SCAN: Scalable Content routing for

Transcription

1 ICC SCAN: Scalable routing for content-aware Networking Munyoung Lee, Kideok Cho, Kunwoo Park, Td T Ted Taekyoung Kwon, Yanghee Choi (mylee@mmlab.snu.ac.kr) Seoul National University

2 Outline Introduction Basic SCAN Operation Design choice of SCAN Simulation Results Conclusion 2/16

3 Introduction (1/2) Current Internet Designed based on end-to-end paradigm More and more traffic on the Internet is attributed to contentoriented services File download and web access Internet routers are not aware of the contents Inefficiency of content retrievals Bandwidth wastes (duplicated downloads) Cannot exploit a nearby copy of the content 3/16

4 Introduction(2/2) -oriented networking New paradigm : redesign the Internet in a clean-slate manner DONA(Data-Oriented Network Architecture) CCN( Centric Network) // NDN(Named Data Network) PSIRP(Publish-subscribe Internet Routing Paradigm) NetInf(Network of Information) Exploit in-network storage attached to routers Cache the contents Support content-aware routing Expected to solve the inefficiency in content retrievals 4/16

5 IP vs. Routing IP Routing +Scalable Routing +-aware SCAN! +Widely used +Efficient delivery -Inefficient - Not scalable - Do not consider the closest or multiple copies - Huge number of contents should be maintained - IDs (CIDs) are usually flat or difficult to aggregate 5/16

6 SCAN Overview Goal: a content routing that solves the inefficiency of IP routing routers (C-routers) perform both content routing and the IP routing IP routing will guarantee the reachability routing (called scanning ) ) is responsible for finding nearby or multiple copies (efficiency) SCAN exchanges only the information of the cached contents in the local content table (LCT) to mitigate the routing scalability issue 6/16

7 SCAN Operations (1/2) Routing (3) A (3) Server A Request (1) (3) A (1) (3) (4) (5) C-router Request Scanning Request Response Range Request Delivery 7/16

8 SCAN Operations (2/2) Delivery (an example) (5) A (5) Server A C-router (1) Request Scanning Request (4) (4) (4) (5) A (3) (4) (5) Response Range Request Delivery 8/16

9 Design choice of SCAN (1/3) Store the information of the subset of the contents t To mitigate routing scalability May cause content reachability problem Use IP routing as a fallback solution 9/16

10 Design choice of SCAN (2/3) Adopt Bloom fl filter (BF) to compress the content information Local exchanges of content information may cause the routing scalability problem due to large number of contents C-routers have one BF for each interface which compresses the information of the contents advertised CID CID % match /16

11 Design choice of SCAN (3/3) All the bits of the BF may be set to 1 When the large number of content information is accumulated (multi-hop distribution) To solve this problem, each C-router will decay the bits of a BF probabilistically before exchanging CRTs Ex) the half of the bits set to 1 can be randomly set to 0 Exact match cannot be used due to the decayed d information during CRT exchange Conclude a content exist when # of matched bits predefined threshold LCT : Local Table CRT : Routing Table 11/16

12 Receives content request Forwarding Decision Performs IP routing LCT lookup Scanning Can I afford Yes CID matched Yes Respond scanning, now? in LCT to the end host No No CRT lookup # of matched bits threshold h Yes Forward the content request to the matched interfaces No Discard Done 12/16

13 Simulation Environments Simulation Environments Simulator Topology Number of routers & end hosts distribution Discrete event-driven simulator 1 transit domain and 5 stub domains generated using GT-ITM 105 routers & 1,000 end hosts Randomly distributed 10,000 contents (1GBytes) 1,000 popular contents are randomly replicated Request distribution Zipf distribution with parameter 1.0 Cache size Bloom filter, threshold Comparison 100GBytes 3,500bit with 14 hash functions, δ=0.1 SCAN-FULL*, SCAN w/o BF**, IP w/ caching***, IP Routing *: SCAN with all content information **: SCAN without BF compression 13/16 ***: IP routing which can exploit cached contents

14 Simulation Results The original server load is reduced when SCAN (and SCAN-FULL) reduces a the requested content is served by C- significant amount of network traffic (per routers link) SCAN achieves 86.4% load reduction IP routing w/ caching and SCAN w/o BF compared with IP routing often utilize the cached content located at 47.3% and 39.6% reduction for SCAN w/o nearby C-routers BF and IP w/ caching, respectively 30.7% and 35.7% reduction compared with IP routing 14/16

15 Conclusion SCAN achieves both reachability and efficiency by using both IP routing and content routing information is compressed to handle the scalability issue SCAN locates one or more closer copies of the requested content Reduces the total volume of network traffic Provides better load balancing among the links 15/16

16 Q&A 16