CloudFlare is currently hosting 2M websites, and is handling 5% of the Internet traffic. EdgeCast has about 6,000 customers, and is handling about 5% of the Internet traffic. Both have similar POP footprints. CloudFlare has raised $72M in funding, and part of that money will be used to build out 50 additional POPs in 2014.
Currently, CloudFlare has 24 POPs located at the major Internet Exchanges. Once CloudFlare is completed with its POP roll out, it will have a total of 74 POPs. The question is “does CloudFlare really need that many POPs”, or can the funds be used to invest in other critical areas, such as buying more servers and bandwidth. We’ll answer that question here.
POPs per CDN
 CloudFlare – 24 POPs
 EdgeCast – 23 POPs
 Aryaka – 25 POPs
 Limelight – 30 POPs (estimate)
Quick Analysis
Once CloudFlare has 74 POPs, it will have about 3x the number of POPs compared to EdgeCast, Aryaka and Limelight. EdgeCast, Limelight, and Aryaka all have similar POP footprints. They’re in the same cities, and sometimes in the same neutral data centers. There is a reason behind this; these CDNs understand that there is a point of diminishing returns when it comes to building out too many POPs.
Next, let’s do an example using logic and the roundtriptime (RTT) performance metric, to see if it’s worthwhile to build out additional POPs. In Europe, the last time I checked, the RTT was 50ms from any two points (major cities). If 50ms RTT can be achieved with 7 POPs, and 40ms RTT can be achieved with 15 POPs, is it worth the expense to build out 8 more POPs for an extra 10ms in performance improvement. Probably not. Thus, 7 POPs should suffice for Europe. Now let’s look at it from a cost perspective.
Annual Cost per CDN POP
 100 Servers x $4k = $400,000
 2 Brocade Routers x $15k = $30,000
 2 Cabinets x $2k/month x 12 months = $48,000
 100Gbps Commit x $1.5/Mbps x 12 months = $1.8M
 Miscellaneous x $1,000 x 12 months = $12,000
 Total annual cost = $2.3M
Based on the estimates above, the average annual cost for a CDN SuperPOP is about $2.3M, plus or minus a few hundred thousand. The $2.3M annual cost is for 100 hundred servers and 100Gbps of bandwidth commit. Anything less than this set up isn’t really a SuperPOP, but more of a miniPOP. The SuperPOP architecture is far superior to the miniPOP architecture model. That’s the reason EdgeCast, Limelight, and Aryaka use it today. If CloudFlare builds out 10 POPs at $2.3M cost/POP, this is going to eat up $23M of cash. If CloudFlare builds out 50 POPs, that’s going to clean out the bank.
Conclusion
Based on the summary analysis above, CloudFlare should build out fewer POPs, but with more servers and bandwidth. Not only does it make sense from a cost perspective, but also from a performance perspective. Let’s see if this post will have any impact on CloudFlare’s plans to build out 50 POPs.
CloudFlare vs EdgeCast vs Aryaka (POP Comparisons)
City 
CloudFlare (24) 
EdgeCast (23) 
Aryaka (25) 

1 
Los Angeles 
Y 
Y 
N 
2 
New York 
N 
Y 
N 
3 
San Jose 
Y 
Y 
Y 
4 
Seattle 
Y 
Y 
N 
5 
Dallas 
Y 
Y 
Y 
6 
Wash DC 
Y 
Y 
N 
7 
Atlanta 
Y 
Y 
N 
8 
Chicago 
Y 
Y 
Y 
9 
Miami 
Y 
Y 
N 
10 
Ashburn 
N 
Y 
Y 
11 
Newark 
Y 
N 
N 
12 
Toronto 
Y 
N 
N 
13 
San Paulo 
N 
Y 
Y 
14 
Bueno Aires 
N 
N 
Y 
15 
Mexico City 
N 
N 
Y 
16 
Valparaiso 
Y 
N 
N 
17 
Amsterdam 
Y 
Y 
Y 
18 
Frankfurt 
Y 
Y 
Y 
19 
London 
Y 
Y 
Y 
20 
Madrid 
N 
Y 
N 
21 
Milan 
N 
Y 
N 
22 
Paris 
Y 
Y 
N 
23 
Stockholm 
Y 
Y 
N 
24 
Vienna 
Y 
Y 
N 
25 
Warsaw 
Y 
N 
N 
26 
Prague 
Y 
N 
N 
27 
Hong Kong 
Y 
Y 
Y 
28 
Osaka 
N 
Y 
N 
29 
Tokyo 
Y 
Y 
Y 
30 
Singapore 
Y 
Y 
Y 
31 
Seoul 
Y 
N 
Y 
32 
Sydney 
Y 
Y 
Y 
33 
Beijing 
N 
N 
Y 
34 
Shanghai 
N 
N 
Y 
35 
Delhi 
N 
N 
Y 
36 
Mumbai 
N 
N 
Y 
37 
Bangalore 
N 
N 
Y 
38 
Chennal 
N 
N 
Y 
39 
Dubai 
N 
N 
Y 
40 
Johannesburg 
N 
N 
Y 
41 
Tel Aviv 
N 
N 
Y 
42 
Moscow 
N 
N 
Y 