Learn How The Internet Works
HUBS AND SPOKES
There are 1,000 megabits in one gigabit and 1,000 gigabits in one terabit of data transmitted.
How the Internet Travels Across Oceans 2019
#Hurricane Sandy hurts Cloud Internet Infrastructure 10/31/12
: Learn How the Internet Works – Hubs and Spokes
The Vast World off Fraudulent Routing As network security engineers have attempted to categorize blocks of IP addresses associated with spam or malware for subsequent filtering at their firewalls, the bad guys have had to evolve to continue to target their victims. Since routing on the global Internet is based entirely on trust, it’s relatively easy to commandeer IP address space that belongs to someone else. In other words, if the bad guys’ IP space is blocked, well then they can just steal someone else’s and continue on as before. In an attempt to cover their tracks, these criminals will sometimes originate routes using autonomous system numbers (ASNs) that they don’t own either.
BAD ROUTE PROPAGATION - USE PROTECTION IF PEERING PROMISCUOUSLY
IETF standard for RPKI-based AS path verification.
China Telecom—National LambdaRail Although our final example isn’t recent, it is worth mentioning in this discussion. During the big China Telecom routing leak of April 2010 that caused an international stir, it is interesting to note where the bogus routes announced by China Telecom (AS23724) propagated the farthest. Before it ceased operations earlier this year, National LambdaRail (NLR) was a “high-speed national computer network owned and operated by the U.S. research and education community.” NLR also had a peering relationship with China Telecom, the state telecom of China. When NLR received the bogus origination announcements from its Chinese peer, it accepted them and routed traffic to China that was intended for numerous other locations around the world.
Resolved: National Lambda Rail (NLR) to cease operation. Last updated on February 18, 2014 at 8:15AM Along with connection to the commodity Internet, Penn State connects to two other national networking infrastructures, Internet2 and NLR. ITS has received notification that NLR will cease operations as of 7:00 PM EST on Monday, February 17th. Commodity Internet and Internet2 will continue normal operations. ITS anticipates no impact to Penn State connectivity or traffic flows. If any impact is detected, either with connectivity to a non-PSU host site or with access to any PSU host from outside the PSU domain, please contact the NOC to open an incident. For more information, please contact Network Operations Center (814-865-4662).
11/6/18 CHINA TELECOM'S INTERNET TRAFFIC MISDIRECTIONS
Great piece by @dangoodin001 on BGP hijacking. BGP hijacking is never just an accident or snarfu, it's a calculated move to consume what you usually aren't allowed to.
BORDER GATEWAY PROTOCOL — Strange snafu misroutes domestic US Internet traffic through China Telecom Telecom with ties to China's government misdirected traffic for two and a half years. For almost a week late last year, the improper routing caused some US domestic Internet communications to be diverted to mainland China before reaching their intended destination, Doug Madory, a researcher specializing in the security of the Internet’s global BGP routing system, told Ars. As the following traceroute from December 3, 2017 shows, traffic originating in Los Angeles first passed through a China Telecom facility in Hangzhou, China, before reaching its final stop in Washington, DC. The problematic route, which is visualized in the graphic above, was the result of China Telecom inserting itself into the inbound path of Verizon Asian Pacific. Madory said the improper routing he reported finally stopped after he “expended a great deal of effort to stop it in 2017.” China Telecom's Internet Traffic Misdirection BY Doug Madory His report went on to endorse a proposed standard known as RPKI-based AS path verification. The mechanism, had it been deployed, would have stopped some of the events Madory documented, he said.
SUBMARINE CABLE MAP reveals the 550,000 miles of cable hidden under the ocean that power the internet. Every time you visit a web page or send an email, data is being sent and received through an intricate cable system that stretches around the globe. Since the 1850s, we've been laying cables across oceans to become better connected. Today, there are hundreds of thousands of miles of fiber optic cables constantly transmitting data between nations.
2016 Facebook and Microsoft are laying a massive cable across the middle of the Atlantic. Dubbed MAREA—Spanish for “tide”—this giant underwater cable will stretch from Virginia to Bilbao, Spain, shuttling digital data across 6,600 kilometers of ocean. Providing up to 160 terabits per second of bandwidth—about 16 million times the bandwidth of your home Internet connection—it will allow the two tech titans to more efficiently move enormous amounts of information between the many computer data centers and network hubs that underpin their popular online services.
SUBMARINE CABLE MAP 2014
TeleGeography Maps -- Undersea cables crossing the Atlantic Ocean - Crossing-2 (or AC-2), are the main data lifelines between continents.
How NSA and GCHQ are tapping internet cables
12/1/14 The NSA's fourth-largest cable tapping program, codenamed INCENSER, pulls its data from just one single source: a submarine fiber optic cable linking Asia with Europe.
Until now, it was only known that INCENSER was a sub-program of WINDSTOP and that it collected some 14 billion pieces of internet data a month. The latest revelations now say that these data were collected with the help of the British company Cable & Wireless (codenamed GERONTIC, now part of Vodafone) at a location in Cornwall in the UK, codenamed NIGELLA.
For the first time, this gives us a view on the whole interception chain, from the parent program all the way down to the physical interception facility. Here we will piece together what is known about these different stages and programs from recent and earlier publications.
10/26/15 Russian Ships Near Data Cables Are Too Close for U.S. Comfort some American military and intelligence officials that the Russians might be planning to attack those lines in times of tension or conflict.
10/2/14 Global Infrastructure explained by Retired NSA Technical Director William Binney.
8/12/14 Google helps build 'Faster' cable under Pacific Ocean with a host of Asian telecoms giants - China Mobile, China Telecom, Global Transit, KDDI, and Sinapoure's SingTel. The cable, dubbed Faster, will connect the US with Japan and cost about $300m (£179m; 225m euros). "The Faster cable system has the largest design capacity ever built on the trans-Pacific route, which is one of the longest routes in the world." The cable will connect Chikura and Shima in Japan to the major hubs on the west coast of the US - Los Angeles, San Francisco, Portland, and Seattle.
About 99% of all transoceanic Internet data is sent via undersea cables.
The Australian Signals Directorate, is in a partnership with British, American and Singaporean intelligence agencies to tap undersea fibre optic telecommunications cables that link Asia, the Middle East and Europe and carry much of Australia's international phone and internet traffic. One former Australian Defence intelligence officer told Fairfax Media that access to submarine fibre optic cable traffic ''gives the 5-eyes [intelligence alliance] and our partners like Singapore a stranglehold on communications across the Eastern Hemisphere''. A major GCHQ interception program, codenamed Tempora, that involves harvesting all data, emails sent and received, instant messages, calls, passwords and more, entering and exiting Britain via undersea fibre-optic cables.The SEA-ME-WE-3 is one of the most important undersea cables accessed by GCHQ and the US National Security Agency. The Australian Signals Directorate and the highly secretive Security and Intelligence Division of Singapore's Ministry of Defence also play key roles in intercepting communications traffic through Asia.
Carriers and ISPs use the cables to pump data across the Atlantic.
Two of the major gateways are in Brookhaven, N.Y., (on Long Island) and in northern New Jersey, where the cables come ashore.
NYC Data Centers:
Transatlantic fiber lands at about 10 different places in Massachusetts, Rhode Island, Long Island and New Jersey that, after having landed, all goes to one of two facilities.
Telecom companies use carrier hotels to interconnect networks to allow data sharing and users of one network to connect with those of another.
There is a high probability that every time you go on a Web site - your Internet traffic passes through 111 8th Ave. at some point.
These 2 buildings are critical to the nation's infrastructure located in lower Manhattan which serves as the major network hubs for the U.S. The buildings, known as carrier hotels, are a 2.9 million square foot structure at 111 8th Ave., and a 1.8 million square foot facility at 60 Hudson St.
Telex has Interconnection and data center company sites and has co-location facilities at 60 Hudson as well as 111 8th Ave. New York City. NYC1, NYC2, NJR1
OPERATIONS CONTACT: 888.835.9832 or TELXTECHSUPPORT@TELX.COM
MOBILE (612) 860-8789 or RSTERBENZ@TELX.COM
Atlantic Metro: LGA1 (325 Hudson Street) LGA4 (121 Varick Street)) LGA1 connects to LGA6.
Low Lying Manhattan’s FLOOD Zone A
Major flooding n NYC Data Centers Hurricane Sandy
75 Broad Street, NY
Internap | Peer 1 | Navisite and other data center providers. ALSO 8th Avenue facility houses several data centers in the low-lying Zone A of Manhattan, was severely impacted, by Hurricane Sandy.
25 Broadway (Telehouse colocation provider )
lower Manhattan has its data center in Chelsea and at the Staten Island Teleport.
Data centers at Google-owned Carrier Hotel 111 8th Ave. NY. Telx Colocation and interconnection specialist is the largest service provider at 60 Hudson Street, one of the the leading carrier hotels in Manhattan operates some 490,000 square-feet of data center space at 111 8th Ave building.
Equinix, Voxel/Internap, XO Communications, INIT7, a Swiss provider of IPv6 infrastructure all run data centes at 111 8th Ave and 75 Broad Street in Manhattan
- Both 75 Broad and 33 Whitehall were located in the “Zone A” flood zone.
HUBS AND SPOKES
TABLE OF CONTENTS:
- Global Internet Primer, Architecture, Finance, Governance, Demand, Voice, What Next?
- International Internet Bandwidth - Providers, Connectivity, Exchanges,
- International Internet Indicators - Network Metrics
GLOBAL INTERNET PRIMER:
Who pays for the Internet? "The answer is either really long or really short depending on what you're trying to say," says Scott Bradner, a leading Internet expert at Harvard University. The Internet does not have a set economic model, so there's no standard way network providers are remunerated for the resources they use.
End of story.
The longer answer is more complicated, precisely because the Internet's provisioning model is not static. Whereas a typical call over the public switched telephone network (PSTN) involves two or three different networks, a typical Internet transmission may involve five or ten. And the connectionless transmission technology on which the Internet is based also means that the role of each of these networks cannot easily be predicted in advance. Smaller networks typically pay larger networks for connectivity, but many larger networks themselves exchange their traffic without charge under a peering, sender-keeps-all basis (see Figure 1, "A Primer on Peering"). They seek to recoup their network costs primarily from their end users and their downstream ISP customers -- not always from other networks, as in the telephony world. To understand why the Internet's schemes for funding international networks are so different from the traffic- based settlement arrangements over the PSTN, a brief digression on technology is useful.
The Connectionless Network
Traditional phone networks, built for voice communications, switch or assign a dedicated end-to-end circuit for every call. That is reliable, but oriented heavily toward a limited set of applications whose bandwidth usage is fairly steady: every connection needs its own circuit. Minute-by-minute and circuit-by-circuit payment methods consequently developed to compensate network providers.
The Internet is a radical departure from this pattern. It is based on packet switching: no dedicated connection is required, and a dedicated route doesn't have to be set up between sender and receiver. Instead, all communication is converted to digital format, broken up into chunks of data called packets or datagrams, given an address, and sent out into the network -- packet by packet. What's most significant is that the path each packet takes may be radically different: all they have in common is that they end up in the same place, ready to be reassembled into a coherent message. That makes it hard to bill Internet communications in the way that traditional phone communications are billed. But it is more efficient for moving traffic around when that traffic may consist of a tiny message sent one moment, a huge graphics file the next, and then nothing for a few minutes -- what traffic engineers call "bursty traffic," because it comes in sudden bursts.
Some have compared the process to mailing a book through a postal service that accepts only postcards: each page must be sent separately, and arrives individually; the receiver must reassemble the pages back into the right order before reading. When first proposed in the 1960s by Paul Baran, in the U.S., and separately by Donald Davies in the United Kingdom -- and later refined in the early 1970s by pioneers like Robert Kahn and Vint Cerf
- it sounded like a crazy idea. But it worked. (For an engaging history of the period, see Peter Salus' book, Casting The Net.) Since the early principles of Transmission Control Protocol/Internet Protocol were published by Kahn and Cerf in a paper entitled "A Protocol for Packet Network Intercommunication" (IEEE Transactions on Communication, May 1974), packet delivery has not changed much. Along the way, routers
- computers acting as "smart" switches, as opposed to the automatic forwarding of bridge switches
- still store and forward packets.
After forwarding, if the first router doesn't receive acknowledgment that the packet has arrived safely at its destination, it resends the packet. The protocol self-adjusts to achieve the best possible service; routers send packets as fast as they can with the lowest error rate. And in socialistic fashion, all packets are treated equally, on a best-effort basis.
It wasn't very reliable, and still isn't. If there's a lot of congestion on a single route, packets are dropped -- not such a good thing for time-sensitive traffic such as telephone calls. But the original designers cleverly built in robustness. Because routers are simply dedicated computers, they can make sophisticated decisions about how to route traffic most efficiently. And when the network is congested or a link lost, they can find out about it and choose another route -- so that if a backhoe digs up a cable, or a fishing trawler cuts one in two, it is a problem, but not necessarily the end of the line. With network information and topology that is well-distributed, properly configured, and actively maintained, the Internet should be able to route around any central point of failure.
With the basic architecture and design principles already in place, the U.S. National Science Foundation (NSF) began funding data networking pioneers at 13 supercomputing centers across the U.S. in 1985. A nationwide circuit for the traffic was commissioned. The academic institutions had to strike deals with local telecom providers to lease local and regional circuits. More and more institutions sought to be connected to the NSFNET backbone, the network's main transport infrastructure.
Things moved quickly. As the network had grown, more and more uses were found for it -- but as long as the National Science Foundation ran the backbone, its Acceptable Use Policy (AUP) was the formal framework for what kind of traffic could run over its facilities:
"NSFNET backbone services are provided to support open research and education in and between U.S. research and instructional institutions, plus research arms of for-profit firms when engaged in open scholarly communications and research. Use for other purposes is not acceptable."
In 1991, three private IP networks
- General Atomics (CERFnet, now owned by MCI WorldCom),
- UUnet (now owned by MCI WorldCom), and
- Performance Systems International (PSInet) danced around the AUP by creating the Commercial Internet eXchange (CIX, www.cix.org), an open peering point for the exchange of network traffic.
The idea stuck. The Internet had proved its commercial viability; by 1992, the U.S. government wanted out, and the NSFNET's backbone transmission network was privatized
- and began to accept commercial traffic, marking what may have been the beginnings of the Internet as we now know it.
- Then, in 1994, the NSF commissioned four network access points (NAPs), essentially traffic exchange points similar in function to CIX, located in southern New Jersey, outside Washington, D.C., Chicago, and San Francisco.
All were run by different telecom operators. In a remarkably short period of time, the basic ingredients of today's global Internet had emerged.
Of course, an application from outside the traditional Internet community would dramatically shake things up. The World Wide Web, developed by Tim Berners-Lee and popularized around 1993, was soon followed by the Mosaic browser, forerunner to Netscape. The exponential growth the Internet had seen until then -- users and host counts generally doubled annually -- hit massive proportions and backbone traffic surged.
What You Pay Depends On What You Do
Back to the economics. Traffic, as we have seen, is routed over the Internet on a virtual pathway without fixed routes or network connections. The physical networks which make up the Internet -- typically leased circuits from telephone companies -- do interconnect, though. And networks do exchange traffic. The economics of how that happens are key to understanding who pays for what on the Internet.
That said, the Internet industry has matured enough to make basic distinctions between different categories of service providers or ISPs. Doing so provides, in part, the answer to how international infrastructure providers, such as telcos, are and will be compensated.
The generic term "Internet service provider" (ISP) has become meaningless. It does not distinguish, for instance, between large international ISPs (IISPs) with global infrastructure (such as MCI WorldCom or PSINet), or local online providers that bundle content with the access services they buy for their customers, such as EasyNet in Europe (www.easynet.co.uk), AsahiNet in Japan (www.asahinet.or.jp), or CAIS Internet (www.cais.net) in the U.S. Nor does it take into account whether the service provider's customers are individual users, who tend to request content; content providers who pay to export data; or other transporters of data. These differences can weigh heavily: unlike telecom finance, Internet cost recovery may involve compensation both for transporting bits, and for the kind of bits being carried -- content.
Better, then, to break down the industry into four classes:
- firms that specialize in Web site hosting,
- downstream ISPs who buy most of their long-haul backbone transit;
- online service providers which bundle Internet access with a focus on content and interface, and
- backbone ISPs.
The cost structure and the money flow is determined by the category to which one belongs. And because most players belong to different categories at different points in their activities -- even most backbone ISPs tend to be downstream from someone else -- those cost structures are complicated affairs. Taken together, though, the worldwide market for Internet access is now big business: it was to have grown from $25 billion in 1997 to more than $100 billion in 2000, according to Zona Research
Let's take a look at the components.
Content Hosting Providers
Web server "farms" emerged from the ISP industry itself, but are now somewhat separate: companies have made Web hosting into a niche business and are growing rapidly. The important fact is that their traffic flow is mostly uni-directional -- the antithesis to the bilateral, unmetered peering arrangements of old. Instead, the few bits of data that trickle in when a user requests a Web page are overwhelmed by the flood of outgoing audio, video, image, and text objects. As a result, backbone ISPs demand that hosting providers, which typically do not maintain a national network, purchase connectivity from a backbone or downstream ISP whose customers seek the content.
This can lead to conflict. Web hosting firms claim that backbone ISPs are already compensated by their end customers, so that to seek compensation from the content provider would mean a double payment. The backbone ISP counters that it is forced to haul the content provider's traffic on its own network to reach its customers -- and it wouldn't need so much infrastructure if the server farm had its own national network. Backbone ISPs thus only agree to accept a server farm's traffic at a price.
In August 1998, a peering dispute erupted between GTE Internetworking and Exodus over this very issue, and both firms' customers came close to losing direct connection to one another. Since then, the issue has continued to simmer. Though access providers still have the upper hand in negotiating peering agreements with content providers, the situation has evolved considerably, and negotiations now typically depend on the perceived value of the content to the access provider's customers.
A similar logic is used for downstream ISPs, who provide Internet access to even smaller providers, corporate customers, and end users. The price of Internet connectivity varies by location and amount of data. In late 1998, for example, a downstream ISP in Cambridge, Massachusetts could lease a 45 Mbps circuit for $2,500 per month. But that paid only for the facilities required to meet the gateway of an upstream backbone ISP. The price to connect with the backbone, which lets the downstream ISP's customers reach other destinations on the Internet -- this arrangement is usually known as "transit," as opposed to "peering" (see Figure 2, "Exchanging Traffic") -- can be as high as $30,000 per month.
While the connection fee may seem a crippling cost for U.S.-based ISPs, service providers outside the U.S. must also pay for the cost of an international private line if they wish to connect directly with the Internet at its core. Such a connection does not come cheaply -- trans-Pacific circuits, for example, were in 1998 going for as much as $60,000 to $80,000 a month for a 45 Mbps line.
However, most downstream ISPs and large corporate users that purchase Internet connectivity do not pay based on their actual usage, bit by bit, but based on a usage profile, broken down into different tiers. It would be too expensive and the tools too awkward to meter and charge every data flow. Indeed, many Internet engineers believe the cost of measuring and billing for exact usage could put a debilitating premium on Internet service. Lest the dilemma seem fanciful, consider the U.S. long distance telephone business. With coast-to-coast U.S. rates of $0.07 a minute or less, up to 40 percent of the rate for long distance telephony may reflect the costs of monitoring and monthly billing.
So on the Internet, the backbone ISP's network measures the overall traffic pattern by glancing at the router's bytes in and bytes out and charging the downstream ISP accordingly. This allows a customer to lease a line with much more capacity than is ever used, pay asum closer to the actual usage, and be assured that should traffic spike, the line can meet the demand for an additional fee. The only drawback with this approach is that it sets up an incentive for the upstream ISP to overbook capacity, under the hopeful (and reasonable) assumption that all customers do not generate peak loads at once. At MindSpring Enterprises Inc. (www.mindspring.net, now part of EarthLink), chief executive Charles Brewer said in 1999 that 30 percent of company costs derive from connectivity fees and 13 percent from customer service expenses. He expected a complete reversal within five years, as bandwidth becomes a commodity and ISPs must differentiate themselves more by the services they offer.
Online Service Providers
Online service providers like AOL earn revenues not by reselling network transmission service, but by bundled Internet access with proprietary content and specialized commerce, selling ads, and providing users with the ease-of-use which comes from special software interfaces and customer hot-lines. They are typically the customers of upstream access and backbone providers, which also manage the network points of presence (PoPs) accessed by dial-up retail users. AOL, for example, has outsourced nearly all of its network transmission needs to MCI WorldCom subsidiary UUNet, though it maintains a multi-vendor strategy to ensure redundancy. Even AT&T relied at one time on BBN's network, since acquired by GTE, to connect leased-line commercial users.
The online service provider is either paid a flat monthly rate by customers for unlimited service, or charges additional fees after a set usage is exceeded. The real payoff is in the eyeballs and the mouse-clicks -- selling specialty content and advertising space and taking a share of e-commerce revenue. Data networking represents online service providers' highest cost, apparently around 50 percent of revenue. Significantly, however, marketing, customer support and subscriber acquisition represent close to 35 percent of revenue.
All networks are beholden to backbone ISPs -- be they content hosting facilities, downstream ISPs, or online service providers -- either to furnish Internet connectivity or to manage the actual network infrastructure. Nor are backbone ISPs an exclusive category. While most network providers are increasingly specializing in specific segments of the market, there are only a very few network connectivity providers that aren't downstream from others. Internationally, the same dynamic applies. Local downstream ISPs in Asia, Europe and elsewhere need the larger, upstream networks, often the incumbent telecom provider, for Internet connectivity. And big Internet sharks outside the U.S. find themselves but tiny sardines when they arrive on U.S. shores with their leased circuit dedicated to IP traffic. They must strike an interconnection agreement with one or more of the Internet backbone networks, just like a regional U.S. ISP.
Some off-shore backbone ISPs have begun to acquire their own national U.S. networks to obtain free peering. Most have not. In 1999, Japan's NTT tried to aggregate its traffic with the large U.S.-based backbone ISP Verio, in which NTT had taken a ten percent stake, and Qwest was keen to piggyback onto EUnet International's peering agreements when the U.S. telecom upstart bought the pan-European ISP in March 1998, forming what would become KPNQwest. The emergence of backbone ISPs to provide connectivity to the Internet
- indeed to determine what actually constitutes Internet connectivity
- is a relatively recent phenomenon. Not surprisingly, there's controversy surrounding their role in the Internet food chain. When the Internet first evolved, ISPs were closer in size and swapped traffic freely. Early Net applications, like file transfer protocol, led to more or less symmetrical traffic among ISPs. In contrast, the Web creates a split between the end users, who import data, and content companies, who are data exporters.
That's new. In the early days, the ganglia of network interconnections were so complex
- since everyone accepted any other network's traffic
- that the only way Internet engineers could map the Internet's topology and traffic flow was simply to draw a cloud. Today, however, the terrific infrastructural investment, and major traffic imbalances due to the emergence of Web hosting firms, have meant that the practice of settlement-free peering is waning. The two noticeable exceptions are that local ISPs peer with their siblings at local exchange points, and that the very biggest backbone ISPs
- the Tier Ones who can move traffic pretty much anywhere without buying long-haul transit from someone else -- continue to peer among themselves. Peering, in other words, is for those who are your peers
- or appear to be.
The point: scale matters. Unless your network is very big and very fast and upgraded continuously, you are always somebody else's customer -- which is one reason why regulators, otherwise leery of interfering with the Internet's dramatic growth, have pondered stepping in to try and keep the backbone market competitive. Whether or not they do so depends on a still-brewing argument over whether or not Internet backbones, currently treated as enhanced services, should be reclassified as public telecommunications infrastructure. If so, then companies operating Internet backbones would be common carrier operators, and therefore required to provide fair and non-discriminatory terms to ISPs seeking interconnection. If not, then as enhanced services providers they continue to be able to pick and choose with whom they connect and on what terms. Like many of the hard questions about an increasingly pervasive Internet, the regulatory status of backbone networks -- basic telecom facility or enhanced ("private") data pipe -- remain unresolved.
2010 Greg's Cable Map is an attempt to consolidate all the available information about the undersea communications infrastructure. The initial data was harvested from Wikipedia, and further information was gathere by simply googling and transcribing as much data as possible into a useful format, namely a rich geocoded format. I hope you find the resource useful and any constructive criticism is welcome. The data is available in ArcGIS .shp file format on request, so long as it's not going to be used for profit.
Attack On Internet Called Largest Ever
The heart of the Internet sustained its largest and most sophisticated attack ever, starting late Monday, according to officials at key online backbone organizations. At the top of the root server hierarchy is the "A" root server, which every 12 hours generates a critical file that tells the other 12 servers what Internet domains exist and where they can be found. One rung below the root servers in the Internet hierarchy are the servers that house Internet domains such as dot-com, dot-biz and dot-info. The DNS is built so that eight or more of the world's 13 root servers must fail before ordinary Internet users start to see slowdowns. Vixie said it was an attack against all 13 servers, only four or five of the 13 servers were able to withstand the attack and remain available to legitimate Internet traffic throughout the strike.The root servers, about 10 of which are located in the United States, serve as a sort of master directory for the Internet.
2013 NSA Says It Can’t Search Its Own Emails: Lacks the Technology
There are actually some email server topologies where it is difficult (by design or coincidence) to search emails by content or sender. I'm talking about hub-and-spoke systems where a central server (or farm) deals with the outside world (the hub), and distributes email to servers that individual users connect to (the spokes). Certain entities have legal obligations for email communication privacy, e.g. not being able to send certain data to or from the outside world or only within their own systems. Law firms, some banks, stock exchanges, those are the ones that I know of for sure. Some of them do this by having rings of servers in addition to the hub-and-spoke architecture. So you have the central hub, and then it talks down the spokes to in multiple layers, and users that connect to servers at different layers of rings can have their senders/recipients checked against various policies, implemented in a number of ways technically. These systems are NOT easy to debug (I've helped fix some before), and depending on what you are logging at various servers on the way in and out, and for how long, it may not even be easy to search the "meta-data", and searching email content is not going to be easy. It will be progressively more difficult the more layers of rings you have in the system and how short the time logs are kept (sometimes such meta-data itself, due to traffic analysis, is considered protected or classified, and is not kept, or at least not on the servers as it would normally).
So no, they might not be lying, except when they say that it is due to their email technology being old. On the contrary, it would be because they are probably using methods that would provide the most secrecy and make centralized searching of email content or sender meta-data difficult by design. Of course they wouldn't admit to what their email architecture is, and that they probably have chosen increased secrecy in the trade-off between ease of use and security. There are of course some agencies that have classified information on them that DON'T use a system with such a design (the one I am aware of is the Whitehouse, and least in the past, from publicly available information on "bad things" people there have done that has come to light), but I know that at least one stock exchange (not users of it, but the exchange itself) uses such a complicated system and that searching its email data would be very difficult. Some domestic utilities use such a system and I know that at least 1 of the 3 largest banks in Mexico do. Those are just systems that I have first hand knowledge of from upgrading or fixing them. Note that "difficult" does not equal "impossible", but if an organization with such an architecture didn't set up a way to do it ahead of time, it would take work to do it (and if logs of sending pairs are gone, it'd take a lot of time).
Hubs and Spokes Email