Bootstrapping evolvability for inter-domain routing with D-BGP Raja Sambasivan David Tran-Lam, Aditya Akella, Peter Steenkiste
This talk in one slide What evolvability features needed in any Q inter-domain protocol? New New Existing protocol Protocol 1 protocol 2 Pass-through Multi-protocol A support structure D-BGP (BGP w/features): rich, evolvable Internet 2
The inter-domain routing infrastructure Allows access to Internet’s content (e.g., ) Today, composed of a single protocol, BGP AS Edge AS Dest Src 3
BGP has many well-known issues Cannot limit ingress tra ffj c High convergence times No QoS Only one best path ASes can be spoofed Proposed solutions Wiser [NSDI’07] R-BGP [NSDI’07] BGPSec [IETFv8] SCION [SP’11] MIRO [SIGCOMM’06] Pathlets [SIGCOMM’09] NIRA [CCR’03] Arrow [SIGCOMM’14] EQ-BGP [AINA’06] HLP [SIGCOMM’05] 4
BGP has many well-known issues Cannot limit ingress tra ffj c High convergence times No QoS Only one best path ASes can be spoofed BGP is rigid: requires neighbors to use it “Grumpy Darwin,” Aaron Birk, http://www.npr.org/sections/krulwich/2012/10/18/163181524/charles-darwin-and-the-terrible-horrible-no-good-very-bad-day 5
Rigidity results in isolated islands s k T Peering link n i l r ) e $ d ( i v o r P Src Dest AS supports BGP AS supports new protocol Island Isolation dis-incentivizes deployment 6
Skirting rigidity with data-plane tunnels s k Peering link n i T l r ) e $ d ( i v o r P Src Dest AS supports BGP AS supports new protocol Island Tunnel path BGP path Incentivizes non-deployers to fj ght evolution 7
Key contributions The two modest evolvability features Pass-through support Multi-protocol structure Makes data-plane tunneling optional D-BGP, which is not far from BGP Only Required 900 lines of code BGP already includes pass-through support Characterization of D-BGP’s bene fj ts Enables a rich Internet w/many protocols Incentivizes adoption by accelerating bene fj ts 8
How we identi fj ed evolvability features Evolvable Internet + R-BGP [NSDI’07] BGPSec [IETFv8] MIRO [SIGCOMM’06] SCION [SP’11] Wiser [NSDI’07] Pathlet Routing [SIGCOMM’09] ! ! ! BGP ➡ mod. BGP BGP // Services BGP ➡ FIA Reqs Reqs Reqs Global reqs Pass-through support Multi-protocol structure (provided by BGP) 9
An evolvable Internet Gulf Wiser Wiser Wiser MIRO d n a SCION l B s HLP I SCION B HLP SCION SCION SCION Runs many routing protocols All ASes support a shared baseline (B) 10
Taxonomy of evolvability scenarios BGP ➡ FIA BGP ➡ mod. BGP BGP // Services Properties Di fg erent ctrl info * Extra ctrl info Ex. Wiser, R-BGP MIRO, Arrow, * SCION, HLP, Pathlets Incentives Deployers Inc. bene fj ts Pro fj ts Inc. bene fj ts Non deployers Joint control Future pro fj ts Joint control " Send across gulfs Send across gulfs Reqs Enable discovery Send in-band Send in-band 11
Evolvability scenarios (FIA) BGP ➡ FIA E.g., extra paths Properties Di fg erent ctrl info Di fg erent ctrl info or link states Ex. SCION, HLP, Pathlets Incentives Deployers Inc. bene fj ts Non deployers Joint control Send across gulfs Reqs Send in-band 12
Deploying SCION, a FIA protocol B I s l a n G d u l f Src Dest Baseline advertisement Packet hdr (IP + SCION) 13
Deploying SCION, a FIA protocol B I s l a n G d u l f Src Dest Baseline advertisement Packet hdr (IP + SCION) 14
Deploying SCION, a FIA protocol B I s l a n G d u l f Src Dest Baseline advertisement Packet hdr (IP + SCION) 15
Global reqs for an evolvable Internet Wiser Wiser Wiser MIRO SCION B HLP SCION B HLP SCION SCION SCION Inform islands about Provide common protocols on paths denominator for e-e paths 16
Features Requirements Disseminate Pass-through across gulfs support Disseminate in-band " Enable discovery Inform islands about protocols on paths Multi-protocol data structure Provide common denominator for e-e paths 17
Outline Evolvability features D-BGP design D-BGP eval 18
D-BGP overview ✚ BGP advs BGP processing with with Multi-protocol structure IA support & pass-through = IA processing Integrated advs (IAs) Island Island 19
D-BGP’s integrated advertisements Dest. address: 128.2.42.52/24 Path vector Abstracts within- AS # Island ID island paths Prevents ASes from [ discounting end-to-end paths that include within-island paths ] 20
An IA for a path Dest. address: 128.2.42.52/24 Path vector Island ID: G Island ID: R AS # Island ID 4 8 D-BGP 3 2 8 G HLP [ 4 1 3 R Dest SCION 2 1 Dest. ] 21
An IA for a path Dest. address: 128.2.42.52/24 Proto desc. Path vector Island desc. D-BGP AS # Island ID Proto(s) Fields Value(s) 8 D-BGP O N r e i g x i t n G HLP h o p 1 [ E 4 3 2 Within- Path 1 3 . 2 R SCION . Path 2 SCION island 5 2 . 1 paths 1 Dest. ] 22
Outline Evolvability features D-BGP design D-BGP eval { Accelerating bene fj ts Control-plane overhead Quagga implementation New-protocol deployments 23
Accelerating bene fj ts evaluation Compared deployment in an Internet with: BGP D-BGP Explored bene fj ts as function of adoption E.g., # paths to dests at upgraded edge domains Experiments done in simulation Used Brite [Mascots’01] to generate 1,000-node topology Used modi fj ed routing simulator [SIGCOMM’14] 24
D-BGP accelerates bene fj ts for SCION 0 0 5 3 Avg. # of paths to all dests advertised to SCION sources 5 2 6 1.84x 2 extra paths 0 5 7 1 5 7 8 D-BGP baseline BGP baseline 0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Fraction of ASes running SCION 25
Summary BGP’s rigidity Evolvability Two features su ffj cient for evolvability D-BGP provides large evolvability bene fj ts 26
Recommend
More recommend