Two important part of the in-network caching
- Cache strategy. (Which? placement )
- Content discovery mechanism.(request-to-cache routing)
This paper focus on content discovery mechanism.
Current related work
By and large, request-to-cache routing can follow one of two approaches: either opportunistic on-path or co-ordinated off-path.
- Opportunistic on-path
Content is searched opportunistically along the shortest path(or a designated path) towards a content oringin.
Drawbacks: very limited search scope(only the nodes along a path) thus limited gain.
- The off-path resolution-based routing
Mapping requests to content items cached in nearby( or not )nodes, usually at predeteminated rendezvous points.
There is an extra routing table at each router to statically or according to a content addressing scheme configure the path towards the corresponding rendzvous points or the off-pat caches.
Drawbacks: too much storage or communication(signaling overhead).
- Co-ordinated off-path
- Flooding requests or meta-requests
- Hash-routing techniques
Mapping requests to physically co-located servers.
- A scoped flooding-based content discovery mechanism.
A ring model,which limits the spread of the flood to the immediate neighborhood.
Drawbacks: higher gain, extra coodination and communication overload.
- Breadcumbs——an opprtunistic forwarding scheme
Nodes keep track of the direction of Data packets towards the end users with LCE placement strategy.
Drawbacks: LCE’s redundancy.
In this paper, authors try to combine the merits of both worlds by using an opportunistic coordination approach, which requires each router to keep track of only a small amount of state and does not require signaling or update protocaols to locate content.
Contributions:
- Introdcing EFIB(Ephemeral FIB) which maps name prefixes to a set of next hops(either in upstream or downstream).
- To maximize network storage utilization, leverage the ProbCache. Each data packet is cached exactly once along the delivery path.
- To decrease overload, propose budget-based forwarding strategy.
Ephemeral FIB
The differences between EFIB and FIB:
- the FIB table of a router is used to forward requests towards content origins. The EFIB table, on the other hand, is a temporary storage of forwarding information used to forward interests towards local (i.e., intra-AS) caches.
- unlike the FIB table, which is traditionally populated via out-of-band means such as a routing protocol [9], the EFIB table is populated opportunistically (see below), using locally available information gathered from the forwarded Data packets.
Placement strategy
Each data packet is cached exactly once along the delivery path. The caching point is decided based on the caching capability of the path and the corresponding position of a given router in the network,
similarly to [13].
In order to implement this content placement strategy and enable the installment of the EFIB entries, we introduce a “Placement flag(PL)” bit to the Data packets. PL set to “1” indicates that a placement is not necessary (i.e., the packet is either responded by a router or has already been cached) and conversely PL set to “0” indicates that a placement should be performed.
Data packet processing
Forwarding strageties
- Multicast: forwarding request to matching entries in EFIB and FIB at the same time.
- Stop and wait: off-path is prior to on-path. The router will not forward interest to next matching enty until receiving last interest’s response.
- Budget-based multicast: forwarding an interest needs cost, initialiaze every interest by setting a budget, when budget can not afford cost, the interest will be discarded.
