucez (INRIA Sophia Antipolis, France)

Published in: Technology
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total views
746
On SlideShare
0
From Embeds
0
Number of Embeds
3
Actions
Shares
0
Downloads
1
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide

CCNxCon2012: Poster Session: Adapting Working-set Theory for Optimal Cache Eviction

  1. 1. Using  the  Working-­‐Set  Theory  to  Model  CCN  Cache  Performance   Florin  Coras 1,  Damien  Saucez2   1Universitat  Politècnica  de  Catalunya,  Barcelona,  Spain   2Inria,  Sophia  Antipolis,  France       Research  challenge   Objective   Evaluate cache performance for CCN traffic Evaluate cache performance Determine best eviction policy Evaluate link between the demand generating process and cache performance Working-­‐Set  Theory  [DS72]   The Working-‐Set (WS) is typically used to model OS page caching performance Early  Results     Model primitives: programs and pages 100k unique content items A program can be seen as a reference string: r1, r2 .., where ri is a page out of a set of pages New content inter-‐arrivals ~ EXP(WEB and ) Applicability is conditioned by the reference Inter-‐arrivals of new demands are biased to increase with time In order to apply WS theory to CCN type of traffic we consider the following equivalences: program demand trace page demand If t measures time in content demands then the WS may be defined as: W(t,T) is the set of distinct content items that have been observed among the T most recent demands demands   tradeoff t-­‐T+1   t   t+1   W(t,T)   time   W(t+1,T+1)   Cache size and miss rate tradeoff is possible Intuitively the WS may be seen as modeling a cache of size |W(t,T)| Future  Work   If s(T) is the average WS size and m(T) is the miss Stationary locality of reference observed for IP rate: packets [FCD12]. Is it for content demands? Validate with real content demand trace unique   s(T)   demands   Further  Reading   m(T)   working-‐ F. Coras, A. Cabellos T   T+1   demands  

×