Network Names in Content-Centric Networking. CCN Names

Transcription

1 Network Names in Content-Centric Networking ACM ICN CCN Names Expressed as URIs /a/b/foo /us/edu/uci/cs/tsudik/papers/acm-icn16.pdf Encoded as TLVs T_NAME Length / Name segment TLVs / ACM ICN

2 Names in CCN Applications Consumer Serve /a/b/foo or forward request Request /a/b/foo Router Network Publish+Serve /a/b/foo Producer /a/b ACM ICN Names in the Network ROUTER CS PIT FIB (1) Exact name in CS? (2) Exact name in PIT? (3) Prefix in FIB? ACM ICN

3 Dual Roles for Names Applications use names to meaningfully express and identify content Human readability is nice! Network entities (routers) use names as sequences of binary strings A router doesn t care about readability or arbitrarily long components Q: Why is the same representation used at both layers & in both contexts? ACM ICN Outline CCN Network Names Name Translation and Its Implications Translation Analysis Related & Future Work ACM ICN

4 CCN Network Names Goal: translate application names into a format that: Facilitates standard network processing (exact match and LPM searches) Removes arbitrarily long names from the network Removes all locationirrelevant information from the name (as seen by Cryptographic Hash Func&on? Translation should: be a bijec&on or very close to one map arbitrarily long strings to fixed-length output only apply to loca&onspecific parts of a name. routers) ACM ICN Name Translation Example /a/b/foo/version=0x00/chunk=0x01/pid=0x02 payload: <bar> T_NAME locator components possibly contained in FIBs T ApplicaTon-specific components not Contained in FIBs Unique Name Fingerprint /H(a)/H(a,b)/H(a,b,foo)/version=0x00/chunk=0x01/PID=0x02/N P =0xFF payload: <bar> N P =H (/a/b/foo/version=0x00/chunk=0x01/pid=0x02) ACM ICN

5 Name Encoding Compressed Unmodified T_NAME Length T_MAPPED_SHA256/32 Length H(a), H(a, b), H(a, b, foo) T_N P 32 (silly example) / < name fingerprint = 0xFF > / T_VERSION 0x01 0x00 T_CHUNK 0x01 0x01 T_PID 0x01 0x ACM ICN Application Processing Consumer: Maps application names to network names for outgoing interests Inverts mapping for incoming content Producer: Pre-computes network names for all its locator prefixes Stores in inversion table Looks up content corresponding to incoming interests based on this inversion table Signed content objects contain N p might also carry app name ACM ICN

6 Current Network Processing Interest pkt (1) decode_pkt hdr name (optional fields) (2) decode_name (3) compute_hash x n (4) HT_lookup x (2+n) CS PIT FIB FIB We obviate the need to hash in order to perform: FIB, CS and PIT lookups ACM ICN FIB Processing Summary En&ty: Impact: Consumer Increased online processing Producer Increased offline computaton & storage of inversion table Router Faster FIB lookup with pre-computed name-prefix hashes Faster PIT and CS lookups with N P (Poten&al) benefits due to fixed-size N P ACM ICN

7 Analysis Setup Questions: How big should a hash digest be? What is the impact on packet sizes? What are the resulting collision properties? What s consumer processing overhead? Unibas URI data set: ACM ICN Name Properties ACM ICN

8 Name Properties ACM ICN Name Size Impact (for interests) RaTo of network name to standard name as size of T()/H() grows ACM ICN

9 N p Size Overhead (for Content) ACM ICN Collision Assessment (32-bit hash) ACM ICN

10 Processing Overhead (consumer) Per-name costs (µs): Average: 1, ( 1ms) Minimum: Maximum: 2, à Reasonable compared to network I/O Throughput (c/b): Average: 1, Minimum: 1, Experimental seong: Intel 2.8 GHz Core i7 Un-opTmized implementaton based on PARC Libparc libraries Ideal seong: Use Intel intrinsics for hashing (~9 c/b) Work on wire-encoded packets Maximum: 3, ACM ICN Wrap-up Motivated separating CCN application and network names A concrete mechanism for name translation function Assessed quality of name translation function and performance implications for all CCN entities ACM ICN

11 Related Work* q CCN names q Requirements [Ghodsi et al., ICN 11] q Location-agnostic names [Van Adrichem et al., Nomen 13] q Focus on FIB algorithm improvements q FIB algorithm modifications based on tries, hash tables, Bloom Filters, etc. [Quan et al., Networking 13], [Perino et al., ANCS 14], [So et al., ANCS 13], [Fukushima et al., Nomen 13] q Several rely on lexicographical name ordering q Our scheme breaks this! q Hop-by-hop optimizations, e.g., passing length of previously matched name in FIB but not on FIB inputs *See paper for references ACM ICN Future Work Compare performance of current FIB techniques with and without network names Play with various hash functions & sizes Explore further uses for translation function T() ACM ICN

12 /this/is/the/end/version=0x00/chunk=0x01/pid=0x02 ACM ICN