Date post: | 28-Dec-2015 |
Category: |
Documents |
Upload: | justin-horton |
View: | 218 times |
Download: | 0 times |
New Directions in Peering for Tier-2 and Content
Providers
Jeb R. Linton
Staff Network Engineer, EarthLink
Topics to be Covered
Peering Motivations and Economics The Cost of Peering - History Disruptions - What do they mean?
New Low Transit Costs New Tier-1 Peering Initiative Metropolitan Multipoint Private Peering
FUD (Teaser: What Happens Next?)
The Role of Peering
To Large Tier-1s Mission-Critical
To the Rest of Us Used to Decrease Cost of Transit Used to Increase Quality of Routes Usually not mission-critical *
* Your Mileage May Vary
Peering Cost – the Legacy NAPs
2-3 years ago, cost justification for NAP peering was easy based on the high cost of Transit
See Bill Norton’s whitepaper: A Business Case for ISP Peering for the origin and explanation of this method of cost justification. Bill’s graphing method will be used throughout.
Transit Cost vs. NAP Peering - Before
$-
$200
$400
$600
$800
$1,000
$1,200
$1,400
10 16 25 40 65 100
160
250
400
650
1000
Bandwidth (Mbps)
Co
st
Old Transit
NAP #1
NAP #2
NAP #3
Peering Cost – the Legacy NAPs, cont’d
Disruption #1: Rapid drop in Transit cost makes legacy NAPs much less desirable
Transit Cost vs. NAP Peering - After
$-
$200
$400
$600
$800
$1,000
$1,200
$1,400
10 16 25 40 65 100
160
250
400
650
1000
Bandwidth (Mbps)
Co
st
New Transit
NAP #1
NAP #2
NAP #3
Peering Cost – Private Lines as Supplement
Evolution: NAP Peering is supplemented with Private Peering.
Private Line to a Single Peer
$-
$100
$200
$300
$400
$500
$600$700
$800
10 25 65 160
400
1000
2500
6500
Bandwidth (Mbps)
Co
stNew Transit
NAP #1
NAP #2
NAP #3
Single PL
Peering Cost –Private Lines for Many Peers
Problem: Private Peering to many Peers can be difficult because of the repeated expenditure and low circuit utilization efficiency
Private Lines to Four Peers
$-
$100
$200
$300
$400
$500
$600
$700
$800
Bandwidth (Mbps)
Co
st
New Transit
NAP #1
NAP #2
NAP #3
Priv. Lines
Peering Cost – Tier-1 Peering InitiativeDisruption #2: Large Tier-1 Providers create the
Next-Generation Peering Initiative
Common facilities are chosen, to which each major Tier-1 will acquire fiber rights
Allows excellent scalability while minimizing repeated cost
Independence, Fault Tolerance, and Provisioning time improve
Next-Generation Tier-1 Peering: Before N*(N-1)/2 Links in
each of 7 Cities Dependence on
RBOCs Slow Provisioning Mostly OC-3 and
OC-12 Poorly adapted to
use of Dark Fiber
Spr
UU
Gen
CW
ATT
L3
Q
Next-Generation Tier-1 Peering: After One-time Dark
Fiber provisioning to a single facility
Each Tier-1 maintains routers in the same room
Cheap, fast cross-connects for OC48 or higher peering
Spr
UU
Gen
CW
ATT
L3
Q
NexGen Peering Center
Peering Cost – Appearance of Metro Ethernet
Disruption #3: Cheap Metro Ethernet services connect to multiple peers over the same port using Point-to-Point and Multipoint VLANs
Metro-Ethernet Multipoint Private Peering
$-$100$200$300$400$500$600$700$800$900
10 25 65 160
400
1000
2500
6500
Bandwidth (Mbps)
Co
st
Transit
Private Lines
Metro Ethernet
Disruption: Low-Cost Transit
Legacy Peering methods lose popularity – NAP attachments and Private Line Peering links are no longer saving money.
Some companies retain NAP presence just to meet contractual requirements, or for marketing reasons.
Increased pressure to find cheaper Peering methods.
Next-Gen Tier-1 Peering: Implications Large Tier-1s draw smaller Tier-1s to new
Centers for Peering
Largest Tier-2 and Content providers drawn to the Centers hoping to peer with Tier-1s
Smaller Content, Tier-2s, Enterprises and others are drawn to new Centers for competitive Transit and some Peering
What happens to all the other collocation?
Price Disruption: Metro-Ethernet Most major Collocation spaces in “NFL Cities”
are already Metro-Ethernet enabled
Pressure to replace individual Private Lines (DS-3, OC-3, OC-12) with Metro-Ether where possible.
Legacy NAP viability may depend on ability to accommodate Ethernet links Many are ATM-based and charge heavily for
collocation Some have adapted or may yet adapt
Metro Multipoint Private Peering: Before
Private Peering NAP Peering In-Building Peering if
you’re lucky.
Low efficiency High Cost Slow Provisioning
Old NAP
Private DS3s
ATM OC3
ME
YOU
HIM
Those Guys
(The situation for Smaller Tier-1, Tier-2, and Content)
Metro Multipoint Private Peering: After
Cross-Connect to an Ethernet port in your current facility
Multiple VLANs for Private and Public Peering, and Transit
Link to NexGen Peering Centers?
ME
YOU
Those Guys
(The situation for Smaller Tier-1, Tier-2, and Content)
HIM
NexGen Peering Center
Transit Provider
PublicPeeringVLAN
?
Efficient, Cheap, and Fast
Metro Multipoint Private Peering Addresses some limitations of Private Peering
Multiple Peer links on the same physical port Private, Public, and Transit share same ports
Many configurations and payment options Pay by port, by utilization, or by allocation P-to-P VLANs per peer, or semi-public
Multipoint VLANs Bandwidth allocated per port or per VLAN Technical options: Broadcasts, Switch Config,
Failover, etc.
Metro Multipoint Private Peering
Probably won’t scale to tens of Gbps May scale further once 10Gig-E is available Scaling is better addressed by Dark
Fiber/Lambda connections to the NexGen Peering Centers
Optimum scalability only reached if all local peers use the same Metro-Ether Provider (!)
FUD – The Tier-1 Peering Initiative
Q. Is this Unfair? Non-Inclusive? Anti-Competitive?
A. No… it’s just common sense, and anyone can join in.
Q. Will this increase Transit prices? A. Competition should improve Transit pricing in
the Peering Centers, and have no effect outside them.
FUD – The Tier-1 Peering Initiative
Q. What about my old CoLo? A. Thanks to cheap metro, most large
Collocation centers are still viable, but you may want to get a foot in the door at the new peering centers.
Q. Are the Tier-1’s really committed to this? A. Yes… It’s the only viable long-term option,
and they don’t want to waste the investment they’ve already made.
FUD - Metro-Multipoint Private Peering Q. Early Entry is scary… why not wait? A. Very low commitment is needed; can start
below 10M
Q. Does the service have the robustness of Private Line?
A. All vendors use diversity wherever available. Most use same or similar configuration to the standard SONET/Lambda services we all rely on
FUD – Metro-Multipoint Peering, cont’d Q. What about the financial stability of the
vendors? Isn’t the cheap pricing based on gaining market share? We saw what happened in the last year… What about today’s vendors? (Applies to Metro-Ether and Peering-Collocation vendors)
A. Reduce your risk with short contract terms…Multiple vendors means you can switch if need be.
FUD – Metro-Multipoint Peering, cont’d Q. What technical gotchas should we be aware
of? A. Management overhead similar to old NAP,
and Broadcast and Congestion issues need to be resolved
Bottom line: Mitigate risk with short contract terms, and do your own financial analyses…
Caveat Emptor!
Points of Contact: Earthlink
Jeb Linton (Architecture/Peering Strategy) – [email protected]
Josh Fleishman (Peering Coordinator) – [email protected]
Backup: What Happens Next
1. In “NFL Cities”, expect most large Internet companies (and many small ones) to migrate to the NexGen Peering Centers.
2. In all areas with Metro-Ethernet, expect off-net Collocation Centers to die off.
3. Expect ATM-based NAPs to continue losing popularity.
4. Metro-Ethernet will predominate for new P-to-P Private Peering links and links to the NexGen Peering Centers.
5. Metro-E will be used for Multipoint Peering. (Services already emerging from several vendors)
6. One Metro-E vendor will begin to predominate for Peering in each Metro area.
7. Peers will migrate from MM-PP to the Centers as they grow.
Backup: What Happens Next… the Big Picture Large Internet companies interested in significant Peering
will migrate toward the NexGen Peering Centers Mid-sized companies will use both:
NexGen Peering Centers, for Tier-1 and High-Speed peering Metro-Multipoint for smaller peers, to maximize savings
Small companies will: Start with a Metro Ether port for Transit, B-B, etc. Add peering VLANs on the same port for cost savings
Point-to-Point for B-B links and opportunistic private peering Metro-Multipoint for additional cost savings
Use Metro Ethernet as a bridge to the NexGen Peering Centers as they grow
Backup - Terminology (for convenience only) Tier-1 – Providers of Internet Access to
Autonomous Systems (a.k.a. “Leased Transit Service”, “Transit”)
Tier-2 – Providers of Internet Access to Individuals
NexGen Peering Centers – Collocation Centers chosen as common locations for Peering by the Largest Tier-1 Providers
Backup - Who wants to peer with whom? Largest Tier-1
Each Other Other Tier-1
Largest Tier-1, each other, large Tier-2, and some Content Large Tier-2
Content, all Tier-1, and each other Content
Everybody but each other Other (Who is Other?)
Academe, Enterprises, R&D, etc… the Usual Suspects
Backup: The “Peering Promiscuity Chart”
LargestTier-1
OtherTier-1
LargestTier-2
Content
Other
OtherTier-2
Other Similar
Other Similar
Other Similar
Other Similar
Content-HeavyTraffic Mostly
Outbound
Eyeball-HeavyTraffic Mostly
Inbound
Arrows point toward Targeted Peers
Backup - Evaluating Available Methods of Peering
Questions to Ask: How much Peering traffic is locally available using this peering
method? How much does this Peering Method cost per Mb for the
expected traffic level? What does the company pay per Mb for Transit in this metro
area?
Things to consider: Should be half the cost of Transit (per Mb) or better Methods of Peering are complimentary, not exclusive