Page 142 - ITU Kaleidoscope 2016

2016 ITU Kaleidoscope Academic Conference

subscriber path, Optimal Cache Placement based on Content
Popularity (OCPCP) [18] - where the popularity of incoming
content is calculated on the basis of cached contents and the
new content is cached based on its popularity value, Network
Coding based Cache Management (NCCM) [19] - jointly con-
siders content routing and caching strategy through Linear
Network Coding (LNC), WAVE [20] - caches contents based
on their access count, Most Popular Caching (MPC) [21] –
caches contents based on their popularity values, Dynamic
Fine-Grained Popularity-based Caching (D-FGPC) [22] – the
modification of MPC, however, unlike MPC in FGPC the
threshold value for content caching is changed dynamically
based on content popularity values. Moreover, some experi-
mental assessment of cache management in ICN are presented
in [23]. Some other research on caching has been proposed
in [24–31].
All of the mentioned strategies investigated only one aspect
of the caching (i.e., either content placement or replacement),
and none of them covers both aspects. Therefore, a flexible Figure 1. An example topology
caching strategy is needed so that to place the contents at the
best possible position and (on the arrival of a new content)
replace one of the cached contents. Actually, the contents is, if(C p >= V t ), where C p is the content popularity and
are cached in the random access memory (RAM) while the V t is the threshold value. The V t is kept 10 in the CPCE to
available static random access memory (SRAM) or dynamic avoid flooding as all incoming contents have the C p value
random access memory (DRAM) is limited in the size. The of 1. Therefore, a content in the CPCE is only considered
DRAM, which is a volatile memory and needs to be refreshed popular when its C p reaches 10. Furthermore, if the cache of
regularly [32] is currently available at 10GB maximum [33, all on-path routers overflows and a new content arrives, the
34]. In other words, cache size is the biggest constraint in CPCE uses the Least Recently Used (LRU) policy to replace
ICN caching. However, in most of the existing available one of the cached contents from each router except that the
strategies, the maximum cache space is occupied. In addition, router which has the maximum outgoing interfaces, denoted
if the memory of a router becomes full and a new content by R max , to accommodate the new arrived content. The rea-
arrives, none of the existing strategies has any policy for that son to leave the content at the R max is to avoid maximum
but simply replaces one of the cached contents by the new bandwidth utilization, as majority of the content requests pass
arrived content. This is achieved using a replacement policy, through that router. However, the question arises of how long
i.e., either Least Recently Used (LRU) or Least Frequently R max will keep contents as the cache of that may also over-
Used (LFU). As LRU and LFU replace the content based on flow. Therefore, in case of cache overflows at R max , on the
the access time, the replaced content may be very popular and arrival of a new content, LRU policy is used to evict one of the
therefore the subsequent requests for the same content will be cached contents accessed least recently. The evicted content
forwarded to the server. In this way, extra content retrieval is cached at the underlying router, denoted by UR 1 (R2 in
delay may occur and thus maximum bandwidth is utilized. Figure 1), placed immediately below R max . Furthermore,
To overcome such problems, we propose in this paper a new if the cache of UR 1 is also full then the Random policy is
caching strategy, Cache Popular Content Everywhere (CPCE), applied here to accommodate the content coming from R max .
which caches popular contents on all network nodes on the The purpose of using the Random policy is to avoid search-
publisher-subscriber path. The CPCE strategy is explained in ing overhead for content replacement as other replacement
the following section. policies, e.g., LRU, take some time to find the contents based
on their access time. Here the replaced content from UR 1 is
moved to the next down cache router, i.e., UR 2 , and the same
3. PROPOSED STRATEGY
procedure is followed until the content reaches the router
placed near the subscriber, i.e., UR n .
The proposed CPCE caches contents at all on-path routers
As the LRU does not care of content popularity and it is
available all the way from publisher to subscriber, as in CEE.
deployed at R max , it may also evict the most popular content.
However, the difference between our proposed caching strat-
egy (CPCE) and the CEE is such that CEE caches every Now, if a new content request arrives for the evicted popular
content from off-path, it may also go through R max ; however,
incoming content while CPCE caches contents once their pop-
if hit does not occur at R max , it can be found from RT that
ularity reaches a specified threshold value. In other words,
a content is cached when its popularity reaches the thresh- the requested content is available at the UR i .
old value in the Request Table (RT) - a table which locally Hence, even the CPCE is designed for on-path caching but
calculates popularity values based on content requests, that off-path nodes can also benifit from its versatility.

– 124 –