Joho the Blog
An Entry from the Archives

Dr. Hossein Eslambolchi, AT&T's CTO, CIO and Fellow, and President of Global Networking Technology Servifces, is giving a talk at Harvard. Later, he'll talk informally at the Berkman Center.

Some notes:

State of the telecom industry: Coming out of a period of: Overcapacity, fraud, regulatory uncertainty, pricing pressure, brankruptices, competitive technologies. In late 2001, AT&T faced a "perfect storm" or "nuclear winter."

Top Ten Technology Trends

10. Ethernet everywhere. Home LANs proliferate. Bandwidth is the killer app. Just about everything will have Ethernet connectivity.

9. Knowledge mining will transform the way we do business — moving from information managing. [So long as I don't have to ever hear the phrase "wisdom mining"?]

Vendor dependency > Open networks, architecture and API

8. Wireless and wired lines will converge. Accelerating virtualization. Wire line communication will be history by 2020. Already, the number of wireless lines exceeds the number of wire lines.

7. Broadband will be the death of locality. When you get an IP-based infrastructure, geography means nothing. Martha Stewart's 212 number rang in her cell in Virginia.

6. e-Collaboration will dominate the workplace, enabled by speech recognition.

5. Sensor networks everywhere. We need lots more addresses. We need IPv6 which gives enough for every millimeter of the planet. E.g., cars will have IP addresses so you can have them tuned as you drive. (Hence, lots of sensors.)

4. "Wireless internet will be big." Moore's Law says that in 2010 we'll have 40mb/second. In 2020 we get 1gb/sec. [That seems unoptimistic. Do we really have to wait that long? LATER: David Isenberg has explained to me that Dr. E was referring to average broadband speeds, not maximums. Sorry!] We'll need quantum computing for this.

3. Convergence of communications and apps will be real: The network will be the computer. Most of the work will be done on the edges.

2. Security is critical. We need a better infrastructure [= not end-to-end?] or we'll have a virus hitting our computer every 5 seconds.

1. IP will eat everything. In 15-20 years, it will be application-based routing. [I don't understand that. Damn smart people!]

In 2010, we'll have self-healing networks. They'll have cognitive intelligence. [Smart networks.] We'll have cognitive radio, eliminating the need for FCC to regulate spectrum. [Yay!] Speech-to-speech translation.

2015: Network moves from hardware based to software based: on-demand, reconfigurable.

202: Last phone number will be retired because we'll all be wireless. Holographic storage. Tele-immersion. Holographic teleconferencing.

Future network direction

Now, we have pipes and ports. Once it's all IP, we'll have application-centric services.

Now we have individual networks > Then we'll have "converged collaborative network."

Users buy fixed capacites now > Buy it as an on-demand utility.

IPv4 > IPv6, multicast, unilink, VPLS [Over my head.]

Heterogeneity required > Heterogeneous by choice

Frame relay, ATM, IP > Converged IP-MPLS

Wired > Wireless, free space, wired

P2P > ebonding, network of things - Think about all the communities of interest you could pull together just looking at the attempts to place calls. [A little scary.]

"IP has eaten everything." [Yup.]

Dr. E gives an example of converged services over IP that sounds like a Semantic Web app, but I think he sees this happening by making the ntework itself smarter, rather than adding layers of metadata.

He says we need to handle different types of data differently via packet routing, e.g., to avoid jitter in VOIP. [Wouldn't increased bandwidth remove the need for this "optimization"?]

He likes Wimax as an alternative to T1/DSL/cable wireline access. It works up to 75mb/sec over 75 miles. By packing more bits per herz, we can get to 100mb/sec at home over the next 5-10 years. [I want 10 gigabits! Don't bogart that bandwidth, Dr. E!]

Dr. E says: The "end to end bigots" missed that it's impossible to scale that architeture when it comes to security. We need to build intelligence into the core of the network. Woiuld you rather have your firewall in one place or in billions of places? The center of the network has a global view. [Actually, I definitely don't want a centralized firewall. Very very dangerous politically and, I'm guessing, actually less secure because there's a single point of infection.]

He ends by saying he's been fighting the forces of control and fighting for the forces of freedom.

From passive recipients to active participation

From newscasts to blogcasts and personal studios.

Proprietary solutions > Standards bodies

Proprietary sw > Open Source for greater reliability

Closed control > Open control [Say more!]

Licensed spectrum > Unlicensed spectrum. either more unlicensed and/or cognitive radios

Regulated access > open access [Say more!]

Ending joke. Some say, "If it ain't broke, don't fix it." At AT&T Labs we say, "If it ain't broke, it ain't got enough features."

Q&A

Q: What about RBOCs and VOIP?
A: The future is IP. But Skype is like a toy. About 6.25% of traffic will be voip to voip [assuming - if I got this right - we assume that 25% of calls start with voip and 25% end with voip, leaving 50% going over the plain old phone network].

Q: How about the weaknessdes and vulnerabilities of non-end-to-end architectures? What will that do to innovation?
A: We need both. E.g., intruder detection systems are only 95% accurate. Firewalls at the edge of the network haven't worked sufficiently and hackers will target your particular weaknesses. So you have to catch this stuff in the network. You need a lot of sensors to be able to catch it, and that has to be done in the center. We have 150 terabytes of traffic (a day?) and we can scan it in 15 seconds. Yes, this may impede SSL handshakes. That's why we need to rethink the network. We could collaborate with Harvard to redesign it.

Q: What data can you keep? You seem to be equating traffic with data...
A: Look at just the voice traffic. About 400M call attempts/day. We don't record the voice data, just who you called. From this we build community of interests. Now, take that model to an IP basis. If you 2.6 petabytes per day, it's too much. So we have a unique sampling algorithm that looks at the header. (We can't look at the content for privacy reasons.) We save the voice data as long as 6 months; I won't go through the detail because of security. Some data we store as much as 7 years.