DPDrPeering.net
Buy the Book
Back matter Answer key

Answers to Practice Questions

Worked answers to the practice questions in each chapter.

TODO (Bill): fill in the answers for each chapter. The 2014 print edition has the worked answers on PDF pages 365–369.

Chapter 2 — Internet Transit

  1. (C) Max(500Mbps, 800Mbps) × $5/Mbps = $4,000/month
  2. YES — commit early.
  3. Max(500Mbps, 800Mbps) × $5/Mbps = $4,000/month
  4. vs.
  5. 1000Mbps × $3/Mbps = $3,000/month

Chapter 3 — The Internet Transit Playbook

    1. The maximum savings from the Optimal Transit tactic can be calculated to be the monthly cost of traffic when transit volume hits commit point minus the monthly cost had they executed the early commit.
    2. The maximum savings would be the entire transit load at the market price.
  2. One can imagine many additional maneuvers including swaps for services in kind, leveraging free trials, or keeping zero commit transit services in waiting indefinitely, bundling free transit with colocation, reselling agreements to consortiums of buyers, bandwidth exchanges, etc.
  3. You and your buddy can share a network, share the costs, and balance the network load, and leverage the aggregation benefits of not needing the bandwidth at the same time.

Chapter 4 — Internet Peering

  1. No. In these situations it is good to draw a picture like Figure 4-2 to see that peering is not a transitive relationship and therefore ISP B and ISP C will not hear each other’s routes announced by ISP A.
  2. Peering is the exchange of access to each other’s routes but not access to each other’s transit service. In this light, peering is not barter. If you look at the ISP service being the exchange of routes and traffic, then peering looks more like barter. If you believe that routes are a commodity, each equally valuable, then peering looks more like barter. Another interesting question is then, what is the basis for taxing? The market price of transit? The price of paid peering? The cost of the next best alternative (cheapest transit in the market)?
  3. Omitted.
  4. Not likely since ISP C is getting paid transit fees by ISP A.
  5. Some argue that paid peering should be cheaper than Internet Transit since only a subset of the routing table is provided. Others argue that peering should be the same or more expensive than transit since it provides a higher performance path. It is an interesting debate to have.

Chapter 5 — The Business Case for Peering

  1. 1500Mbps × $2/Mbps = $3K + $5K = $8K
  2. No: Transit is cheaper @ $2/Mbps. Peering = $3K + $1K + $3K = $7K / 2000Mbps = $3.50/Mbps
  3. No, you are still on the hook for 2Gbps × $2/Mbps = $4000, adding in the $300 for the peering port puts you at $4300/month. Cheaper to just send all 2Gbps through transit.
  4. The Effective Peering Range is between 2.5 Gbps ($5000/2500Mbps = $2/Mbps, the cost of transit) and 7Gbps.

Chapter 6 — Selecting an IXP

  1. A carrier or ISP would be competing with their customers by launching an IXP. ISPs and carriers try to avoid paying their competitors or reinforcing their competitor’s position in the market, so it will be difficult to get critical mass. While they may try and be “neutral” by allowing competitors in to sell to the population in their data center, this may cause customer churn. History has proven that ultimately, the clash between the operation of the IXP business and the operation of the carrier or ISP business will cause the IXP to fail.
  2. An ISP might prefer a more expensive IXP if the peers that they seek are in that IXP. An ISP might prefer to support a local IXP operated as a non-profit over a less expensive commercial operation for business alignment reasons. An ISP might prefer a more expensive IXP if there are transit sales opportunities or if they are already in the building so extending transport is inexpensive and logistically easier than building into the less expensive IXP.
  3. I would avoid talking about the telecommunications access unless there were some providers scheduled to install. I would talk about the deployment process being easy, about the customer focus and attentiveness. I would avoid talking about the ISP Current Presences since there are no ISPs in the IXP yet. Operations support can be a strong card to play, and business alignment might be a differentiator worth mentioning. If there is flexibility on cost, then I would raise that. If there is strong financial support behind the next IXP, I would play that card but I would avoid to the extent possible the exchange population and the existing versus emerging exchange. If the intent is to build a strong regional IXP and the routes are underserved, I would play the regional routing strength card.

Chapter 7 — Public vs. Private Peering

  1. If an ISP can max out their peering port with 7Gbps then they will realize a cost of $1400 / 7000 Mbps = $0.20/Mbps. To get the equivalent economics with a $300 cross-connect the ISP would need to peer $300 / $0.20 = 1500 Mbps across the $300 cross-connect.
  2. The larger ISPs tend to have a restrictive peering policy so peer with only a few others and peer large amounts of traffic. These ISPs are less likely to benefit from peering with a lot of the smaller players on a public switch.
  3. The smaller ISPs tend to prefer the public switching fabric to peer small amounts of traffic with as many small players as possible.

Chapter 8 — The 20th Century Peering Ecosystem

  1. Poaching customers is easier when competitors know the customer name, contact information and their selected transit provider.
  2. Not necessarily. The NSPs have routes that they learn from peers. These routes would not be propagated to other peers, so there is value in buying transit from these players besides customer routes.
  3. D.

Chapter 9 — The Global Peering Ecosystem

  1. ISP Y will say “No, we prefer the revenue.”
  2. ISP Y will say “No, we prefer the revenue that we receive indirectly from ISP B, and peering would both reduce that revenue and annoy our customer ISP B.”
  3. While ISP A is likely to correctly point out that peering will not reduce revenue for ISP Y, ISP Y is likely to respond by saying that it already hears the ISP A routes for free through its peering with ISP X. Having no desire to annoy his peering relationship and gaining no significant benefit from peering with ISP A, ISP Y is likely to respond by trying to sell ISP A some transit.

Chapter 10 — The 21st Century Peering Ecosystem

  1. The content providers seem willing to pay more for a better end-user experience. The CDNs promise the content providers that they will get the content as topologically close as possible to the eyeballs, and since the CDN competitors all pay for paid peering, every other CDN must do the same. ISPs may decide that the cost is about the same and it performs better, so why not?
  2. The access networks, since there is no alternative path to reach their eyeball customers.

Chapter 11 — The ISP Peering Playbook

  1. All of these tactics are manipulative and apply social engineering to some degree. These are effectively identifying the levers (the motivations) and applying pressure (the tactic) to cause the desired effect.
  2. This of course is in the eyes of the beholder. Some believe that even the fake NOC outage can be justified as a counterbalance to the antisocial refusal-to-peer-openly tactics employed by the target.

Chapter 12 — Taxonomy of Data Centers

  1. European model IXPs are generally cheaper. In many cases I would see European model IXPs sell peering ports 30% cheaper than the U.S. model IXPs.
  2. The European model IXPs are owned by members. Since the members compete with each other they probably do not want to see their competitors pay a lower price for exchange point services than they pay.
  3. The U.S. model IXPs operate a data center in addition to the interconnection services and therefore have a higher operations cost than the European model IXPs.
  4. D. Even though they invite competitors into the building it is still a carrier-owned data center.
  5. C. This is a classic telco hotel model.
  6. B. This is a carrier-neutral data center.
  7. A. By reselling Internet services to the customers they are a hosting company.
  8. C. This is the U.S. model IXP where the company owns the data center and interconnection facilities.
  9. A. This is the European model where they only operate the peering fabric.

Chapter 13 — The IXP Playbook

  1. Yes, the IXP is past critical mass. On average each member is paying $2,000/mo for 30 Gbps = $0.06/Mbps. In a market where the price of transit is $1/Mbps, peering provably makes sense here.
  2. The question goes to the IXP legitimacy. Are there ISPs there that value the exchange and therefore likely to stay? Are these ISPs going to peer with me now or in the future?
  3. Omitted. Send your answer to the author: ipp133@DrPeering.net
  4. Omitted. Send your answer to the author: ipp134@DrPeering.net

Chapter 16 — The Future

Answers coming soon.

Chapter 15 — International Peering

Answers coming soon.

← Chapter 15 — International Peering Table of Contents →