Future Directions for Internet Researchers Panel

The Sally Floyd Channel

Sponsored by the NeTS-VO

January 10, 2022

8:00AM‎ – ‎9:30AM Pacific Time, 11:00AM – 12:30PM Eastern Time

Main Speakers: Vinton Cerf, Alberto Leon-Garcia, Jennifer Rexford, Amin Vahdat

Please use this link for registration: https://docs.google.com/forms/d/e/1FAIpQLSewSX04kNrH6dI0VaxyOv8O9TEYybV-xAAd-krTM7QqaTIVKg/viewform?usp=sf_link

About the Speakers
Vinton Cerf

Vice President and Chief Internet Evangelist

Vinton G. Cerf co-designed the TCP/IP protocols and the architecture of the Internet and is Chief Internet Evangelist for Google.   He is a former member of the National Science Board and current member of the National Academies of Engineering and Science and Foreign Member of the British Royal Society and Swedish Royal Academy of Engineering, and Fellow of ACM, IEEE, AAAS, and BCS. Cerf received the US Presidential Medal of Freedom, US National Medal of Technology, Queen Elizabeth Prize for Engineering, Prince of Asturias Award, Japan Prize, ACM Turing Award, Legion d’Honneur, the Franklin Medal, the Catalunya International Prize and 29 honorary degrees.

Alberto Leon-Garcia

Alberto Leon-Garcia is Professor in Electrical and Computer Engineering at the University of Toronto.  He is a Life Fellow of the Institute of Electronics an Electrical Engineering “For contributions to multiplexing and switching of integrated services traffic”. He authored the textbooks: Probability and Random Processes for Electrical Engineering, and Communication Networks: Fundamental Concepts and Key Architecture.  Leon-Garcia was Founder and CTO of AcceLight Networks in Ottawa from 1999 to 2002.  He was Scientific Director of the NSERC Strategic Network for Smart Applications on Virtual Infrastructures (SAVI), and Principal Investigator of the project on Connected Vehicles and Smart Transportation.  SAVI designed and deployed a national testbed in Canada that converges cloud computing and software-defined networking.  CVST designed and deployed an application platform for smart transportation. Leon-Garcia was Founder of StreamWorx.ai which developed massive-scale, real-time streaming analytics software for network operations and cybersecurity applications.

Jennifer Rexford

Jennifer Rexford is the Gordon Y.S. Wu Professor of Engineering and the Chair of Computer Science at Princeton University. Before joining Princeton in 2005, she worked for eight years at AT&T Labs–Research. Jennifer received her BSE degree in electrical engineering from Princeton University in 1991, and her PhD degree in electrical engineering and computer science from the University of Michigan in 1996. Her research focuses on computer networking. She is co-author of the book “Web Protocols and Practice” (Addison-Wesley, 2001) and co-editor of the book “She’s an Engineer? Princeton Alumnae Reflect” (Princeton University, 1993). Jennifer received the ACM Grace Murray Hopper Award for outstanding young computer professional, the ACM Athena Lecturer Award, the NCWIT Harrold and Notkin Research and Graduate Mentoring Award, the ACM SIGCOMM award for lifetime contributions, and the IEEE Internet Award. She is an ACM Fellow, an IEEE Fellow, and a member of the American Academy of Arts and Sciences, the National Academy of Engineering, and the National Academy of Sciences.

Amin Vahdat

Google Fellow and Vice President, Systems & Services Infrastructure

Amin Vahdat is a Fellow and Vice President of Engineering at Google, where his team is responsible for Engineering, Product Management and UX for Compute (Google Compute Engine, Borg/Cluster Scheduling, and Operating Systems), Platforms (TPUs, Accelerators, Servers, Storage, and Networking), Network Infrastructure (Datacenter, Campus, RPC, and End Host network software), Cloud Networking (Google Compute Engine, NetLB, Google Private Cloud ), Storage (Filestore, Google Cloud Storage, Backup and Disaster Recovery, and Transfer), and the Systems Research Group. Until 2019, he was the Area Technical Lead for Networking at Google, responsible for Google’s Technical Infrastructure roadmap in collaboration with peers in Compute, Storage, and Hardware. Vahdat is active in Computer Science research, with more than 41,000 citations to over 200 refereed publications across cloud infrastructure, software defined networking, data consistency, operating systems, storage systems, data center architecture, and optical networking.

In the past, he was the SAIC Professor of Computer Science and Engineering at UC San Diego and the Director of UCSD’s Center for Networked Systems. Vahdat received his PhD from UC Berkeley in Computer Science, is an ACM Fellow and a past recipient of the NSF CAREER award, UC Berkeley Distinguished EECS Alumni Award, the Alfred P. Sloan Fellowship, the SIGCOMM Networking Systems Award, and the Duke University David and Janet Vaughn Teaching Award. Most recently, Amin was awarded the SIGCOMM lifetime achievement award for his groundbreaking contributions to data center and wide area networks.


Panel Chat

08:15:52 From  Gary Minden : How do you separate networking research from the web? 

08:17:56 From  Ankit Singla : To Gary’s question: should we? 🙂 I would also love to hear the application-side of the evolution, e.g., the “web3” folks are making a lot of noise? Is it just that (or worse, entirely a sham/scam/ponzi) or do folks here see value in this? (IMO, applications should be our concern in networking research!) 

08:20:06 From  vint cerf/google : I wonder to what extent layering can help us with the dynamics of evolutions? 

08:20:20 From  Audrey Randall : +1 to Ankit’s web3 question 

08:20:40 From  Henning Schulzrinne : Most of the outages recently seem to be related to automated behavior going awry. What does this say about increased calls for use of ML in networks, where rare conditions drive outages? 

08:21:12 From  Anees Al-Najjar : Question to Jennifer, can we push the network programmability functions to HPC for better performance? 

08:21:26 From  kfall : fwiw, it strikes me that one of our biggest issues is the veracity of the information our networks carry.  In some cases, hints of this are derivable from lower level information (including network operations level info).  Do we have a responsibility to help this issue [?] 

08:21:33 From  Gary Minden : Ankit: Then we might separate our research into networking and social networking? 

08:22:19 From  Paul Gleichauf : The conversation about evolving/replacing the Internet to be reliable, programmable and secure has been ongoing for decades, but it seems that evolution vs. replacement efforts have been slow as it has become more pervasive. Is the infrastructure to change the Internet to meet these challenges adequate to get there or is something more needed? 

08:22:40 From  Gary Minden : Amin:  The Internet started as peer-to-peer. 

08:23:00 From  Ra yavatkar : Jen and Amin – to Vint’s point about why Internet was built to be best effort, I wonder whether building programmability inside the network would compromise that aspect which has made the Internet most functionally robust architecture to survive. Thoughts? 

08:23:57 From  Dave Taht : https://moxie.org/2022/01/07/web3-first-impressions.html 

08:24:29 From  H B Acharya : “Functionally robust” as in “cheap and simple”? (better is worse, a simple network with dumb pipes was easy enough anyone could join) 

08:24:46 From  Ra yavatkar : Building trust is a part of making BGP and DNS less fragile which they are IMO. Do you think blockchain based distributed ledgers could help here? 

08:25:38 From  Dave Taht : web3 rrinventing artificial scarcity and actually centralized in all but name. 

08:25:41 From  H B Acharya : @Ra : why would you prefer this to existing DNS-over-TLS etc.? is there any advantage to keeping a whole immutable history? 

08:26:33 From  DJ : are there plans for the next gen Internet, e.g. Internet 3.0?  secure at every layer, ipv6, 6g, edge, account for enormous video traffic volumes, etc. 

08:27:36 From  Gary Minden : DJ: Yes, thought about it about 25 years ago. 

08:28:05 From  H B Acharya : Peter Steenkiste is building one – XDR 

08:28:52 From  H B Acharya : *XIA sorry 

08:29:18 From  DJ : the need is NOW – with major xxMUSD hacks on weekly basis, the risk is for folks to soon stop trusting the Internet. 

08:29:19 From  jeff tanrsura : @HB – for BGP, for sure( rather than putting together 10 different feeds that emulate it) DNS – not so 

08:31:03 From  Hongwei Zhang : What is the panel’s take on the potential strategies we can leverage to bring URLLC-type safety-/mission-critical services beyond RAN and as far as possible into the backbone/wired-Internet to enable a broader set of applications feasible? To what extent are some of the intellectual learnings from the 90s applicable (algorithmically) or not? 

08:31:36 From  Hongwei Zhang : And implications to network architecture, programmability, and innovation paradigm? 

08:33:50 From  H B Acharya : @Dr Rexford – you did this too right? MIRO? 

08:36:53 From  vint cerf/google : i deliberately tried to ignore the specifics of the underlying networks to allow for multinetwork operation 

08:37:01 From  Dave Taht : dtn – what applications are needed for some useful work (and debugging) here on earth? what mental  models need to change? 

08:38:26 From  Dave Taht : is there some way to have a fire department for the internet?  https://xkcd.com/2347/ 

08:38:58 From  Ankit Singla : to Vint’s point of allowing multi network operation, I was quite surprised to find that one of the early networks (among the “multi networks”) supported in the early design was the SATNET *satellite* network. 

08:39:08 From  Tony Tauber : So much of the action and control-plane for “the Internet” (vs. generally networks) is actually in the DNS.  Seems not so much research and knowledge-sharing has been focused on understanding it’s structure, behavior, and changes over time. 

08:39:47 From  Henning Schulzrinne : I see a lot of student enthusiasm shifting to areas other than systems (OS, networks), particularly in the US. Is this a concern for the long-term viability of the research community? 

08:41:00 From  DJ : what is the role of quantum computing long-term evolution on security?  are there viable alternatives to current one-way-functions-based algos? 

08:41:45 From  Vividh Siddha [Apple] : The pandemic proved that the internet stood the test of scale, reliability and availability. Most ISPs and providers were running at unprecedented capacity and managed to keep the internet stable. That has been great, but its 2022 and the video conferencing or video calling experience is still not great. Can-you-turn-off-your-video has been a common phrase used on video calls. With more demanding workloads such as XR (real-time communications) coming up what are your thoughts on making this great. 

08:43:09 From  Gary Minden : VC: Yes, but at the same time you abandoned the technical details of the basic internet. 

08:43:32 From  Jason Livingood : Re video conf… It’s all working latency (bloat)… Something folks are not measuring extensively nor discussing much. 

08:43:41 From  Srujan : With Nick Meckeon in intel and acquisition of barefoot, do you think custom protocols are going to get new traction ? 

08:44:40 From  John Grant : @Vividh Siddha: ETSI ISG NIN is working on a better experience for live audio/video/etc https://www.etsi.org/technologies/non-ip-networking 

08:45:13 From  Dave Taht : thx vint. fq-codel needs to move Into more offloads AND we need ongoing passive insight into the behaviors of flows. 

08:45:49 From  John Grant : PS “non-IP” actually means “not restricted to IP” 

08:46:12 From  Dave Taht : (I’m rooting for Nick to get traction in Intel too!!) 

08:46:52 From  Jennifer Rexford : The telephone network? Hmmm, that rings a bell… 🙂 

08:47:08 From  Katie Wilson : 😉 

08:47:11 From  Ankit Singla : BELL, you say? 

08:47:24 From  Victor Liu : Any comments on AI/ML?  AI/ML in many fields has been progressed significantly, and network operation should be something similar to the chip industry to use ML, but so far not much yet. We have network verification using Intent based networking similar to formal verification for chip industry. We do not have ML based failure detection for network operation yet but might be something can be done… 

08:48:09 From  Gary Minden : What if the Internet is the next scaffolding? 

08:48:23 From  Dave Taht : I’m dying to hear more about DTN!? 

08:50:11 From  Anita Nikolich : next Internet: needs content that’s not controlled by a few commercial entities. i think students are more attuned to this these days. unclear if this can be achieved. 

08:50:25 From  John Day (USA) : Vint is right. CCN is like IP before CIDR and a lot more than 32 bits. 

08:50:31 From  Ankit Singla : Ion Stoica et al. had an intriguing HotNets talk awhile back, that argued that HTTP _is_ content centric, and is (sorta) good enough 🙂 

08:50:52 From  John Day (USA) : CCN fundamentally doesn’t understand routing 

08:51:41 From  Adam Waters : It’s a good paper by Ion: HTTP as the narrow waist of the future internet


08:52:12 From  Ankit Singla : ^ that’s the one, thanks! I quite liked it 🙂 

08:52:56 From  John Grant : Lots of folks are happy to have Alexa listening in on all their conversations 

08:53:13 From  Rosa Zheng : Why do we need 5 layers? Would 3 layers sufficient? 

08:54:43 From  jeff tantsura : We hit 30% with BGP RPKI, there’s a good progress! 

08:54:48 From  kfall : Brylcreem? 😛 

08:55:15 From  Audrey Randall : People have tried distributing the root of DNS using blockchain. The end result was that if there is no central authority that can take down abusive domains, abusive domains tend to predominate. 

08:55:20 From  Hesham : @Rosa, I think 3 layers are enough. 

08:56:36 From  Dave Taht : theoretical question, why can’t Google route all it’s quic traffic through 8.8.8.x anycast? 

08:56:50 From  Alefiya Hussain : The “HTTP  as the narrow waist” is now 11 years old… do you think it will continue to stay true? 

08:57:09 From  Xiaowei Yang : We, as a community, have invented many innovative mechanisms.  I sometimes feel the lack of incentive to deploy them is the bottleneck to innovations. Providing incentives is more or less or policy or human issue. Not sure how to address that. 

08:57:16 From  Terry Benzel : There is lots of focus, rightly so, on social networking and disinformation. But in the end all of this does run on a network, moving bits so what can we do at the network layer to address all of the social networking and disinformation 

08:58:06 From  Ankit Singla : what kind of blind spots do the “most prestigious” venues for networking have in research? e.g., I have never seen a paper at SIGCOMM/NSDI that looks at say … security/privacy vulnerabilities for users with disabilities (given that our methods of user security on the Web are often visual elements). 

08:58:33 From  Giri (Nvidia) : Surprised Balkanization of Internet is not discussed. 

08:58:56 From  vint cerf/google : we will get there – stand by. (re balkanization) 

08:58:58 From  Anita Nikolich : under studied: data. use and sale of personal data, data ownership. (Vint mentioned data sovereignty) 

08:59:58 From  Henning Schulzrinne : A lot of the privacy research has been about “more encryption”, which fundamentally misunderstood the problem. 

09:00:02 From  Stuart Cheshire [Apple] : I’ll echo Vint’s comment about the importance of fixing bufferbloat. In terms of the viewing the current Internet as the scaffolding for what comes next, we can *add* new capabilities layered on top of an existing Internet, but we cannot *remove* delays occurring in the underlying network. If the underlying network imposes delays (in the hundreds of milliseconds, or more) before it delivers a message (because the message spends an excessive time sitting in an oversized buffer) no higher layer can do anything to remove that delay after it has occurred. 

09:00:06 From  Dave Taht : Ankit, why are we still publishing “papers”? why are we not refuting and building on research dynamically 

09:00:20 From  Anita Nikolich : ⬆️ 

09:01:02 From  vint cerf/google : +1 Stuart – nice to see you on the call 

09:01:03 From  Ankit Singla : I don’t disagree with that at all, Dave. In fact, papers are pretty poor even just interface-wise! 

09:01:57 From  John Grant : @Stuart: Wasn’t a problem with ISDN. Maybe some babies thrown out with the bathwater when switching to best-effort packet nets. 

09:05:22 From  Dave Taht : (selfishly I am dying to hear of future directions and the state of the art in dtn) 

09:05:58 From  Mohammad Faghani : Security used to be an afterthought. Do you think the new technologies and evolutions are considering security by design? 

09:05:59 From  Mark Kuehner (SAP) : @Vint What are your thoughts in regards to sovereignty and control.  On one hand we want the freedom of the net, on the other hand we have trends that countries try to get control for several reasons ( dependancy, technical, political)  or also companies try to get over it ( technical). 

09:06:17 From  Dave Taht : I’m willing to go down that dtn congestion rabbit hole with someone. 

09:06:53 From  vint cerf/google : @dave, we should discuss offline with IPNSIG.ORG team 

09:07:01 From  LIXIA ZHANG : Agree with Vint that crypto is a basic tool that should be used everywhere to serve security purpose (and to Mohammad comment: yes it needs to be designed into, instead of being afterthought) 

09:07:22 From  vint cerf/google : Hi Lixia! great to see you on the chat. 

09:07:38 From  Stuart Cheshire [Apple] : @John Grant You raise a good point, which has a lot of depth. With ISDN you had a monolithic homogeneous path end-to-end, the data flows through smoothly. That made things easy, but it was very slow and expensive. The insight of the Internet is to make a heterogeneous network — a mix of technologies with different data rates. But when you have different data rates you need buffers at the transition points to smooth out traffic bursts, and that’s when smart buffer management becomes crucial. 

09:07:45 From  Ankit Singla : (So good to see so many other networking idols here in the chat! e.g., Lixia) 

09:08:19 From  Jennifer Rexford : https://en.wikipedia.org/wiki/Unsafe_at_Any_Speed 

09:08:23 From  Gary Minden : AI/ML is bogus 

09:08:30 From  Dave Taht : Larry Niven once wrote a story called “safe at any speed” which is also worth. thinking about. 

09:08:45 From  vint cerf/google : +1 Stuart – significant variability of demand creates potential for congestion 

09:09:55 From  LIXIA ZHANG : Vint: can’t afford miss such a great panel:-) 

09:11:23 From  John Grant : @Stuart: In ETSI ISG NIN we’re standardising technology that doesn’t assume fixed bandwidth (as of course ISDN did) 

09:11:34 From  Paul Gleichauf : Can we do a good enough job managing tradeoffs—for example security/cryptography have computational and storage overheads that produce delays, so does policy-based management and network visibility, etc. Does a future Internet and research on getting there have to make tradeoffs explicit, or should can the goal be to have it all? 

09:11:35 From  kfall : I’d think that with (~link) admission control you have the ability to move where that congestion may occur.  So, in a portion of the high latency e2e path you may still have low latency portions for which reactive cc could be achieved.  How that integrates with the overall path (and, say, scheduled connectivity) is probably open, but not untought of. 

09:12:31 From  Giri (Nvidia) : Unfortunately, Web3 name is hijacked by Crypto folks. Web2 where companies aggregate data is not scalable. What will enable True-Web3 where data is decentralized and Companies provide services? 

09:13:43 From  Ankit Singla : ^ I have been thinking about how much networking folks should engage in that (Web3) conversation. Right now, it seems too suboptimal, leaving it entirely to the web-dev folks. Ultimately, applications do (and should!) drive where networking goes as a discipline at least in part. 

09:13:45 From  Victor Liu : We think treating network snapshots as pictures are the key to adopt computer vision or NLP techniques into network management.  Will try to see if we can have something like ML research field to have the SOTA – state of the art type of comparable research datasets for us as a community to grow that field quickly.    

I do agree statistics part of ML, which is not exactly what most protocols are built at accurate system. It is exactly similar situation to the chip industry where design are high-9s accurate but Computer vision and ML are also used in the production and manufacture fronts where errors are happening and requiring control. 

09:13:59 From  John Day (USA) : ML can only react to what it has seen. If it sees something entirely new. It will assume it is something it has seen which can be very wrong. 

09:15:04 From  Victor Liu : We recently use IBN startup tools to replace ping for network flow reachability tests 

09:15:48 From  Shivendra Panwar : @John Grant Circuit switching is still a solution fig you are willing to waste some bandwidth, which may be an acceptable cost. 

09:16:28 From  John Day (USA) : Not always. Some decisions are irreversible 

09:16:45 From  Shams Jawaid : My question for the panel on security – how do you best deal with the rise of something like ransomware? 

09:16:50 From  John Grant : We use all the unused bandwidth for the best-effort traffic (unused, not just unallocated) 

09:16:51 From  Victor Liu : Working at a Sat company, many assumptions that satellite links can be our backup for terrestrial links might have to be revisited… 

09:16:58 From  Vividh Siddha [Apple] : @ankit that’s right. Every time a application layer treats the network as a layer to be used does not end well. Look at the evolution of http. So yes, there needs to be a active engagement with the web3 folks (whoever they are). 

09:18:53 From  Dave Taht : awesome question 

09:19:25 From  Allison Mankin : It strikes me network innovation just gets deployed and tested at scale under pressure of external conditions (pandemic, industry changes, outages) – what’s a good learning of yours from a “battlefield” experiment…I like the scaling of video and the rapid addition of security features as people got attacked early in the pandemic 

09:20:03 From  John Day (USA) : No, it doesn’t 

09:20:09 From  Dave Taht : I’d add globally synced timestamps and a flowid at l2. aloha almost had ttl right. 

09:21:53 From  Shakil  : Hi, Not sure if this is applicable topic here but if you could kindly answer. How quantum computing in cloud will affect the application (workloads) landscape, would all need to be refactored and rewrote? Or will quantum computing be specific for some applications only, not so for general purposes. 

09:21:54 From  Jennifer Rexford : https://mitpress.mit.edu/books/designing-internet 

09:22:20 From  vint cerf/google : i would like to respond to the quantum question if we hafve 

09:22:21 From  vint cerf/google : time 

09:23:10 From  LIXIA ZHANG : As Vint mentioned earlier: today’s internet could be used as scaffolding a future design 

09:23:16 From  Dave Taht : virtualization adds latency. I’d like cpus that could context/priv switch in five clocks and the. rethink about virtualization… 

09:23:23 From  Reza : Giri’s question about Web3 and decentralization is very interesting. I’m waiting to see what the answers to that question are. 

09:23:33 From  LIXIA ZHANG : IP was a virtualization itself. 

09:24:01 From  Paul Gleichauf : One needs to be careful analyzing how decentralized blockchains really are. 

09:25:04 From  LIXIA ZHANG : The panel asked a critical question earlier: what is the foundation of trust for Web3? 

09:25:33 From  Dave Taht : @paul see Moxies post I linked to earlier… 

09:25:36 From  Paul Gleichauf : Quantum networks don’t copy state, so they will be quite different from classical networking. 

09:25:40 From  Dave Taht : faith. 

09:26:22 From  LIXIA ZHANG : Agree with Vint; my comment was a response to Reza. 

09:26:46 From  kfall : (but trust isn’t a binary or static thing) 

09:27:02 From  Shakil  : Thanks, meaning applications need to be refactored and rewritten 

09:27:16 From  Shakil  : In quantum computing 

09:27:27 From  Paul Gleichauf : Trust from untrusted components is at the core of blockchain design motivations. 

09:27:35 From  Dave Taht : tailscale is doing some interesting work on scaling trust from the bottom up 

09:27:37 From  vint cerf/google : PGP didn’t scale very well unfortunartely 

09:27:38 From  Mazin Yousif : You can always assume zero trust and establish trust at a later stage 

09:27:39 From  Mark Kuehner (SAP) : Mean question: Assume trends continue , traffic is processed by large cloud providers, most likely E2E in the future (5G, satellite),  governemnts  try to control the the traffic.  Is the internet, how we know it today, still relevant in 10-20 years? 

09:27:54 From  vint cerf/google : yes to zero trust idea 

09:28:06 From  LIXIA ZHANG : Signing parties to get authentication. Trust can then be built on top of authentication. 

09:29:39 From  Dave Taht : please share! (the internet is a copying machine) – John Perry barlow 

09:29:47 From  LIXIA ZHANG : Great panel, thanks for organizing it! 

09:30:00 From  Audrey Randall : Thank you all! 

09:30:22 From  jeff tantsura : Thanks everyone, great panel! 

09:30:35 From  Shakil  : Thanks a lot, great panel and host! 

09:30:44 From  Morteza Kheirkhah (IDE) : Great panel. Thanks for the amazing dicussions! 

09:30:52 From  LIXIA ZHANG : Need to teach students “architectural thinking” 

09:30:58 From  David Hay : Thanks a lot! 

09:31:00 From  Victor Liu : Industry practitioners should post and publish their problems for academics to jump into the solution competitions.  That how we survive 

09:31:12 From  qiang Guan : Thanks 

09:31:20 From  UC3M – Víctor P. Gil : Thaanks