The Great Public vs Private Peering Debate Peering
The Great (Public vs. Private) Peering Debate Peering at 10 G William B. Norton Co-Founder & Chief Technical Liaison Equinix, Inc. RIPE 50 – Stockholm EIX-WG May 5, 2005 wbn@equinix. com
Research Question • Does 10 Gbps Public Peering make sense financially? • Private Peering (Gig. E) is the next best alternative • Perception: 10 G router HW is too expensive • 10 G Public vs. 1 G Private Peering
Public Peering Model P P ISP Gear 1. Router w/ 2 10 Ge 10 G Interface Card 2. 10 G Public Peering Port 3. Colo Modeling… P R 10 G IX sw Assumption: We are ignoring the transport cost into the IX :
10 G Public Peering Modeling Assumptions 1) 2) 10 G Public Peering Port at IX: $10, 000 per month 3) Colocation: $1, 000 per month $20, 154 per month Assumption: 36 month amortization of capital equipment
10 G Public Peering Scaling Add 2 nd 10 GEs Assumption: Aggregation Efficiency ~1. 5: 1
Graphically Assumption: If Peering towards 10 G, probably buying transit ~$10/Mbps
Private Peering Model ISP Gear • Router w/10 G NICs 10 G R 10 G sw • 10 G Public Peering Port • Colo Add Private Peering Switch P P P :
Scaling 1 G Private Peering
Graphically
Religion: Public Peering is Better than Private Peering 1. Aggregation Benefits • • Peaks+Valleys=Efficiency Large Gradations=headroom 2. Ease of Administration • • • Easy+Fast Trial Peering No additional Cost/No Signing Authority Financial Predictability Scales Large (950 M) Peering Sessions 3. Public Peering as Selection Criteria
Religion: Private Peering is Better than Public Peering 1. Easier to Monitor • • No Net. Flow needed! No Blind Over Subscription Problem 2. More Secure 3. Cheap • $200/40 Mbps =$5/Mbps 4. More Reliable + Easier to Debug 5. Signals a more attractive peer
Religion: Hybrid Approach • Publicly peer small sessions • Privately peer large sessions at 40% • Vijay Gill’s Peering Lifecycle Argument 1) 2) 3) 4) 5) 6) Publicly Peer w/anyone Large Peers to Privates Peer with peer’s peers More selective to get benefits Depeer to reduce ops load Exclusively privately peer Some debate here
Bottom Line: Public 10 G vs. Private 1 G early break even points 1) Similar curves, early break even points (2 peering sessions) 2) As traffic grows, difference shrinks 3) European 10 G IX Port Fees are less
What if 10 G Port costs $4400/mo? Public Peering Ease of Administration Watch Assumptions: Cost of 10 G HW Dropping Sub 10 G Architecture
Transition Peering Dynamics Historically, two competing dynamics here… Peering migrates – Move Public Private Peering • Can’t afford the 10 G router – let’s private peer – Move Private Public Peering • Can’t trunk or load share across 2 Privates
Acknowledgements Ren Provo (SBC), Richard Steenbergen (n. Layer), James Rice (Lo. NAP), Todd Underwood (Renesys), Stephen Wilcox (Tele. Complete), Vanessa Evans (LINX), Niels (ANS-IX), Chris Malayter (TDS Telecom), Patrick Gilmore (Akamai), Frank Orloski (T-Systems), Vijay Gill (AOL), Vish Yelsangikar (Net. Flix), Nathan Hickson (e. Bay), Steve Feldman (CNet), Lane Patterson (Equinix), Joy Fender, Falk Bornstaedt (T-Systems), Remco Donker (MCI), Danny Mc. Phearson (Arbor Networks
- Slides: 16