This is one long post here…
The day started with Professor Gaetano Borriello, from University of Washington and his talk entitled “Invisible Computing.”
Prof. Bariello started with the classical Weiser’s quote “the most remarkable technologies are those that disappear”, but remarked that currently computers are not really disappearing. He stressed an important aspect in the metaphor of the disappearing computer: it is not about the “visually” disappearing, but “cognitively” (you can still see the computer and know it works fine, but his role is to assist you and not to frustrate you). Also, he presented the Labscape project, which is a system for automatic work capture for cell biologists. In science, methodological procedures and research results are hard to annotate, to index, and to share, as there is no “standard language” to formally describe them. Automatically capturing the methodology and organizing the data of experiments, search all that, would be extremely helpful to share data and eventually automatically write documentation about the experiment. Labscape is here to assist the researcher in doing science and not spending time on meta-science (sensors on what the guy is doing, how much liquid is poured where, etc). It’s visually visible, but not cognitively, however visibility is still necessary (need to know it’s there and working).
Also, he presented the RFID ecosystem, which is an large scale experiment of the pervasive use of tagging (people and objects are both chipped). The goal was to create a microcosm of a world saturated with identifiable objects, 160 tag readers, and thousands of tags (every interaction is recorded, all read events are stored in a DB specially made for users, with probabilistic event detection, particle filters for location estimation). Twitter was also used to track last location and movements of colleagues. They also setup digital diaries for when/who/where/how long you meet. Challenges: noisy sensors (water in the body affects, data management, and security). Limited scenarios only for now in the lab, not real deployments.
He also discussed the panopticon - asymmetric visibility (guard tower in the middle of prison and everybody can be seen). Institutional privacy: what is socially appropriate. Physical access control is used, where only viewable things that anybody else could see anyway can be accessed (from your physical presence) – no superman Xray vision is allowed by the system. Tag IDs are sent in clear, provenance trail (they can revoke/delete anything anytime, incl. the composite events), transparent queries. He also discussed about Joshua Smith from Intel who is working on adding accelerometer and light sensors to passive tags, which can sense without any external power (which would be cool to know if a chicken has been exposed to dangerously high temperatures).
Philosophy of both projects: future world technologies will really disappear, process of gathering many bits of created data/interactions. Not physical invisibility, just cognitive! For now, benefits are only small tricks and hacks that can significantly change little things in our lives, but it is still too early for seeing large benefits.
Two cities in China are doing exactly that: monitoring people (he didn’t mention which city). Also can be used to detect anomalies in mobility patterns (2 people not moving for long, etc…).
Afterwards, Prof Andy Hopper FRS from Cambridge University presented his talk.
How can developing countries become successful without using too much resources (as we did), computer tool for enabling improvement in developing countries.
Optimal digital infrastructures
Provisioning appropriate availability: redundancy of data centers doubles costs, so develop new tools to reduce the redundancy. E efficient computing: adaptive datacenters. Scale energy use with useful work done at all levels. Develop principles: switch off if not use, don’t send data if not wanted, know where traffic is coming from, use tech that linearly scale energy (PROPORTIANAL consumption to task). Servers consume 50% when idle, and as server are sometimes idle for a long time it’s quite bad (Energy balancing in cloud computing – start/stop more or less servers based on predicted load, XEN virtualization, jobs are moved around in 250ms, non-interactive jobs are delay-tolerant). Virtual battery something spatial about computing tasks, can be virtually done anywhere in the world, where energy is available and cheaper (because it’s cheaper to “ship” the task anywhere else) (quote article for supercomputers in wired) - > what granularity should be the jobs, what to ship? Data, program, or both? I like the idea of “shipping/outsourcing” computing. Where to put the farms? Look at wind maps to use energy that would be otherwise lost. Global goal: energy proportional computing/communications at all levels. BAN THE WORKSTATION, because it’s very inappropriate.
Prof. Hopper.
Sense and optimize
Sensor based digital model of the world: Googling earth, space-time. Sensing: already many sensors, and people publish/share their data (incentives to do so). Storing: create a global repository, data & comp models. Indexing. Sensing indoors location (using beacons and fixed infrastructure) or outdoors with cars & mobile phone. Mapping 3D cell reception. Human sensing. Reward for content creation enticing and wealth creating for developing world?
Thermal maps, personal energy meters. Collect info about individual consumption. Dilemmas: who to trust, of values, governance… Also, which surveillance scenario will prevail?
Predict and react: prediction of travel times and routing. Why a particular route is chosen? Gives several choices (www.camvit.com).
Global computing standards for digital alternatives to physical activities: growing tendency to move bits rather than people and product. Do webcams make us travel less, or more? Can we create virtual worlds where we can conduct our lives?
Control through price, where you empower people, and they can make choices to spare money. How much computation is really “necessary”? Evidence: the more computing is there, the more energy is used. Having more computers is really actually going to save energy?
Professor Margaret Martonosi, from “Zebranet and beyond: exploiting the unique characteristics of mobile sparse networks“.
All computing is mobile, distributed, as lots of devices became computers. So we have now a heterogeneous network of different computing platforms. How to program/manage them?
Prof. Martonosi.
She presented the Zebranet project, that focuses on analyzing the interactions within species and between species, and to measure the impact of human development on animal behavior. In Kenya, there is no communication infrastructure. Triangulation with VHF signals to locate animals is good but not really robust. Also, you have to be out actively monitoring. On the other hand GPS devices are totally restrained (either local storing of data or sent, neither of which is optimal). So what to do when no good infrastructure is available? Mesh p2p data forwarding!
Mobile sensor network, not just a fixed mesh, but by encounters between nodes, for sparse networks. This requires a very different protocol/system design, including custom hardware and software that allows reprogramming of nodes at runtime through viral propagation to the neighbors. Data compression (very aggressive) and look at tradeoff between compression and communication costs, data abstraction layers for fast communication & queries.
Efficient non-GPS localization: why does it matter? Localization is essential for many applications, eg. social networks to locate your friends, geographical routing for optimized communication, spatially aware computation, etc. GPS is expensive and power hungry, so she proposed a collaborative localization algorithm called LOCALE (ISPN 08 Zhang and Martonosi) which emphasizes low power, accurate low device cost, low infrastructure, etc. Also, how can the aggregation of multiple position estimates increase the confidence in localization results, tracking position between encounters.
Each node keeps an idea of where it is with a cloud of confidence. If devices have high confidence, then they can reduce the cloud size. Use fixed beacons with GPS and accurate location, and they can share their info with neighbors. Collaboration is a powerful mechanism for improving accuracy in localization.
Then what if we replace the zebras with people? Store and forward to replace phys, wired infrastructure. SARANA: system to support collaborative, low-infrastructure computation
SA - Spatially Aware where are services located? New languages are being developed for location aware computing (SpatialViews) The language allow the programmer to create quality assessments as well, RA - Resource Aware, NA – Network Aware.
Sparse mobile networks are here and growing, time sync is still a big issue when you don’t have cheap access to a global clock (as the GPS would provide).
Finally, Professor Timothy Roscoe, from ETH Zurich closed the morning session with his talk entitled “Network architecture for ubiquitous computing”.
Prof. Roscoe
Obstacles to progress: past - what the internet is, present - internet problems, roadblocks for progress and a research agenda, future - new architecture, and implications for ubicomp.
Very few ubicomp, long-lived, or large deployments have been made, and only few of them addressed real-world real challenges such as heterogeneity, scalability, and evolvability. Besides it is very difficult to evaluate these systems. Unfortunately, there’s not much motivation for these issues in Ubicomp.
IP is our “universal” protocol, end-2-end arguments, and edge vs core. There are things it supports but it shouldn’t (spam), and the other way around. The net was not designed with a security or resource control of mind for example, and there are many other things the net doesn’t also have, and this should be a research opportunity! But how to know if your idea works, because usually the problem is bigger than your lab network? How can anyone else reuse your idea? According to a report in 2001 (US NRC), the Internet is too hard to change and too important to change (“ossification” is the elegant term he used for that) because of the impact a failure would have. A suggestion was to use overlay nets, but this leads to other problems: how to deploy, evaluate, and access overlay networks?
Large scale internet services: akamai, google, with many geographical locations. P2P nets: millions of nodes, increased variety of applications. Lots of simulations, but who believes them? And then you also need to validate against something. Emulation on large clusters has also be done, but would you really meet the real-world problems?
Lots of small scale deployments. A meeting for building a collective platform (march 2002 - paper came out blueprint for introducing disruptive technology into the internet), remarkable consensus was reached, community-built platform, address both issues: enable a real wide-area distributed systems tesbed, and deploy new nets overlays with real users. Key ideas: slices! Like Wuala, you give some, get some. PlanetLab had huge impact and now is made of over 800 machines throughout the world! It even changed the publishing climate by raising the bar of validation, and no more excuses for lack of deployments. You can virtualize links as well as servers! GENI.net large-scale project – for research for the future of the internet. FIRE FP7 program in EU. Now the problem: how do you share resources between multiple users, with resource guarantees, over short timescale, and securely?
He suggests that it is time to rethink two basic principles of the internet: core vs. edge applications increasingly resemble overlays and links are more important than the end-2-end (also often there are more than one end…). Is really a new network that we need? Or rather a different approach when using the same, existing infrastructure?
Key ideas
- Low-level resource provisioning VM, links, radios. etc
- Overlay networks that can adapt in an application-specific manner
- Rich resource descriptions knowledge representation.
A much more “re-configurable” infrastructure is therefore required. Mixture of net technologies! Not just a homogeneous set of techs, but mix of mobile low power and high-end computers.
Intervention from Jeanette Wing: First decide the agenda and research topics, then define the experiments to validate models, and that will finally drive the infrastructure choice, rather than the other way around. It’s not just about the internet, but about a much wider substrate that spans different physical layers, “internet is just an artefact”.
