Large-scale Edge Services: 1 million lines of code, 1 million devices, 1 
week

	Anthony Joseph

Computing power has been growing rapidly while cost, size, and power
utilization have all been declining, with the result being the
proliferation of inexpensive, yet powerful novel edge devices, such as
Personal Digital Assistants, SmartPhones, and sensor motes. These
novel devices do not operate in isolation, rather they leverage new
communication networks (e.g., WiFi, GSM, Bluetooth, etc.) to share
data and access services.

As a concrete grand challenge, I propose that we enable a one million
lines of code service/application to be developed and deployed to one
million edge devices in one week's time. A problem of this scale
cannot be solved with today's communications protocols and
infrastructure, application development environments, and deployment
and management tools. The last two areas are primarily the domain and
focus of distributed systems research, however there are important
differences that must be considered (e.g., power,
computational/communications limitations, fragility, etc.) in
designing the environments and tools.

There are many sub-challenges to this problem, including:
  - Cross-layer and cross-network services: services vertically
      spanning the protocol stack and horizontally spanning
      access/backbone networks
  - Endpoint-mobility: moving live services across edge devices
  - Service-mobility: moving live services across access networks
  - Personal-mobility: network-/device-independent identifiers
  - Automated, dynamic service composition

New edge devices introduce new opportunities for interesting services.
However, there are further services opportunities when the devices are
combined with communications infrastructure that spans multiple access
and backbone networks. We envision such services as ``Potentially Any
Network Services'' (PANS) that can be accessed across diverse
networks. There are limited commercial instances of these services
(e.g., Google Voice Search by Phone and Google SMS Search by Phone
work on small edge devices). Even more interesting are the
opportunities for service-level mobility (e.g., a multi-function
wireless communications edge device that lets a user switch an active
phone call from GSM to a VOIP over IEEE 802.11), and endpoint-level
mobility (e.g., a user being able to switch an active phone call from
a GSM handset to a VOIP desk phone).

There are many challenges associated with PANS, service-mobility, and
endpoint-mobility. Many are centered around the infrastructure,
specifically, the lack of cross-network infrastructure services, such
as network/device independent naming, personal location services,
data/control transcoders, and communication protocol adapters.
Addressing these challenges will require close collaboration between
multiple layers of the protocol stack, and across network boundaries.
For example, while the infrastructure itself can (and already is in
some cases) be leveraged to ``track'' users’ activities and locations,
such capability does not exist in a cross network fashion. Consider a
user with a PDA phone who moves from outside (where GPS-based tracking
can be used) to indoors (where WiFi fingerprinting or other indoor
tracking can be used). The challenge lies not only in providing a
single location interface to an application on the device, but also in
providing the same information to services in other networks. Clearly,
there are also important privacy and security implications that need
to be enumerated, considered, and addressed.

Novel services also introduce other challenges. In particular, a user
with multiple devices shouldn't have to worry about individual
identifiers, devices, and networks. Instead, they should have an
``identity'' that spans the devices and networks providing universal
endpoint access for people and services across diverse networks.
Consider the situation today, where a user may have multiple e-mail
addresses, home/work/cellular telephone numbers, a Blackberry,
etc. Which identifier/network should be used to reach the user?  Who
decides, the recipient or the person trying to reach them? People
naturally move across device endpoints and networks, and their
identities should also follow, in a form of ``Personal Mobility'',
where the person is the endpoint, not the device.

Rapid service development and deployment will require support for
reusable, composable components. Monolithic, one-time services should
be replaced by componentized, extensible services. Furthermore, given
the rapid growth in new types of edge devices/sensors, networks (e.g.,
WiMAX, UWB, satellite, etc.), the composition process should be
automated as much as possible. By automating the process, new endpoints
and networks can be fully supported and integrated when they are
turned on. Furthermore, enhancements/upgrades in components would
automatically be made available to all users of those components.

Many of the enabling technologies for large-scale services at the edge
exist, however the challenge is what is missing: the necessary intra-
and inter-network communications protocols and services infrastructure
for enabling the rapid easy development and deployment of these novel,
innovative, composable services and new edge devices.