Questions
1. Assuming increased funds supporting distributed high performance
computing applications and networking middleware:
2. For Networking Researchers:
3. Assuming that universities' connections to an advanced network
infrastructure evolve into a commodity service (to be purchased
from various commercial providers, gigaPoPs, or Abilene):
4. Assuming that NSF/ANIR desires to encourage increased commitments
from universities as a prerequisite for future network-related
funding
Answers:
Bob Braden
1. Assuming increased funds supporting distributed high performance
computing applications and networking middleware:
- What is your vision of where the computer science community
(including network research) could be in five-years,
particularly in terms of distributed applications and
computing across high performance networks?
This is a very broad question that I don't feel prepared to answer.
My
crystal ball is perpetually cloudy for such questions. However,
here
are some comments on the underlying assumptions.
First, there may be two distinct visions at issue; the needs of
the
computer science community may be quite different from the needs
of the
network research community, just as the experimental infrastructure
needs in civil engineering are different from the needs of the
users of
our highway system. At the most fundamental level, network researchers
need to "break" the network, while most other computer scientists
want the network to work as reliably and transparently as possible.
When people say "high performance networks", they usually are
thinking
about pure speed in bits or bytes per second. The significance
of high
network speeds is much over-rated. One thing that is not going
to
change in five years is the speed of light, and we are already
close to
the point where the speed of light dominates the end-to-end delay.
There are, however, other dimensions of performance than speed;
I would
suggest that the two most important dimensions are scale and diversity.
Scale has an impact through the number of nodes that are interconnected
as well as the number of distinct data flows that are multiplexed
into
a data pipe. The designers of the Internet continue to be humbled
by
its exponential growth. Diversity comes in several flavors: diversity
of underlying data transmission technologies, diversity of network
administrations, and diversity of user requirements. Scale and
diversity don't sound as sexy as gigabits, but they are more
fundamental.
The Web dominates Internet traffic today. Based upon past experience,
we would not be surprised if some other major computer communication
application emerged within the next five years. For example, a
new
major application area might result from some realization of
"ubiquitous computing". The continuing growth of gates on a chip
seems
like have a much bigger impact than the further evolution of fiber
optics.
I believe that the Internet design is far from "finished". It
will
not be just a "simple matter of engineering" to allow the Internet
to
function well as its size and diversity grow. The basic Internet
architecture was designed by an earlier generation of computer
scientists using abstract principles, and that design has generally
proven highly successful. However, I believe that the Internet
architecture is in something of a crisis today. The original design
is
deficient for tomorrow's Internet because the constraints and
assumptions are dramatically different in 2000 than they were
in 1980.
Neither the scale nor commercialization were considered in the
original
design. These changes in the problem space, as well as the distortions
created by the huge economic importance of the Internet, are leading
to
a continuing series of ad hoc engineering solutions. The result
is an
exponential increase in complexity as engineered features interact
with
each other.
It may be that the network research community can no longer have
any
influence on the direction taken by Internet technology develops;
however, it would be nice to try to pump some entropy out of the
design. The next five years may be our last chance to make a "mid
course correction" in the Internet architecture. What seems to
be
required is a medium-scale (neither below critical size nor too
large
to be manageable) federally-funded research effort in networking.
This
effort will have to tread a fine line between innovation and reality.
- What enabling technologies or infrastructure support would you
require to achieve this vision?
The fundamental "enabling technology" is people -- network researchers
-- funded and organized into a collaborative effort to work towards
solutions of the problems of scale and diversity. Secondarily,
links
and hardware platforms will be needed to support collaboration
within
this group.
2. For Networking Researchers....
- What network research is ready for deployment and
experimentation in the context of an emerging, advanced
network infrastructure?
This question seems a bit strange in the context of network research,
although it is probably a sensible question for other areas of
computer
science. I would expect network research to *precede* or be
*concurrent with* the emergence of an advanced network infrastructure.
Key resarch areas in the short term would include Web caching
(and its
generalizations), the many aspects of quality of service, and
the
architectural support for management of administrative diversity.
- What classes of network research problems require
an advanced research infrastructure for experimentation,
as opposed to testing on the commercial Internet?
Those research problems that require either "breaking" the network
in a fundamental way, or that require performance guarantees that
cannot be met by the commerical Internet.
- What policies or strategies might be used to truly effect
a paradigm shift in the nature of the advanced
infrastructure?
What is needed is a fairly large measure of both flexibility and
wisdom from the funding agencies. I do not know whether this can
be accomplished within the current agency structure and applicable
laws.
3. Assuming that universities' connections to an advanced network
infrastructure evolve into a commodity service (to be
purchased from various commercial providers, gigaPoPs, or
Abilene):
- Can the needs of the computing research community be effectively
addressed using commercial R&E services?
- What are the most important end-to-end infrastructure issues,
e.g., constraints upon a user's ability to access or
manipulate remote resources (such as computational resources,
instruments and storage media) from his desktop? And, is
there any on-campus support available to assist PIs with
their wide-area-networking requirements?
- What responsibility (if any) should NSF have in addressing these
problems/issues?
These seem to be questions about commodity Internet service for
universities. That is not an area in which I can give an opinion.
4. Assuming that NSF/ANIR desires to encourage increased commitments
from
universities as a prerequisite for future network-related funding:
- What methods might computing researchers use to engage their
universities, collaborators, or funding institutions as
partners in enhancing campus and/or related network
infrastructure and network support?
I cannot answer this. I do believe that what is needed is a renewed
sense of committment from the US government, to fund the sort
of
coordinated, intensive network research effort that might make
a
difference. Asking for committment from universities seems like
asking
for automobile drivers to have an increased committment to their
roads. Thanks, I am deeply committed to the freeways I use, but
if a
new freeway is needed, or even a new road-building technology,
I and
my fellow drivers are not going to be very much help.
- What can be done to encourage development of next generation
distributed applications and middleware and the emergence of
a strong user community for advanced networks?
This is a tough poltical and social question. It may be difficult
to get where we need to go within the current laws and governmental
structure.
Judy Brown
1. Assuming increased funds supporting distributed high performance
computing applications and networking middleware:
- What is your vision of where the computer science community
(including network research) could be in five-years, particularly
in terms
of distributed applications and computing across high performance
networks?
In 5 years, collaborative applications will be commonplace.
Remotely-located scientisits will gather to discuss research results
or to
run "what-if" scenarios. As the environmentalists run simulations
along the
lines of "what will happen if xx pollutant is not stopped," participants
will see visualizations of results, and each will be able to steer
the
input to the simulation. The computational resources will be distributed,
and the scientists will not need to know where they are. Other
scientists
and educators will explore information together in networked virtual
environments.
- What enabling technologies or infrastructure support would you
require to achieve this vision?
For the environmental scenarios, we need fast computing, seamless
distribution of the problem to the computing resources, visualization
distributed to all the participants, and an ability to pass control
among
the participants.
For the tele-immersive activities, we need higher bandwidth to
allow more
realistic rendering and transfer of voices, low latency for moving
around
in the scene and for getting the audio transfer without a time
lag, and
ability for all the participants to interact with the model and
with each
other.
2. For Networking Researchers....
- What network research is ready for deployment and experimentation
in the context of an emerging, advanced network infrastructure?
- What classes of network research problems require an advanced
research infrastructure for experimentation, as opposed to testing
on the
commercial Internet?
- What policies or strategies might be used to truly effect a
paradigm shift in the nature of the advanced infrastructure?
I'm not a networking researcher.
3. Assuming that universities' connections to an advanced network
infrastructure evolve into a commodity service (to be purchased
from
various commercial providers, gigaPoPs, or Abilene):
- Can the needs of the computing research community be effectively
addressed using commercial R&E services?
I think some could. However, research needs scale to fit the available
resources, so there are never enough resources. With enough increased
resources, scientists will be able to solve problems they can't
currently
consider.
- What are the most important end-to-end infrastructure issues,
e.g., constraints upon a user's ability to access or manipulate
remote
resources (such as computational resources, instruments and storage
media)
from his desktop? And, is there any on-campus support available
to assist
PIs with their wide-area-networking requirements?
The user interface is one of the most important. Parallel programming
is
still very difficult for some problems. Collaborative tools are
still
rudimentary.
There is limited on-campus support to assist PIs with
wide-area-networking requirements, but not the level of support
needed. For
example, to parallelize CFD code, is a major task.
- What responsibility (if any) should NSF have in addressing these
problems/issues?
The PET program made some good strides in helping researchers
parallelize
code. In addition, NSF should support the development of the tools
and
applications to simplify the use of distributed computational
resources and
distributed interdisciplinary collaborations.
4. Assuming that NSF/ANIR desires to encourage increased commitments
from
universities as a prerequisite for future network-related funding
- What methods might computing researchers use to engage their
universities, collaborators, or funding institutions as partners
in
enhancing campus and/or related network infrastructure and network
support?
There could be a commitment to use indirect costs for local network
infrastructure. Researchers used to be able to write in a lime
item of
"computer use" in grants that helped support and upgrade campus
computers.
Is there a potential line item that could help support and upgrade
campus
networks?
- What can be done to encourage development of next generation
distributed applications and middleware and the emergence of a
strong user
community for advanced networks?
Funding specified for a distributed application and middleware
is an
obvious answer to get them developed. However, scientists will
use these
tools and applications only if it either seamless to do so or
if a value
has been demonstrated to them. Demonstrations at Internet2 meetings
don't
reach the scientists. Demonstrations need to be held at disciplinary
(physics, chemistry...) meetings, preferably by a colleague in
the
discipline.
1. Assuming increased funds supporting distributed high
performance computing applications and networking middleware:
-
What is your vision of where the computer science community
(including network research) could be in five-years,
particularly in terms of distributed applications and computing
across high performance networks?
one driver: mostly fueled by entertainment
bandwidth is finally almost enough for remote high
speed/high realism graphics, and will certainly be
there (although perhaps expensive) in 5 years
second driver: locating data
distributed nature of our networking is because
the *data* we want to look at is distributed
(data mining just one example)
challenge now is not storage but *finding* information
since we're drowning at this point.
status of computing paradigms:
cluster technology
as networking catches up to memory bandwidths
and latencies (still behind but closer),
still doesn't seem like we'll see many conventional
CPU-bound distributed application in the next 5 years,
since per-box computing power increasing so fast.
how many applications really need 100 machines
at a time anymore.
parallel applications
...indeed, parallel applications that work well over
distributed networks are a pretty small subset of all
parallel applications because distributed networks have
inherently large latency (ms range), while
most large parallel applications (in current parallel computers)
barely tolerate 10s of microseconds of latency.
There are applications that don't care about latency:
cracking codes, optimizations that can use multiple
processes to search a large solution with different
initial conditions, and so on. But most
current parallel codes are tightly coupled.
one good result of experiments with distributed networks
could be better latency-tolerant parallel applications.
would be major benefit to traditional parallel computing.
but we shouldn't expect anything like the
"information power grid" to be practical for big
applications within five years. most big applications
have big i/o requirements, and if code runs on some big
machine somewhere, you really need to be concerned
about where the data ends up and how it will be moved
to some place where you can use it. large parallel
computers have i/o requirements of many GB/s today,
and that will only increase. current WANs are a factor
of 10 below that, even if you have the whole link to
yourself.
reckon most likely use of distributed computing
resources in 5 years will be for researchers at
universities x, y, and z to agree to work on a
particular computation at their home sites. the
application will divide a problem in some
coarse-grained way, and the components will
communicate at low bandwidth.
it's possible that some sites will try to sell their
excess cycles over the network to needy buyers.
cycles will have to be cheap, reliable, or fast to
make them more attractive than buying local hardware.
for distributed applications, compute power will have
to be available from several sites at the same time,
so distributed scheduling algorithms will be needed.
1) still blocked on last mile thing (copper ugh)
2) and actually physical support for long haul as well,
fiber, routers, switches
3) routing is a mess
4) network protocols with less overhead
5) transport protocols designed for much different
infrastructures than they run on today.
6) (related to (4,5))
eliminating (or at least reducing) the basic conflict
between performance and network service transparency.
e.g., web, multicast, mailing lists, usenet.
all have same goal: allow a sender to disseminate
information to multiple recipients.
but each mechanism is optimized for different kind of usage:
web: quick, low-intensity random access to
information of limited interest,
with optional access control;
usenet: efficiently flooding limited amounts of
information of interest to
many unspecified readers
with no access control;
mailing lists: tight control over recipients,
IP multicast: real-time apps needing minimum delay
combine best features of each into unified, general purpose
multicast service that adapts to patterns of usage
without the user having to commit to one in advance
doesn't need new/separate network. just resouraces to
build necessary system infrastructure
(e.g., caching servers with big disk farms and
access to existing high speed backbones with
multicasting support)
2. For Networking Researchers: - What network research is ready
for deployment and experimentation in the context of an
emerging, advanced network infrastructure?
1) traffic engineering (MPLS at least. early diffserv prototypes
soon)
2) multicast
3) measurement
not sure anything does, currently.
if it's ultimately intended for the commercial Internet,
it should design that in from start etc...
(glib but legitmately arguable)
don't have fed govt support infrastructure.
3. Assuming that universities' connections to an advanced
network infrastructure evolve into a commodity service (to be
purchased from various commercial providers, gigaPoPs, or
Abilene):
yes, if they're willing to pay for premium services
1) security
2) lack of any user control over path (route) selection
3) campuses have to take seriously the
support their own infrastructure
1) generally none. don't get in the way.
if you have to do something, make sure it's something like:
2) foster research collaboration with industry
3) if possible (unclear how), drive some consensus
on only one or two physical layers and protocol stacks
4) make sure some core software infrastructure remains
public/open-standard (sendmail, bind, squid, nntp)
4. Assuming that NSF/ANIR desires to encourage increased
commitments from universities as a prerequisite for future
network-related funding
networking should be part of university overhead.
(like phone system). get over it, already.
you don't ask NSF to pay your phone bills.
1) infrastructure-wise: Need to address scaling issue.
fund research that will help the types of services
that a few researchers have found useful
become available on a massive scale.
it's not enough to provide nifty bandwidth hog application
with stringent delay and loss requirements to a handful
of big science researchers and other big funding awardees.
completely different problem to make high bandwidth,
real time or strict requirement applications available
as a commodity.
2) economics-wise:
note that (1) does _not_ mean bit charging techniques
except for very high end applications.
(Keep in mind how folks scheme to avoid such charges
just like they do today with the RBOCS).
assume a model with flat rate calling within the IXC (ISP)
and a flat rate charge (like the current T1 to the IXC)
on the access side. Rotally flat rate voice then
perhaps some fixed charge for some real time
allocation they can burn as desired..
goal: not priced per usage, just scalable and flat
like IP today.
(telcos won't like, kills cash cow, oh well)
3) political(econ/infra)-wise:
it is _vital_ that the institutions enhance their own
infrastructure. in more than one case, a new school
hooked up to the vBNS, but was unable to take full
advantage of the connection because the campus
infrastructure wasn't up to snuff. waste.
4) application-wise: concentrate on tools for
building distributed applications.
need common API another layer up.
Mario Gerla
1. Assuming increased funds supporting distributed high
performance computing applications and networking middleware:
a. What is your vision of where the computer science community
(including network research) could be in five-years,
particularly in terms of distributed applications and computing
across high performance networks?
An individual or team will be able to access unlimited computing resources and to mine large databases, in order to deploy very advanced applications, for scientific or business or community purposes, or just simply for fun and entertainment. This will require an ubiquitous, heterogeneous network providing integrated media services to high end desktops (Gigabps) as well as to cellphones (Mbps). Example: there has been a chemical threat in the LA subway system by a terrorist group. The fire dept downloads LA maps with detailed terrain/building layouts from local data bases; it moves them to supercomputers where simulations are run to evaluate the possible dynamics and spread of the spill. At the same time, highway traffic simulations are run to evaluate evacuation routes etc.
b. What enabling technologies or infrastructure support would you require to achieve this vision?
A very elastic, integrated ubiquitous network, capabale to carry streams at different QoS levels, with soft degradation when capacity mismatch is encountered on the path. Also, feedback (via proper interfaces) from network to applications, so that the latter can (possibly via appropriate middleware) take full advantage of the network resources.
2. For Networking Researchers:
a. What network research is ready for deployment and experimentation
in the context of an emerging, advanced network infrastructure?
Middleware level techniques to match applications to network resources. These include concepts such as agents, proxies and active nets.
Also, at lower levels, bandwidth allocation and QoS routing techniques. End to end congestion control alternatives/complements to TCP
b. What classes of network research problems require an advanced research infrastructure for experimentation, as opposed to testing on the commercial Internet?
Network research aimed at interfacing advanced applications with advanced nets . Advanced applications require a high bandw network and may involve several participating partners (across country) to be realistic. Moreover, it is important to be able to control/monitor the experiment (traffic loading, specialized measurements etc). The commercial internet does not provide the required high bwd, controlled environment.
c. What policies or strategies might be used to truly effect a paradigm shift in the nature of the advanced infrastructure?
This question is a bit too vague..
3. Assuming that universities' connections to an advanced network infrastructure evolve into a commodity service (to be purchased from various commercial providers, gigaPoPs, or Abilene):
a. Can the needs of the computing research community be effectively addressed using commercial R&E services?
If during the evolution from research to commodity operation, the research community has stressed the network hard enough to make sure that the key features required for advanced research are in place (namely, it has "debugged" it for advanced operations), then the resulting commodity network will probably meet the needs of the comp research community for the next 3 to 4 years. Beyond that, the "next" vBNS or Abilene should be brought in.
b. What are the most important end-to-end infrastructure issues, e.g., constraints upon a user's ability to access or manipulate remote resources (such as computational resources, instruments and storage media) from his desktop? And, is there any on-campus support available to assist PIs with their wide-area-networking requirements?
Using remote computing/database resources for advanced research experiments typically requires quite a bit of admin/logistic hassle (thus possibly discouraging more widespread use/experimentation ). Incompatibility issues deriving from lack of standards are another problem. But, of course, standards often run contrary to innovative research. As for wide area net access, the Campus Computing Services are getting better at that, since they also stand to gain visibility with the Chancellor.
c. What responsibility (if any) should NSF have in addressing these problems/issues?
NSF has done a great job in promoting access to supercomputers and to vBNS via infrastructure and research grants. For vBNS, the success has been only partial and, certainly, much delayed because of the difficulty of leasing OC3 access lines. I expect this to be a singular problem not to recur. I think NSF should retain its role of providing high tech infrastructure support. In fact, there should be a time when any NSF PI should be given access to such resources. Ie, no need to apply for a special grant. Just get the access as a bonus along with the research grant.
4. Assuming that NSF/ANIR desires to encourage increased commitments from universities as a prerequisite for future network-related funding
a. What methods might computing researchers use to engage their universities, collaborators, or funding institutions as partners in enhancing campus and/or related network infrastructure and network support?
Actually, most Universities and industrial partners are very willing to cost-share in these projects (see Internet II etc). They all stand to gain in many different ways. The vendors have obvious marketing reasons. But also Universities are enticed by the visibility of high- profile projects. So, good research track record by the PI is still probably the best method.
b. What can be done to encourage development of next generation distributed applications and middleware and the emergence of a strong user community for advanced networks?
The cliche' answer to this question is the marriage of advanced applications developers with network researchers. However, collaboration among universities and research lab groups is not enough. The success of European projects like Esprit and ACTS (where Industry and University collaborate) is telling us that there should be more active industry committment/involvement (well beyond just paying the network bills) in the US next generation programs. This of course implies more liberal intellectual property rights.
Farnam Jahanian
1. Assuming increased funds supporting distributed high performance
computing applications and networking middleware:
Several broad trends in information technology are transforming
the era of workstations and personal computers into an era of
Internet-based networked applications: the proliferation of inexpensive
powerful microprocessors running primitive run-time environments;
the rapid convergence of voice and data networks into a single
"universal network" based on IP; significant increase in bandwidth
in the backbone networks; the promise of a global mobile infrastructure
enabled by advances in wireless technologies; and the emergence
of commercial applications that promise to exploit an immeasurable
number of electronic devices that are intelligent and connected.
Emerging networked applications enabled by these trends are diverse
and ever-expanding. They include: collaborative computing; tele-medicine;
interactive distance learning; remote sensing; secure video conferencing;
electronic commerce; virtual and interactive engineering and prototyping;
anytime-anywhere access to data; intelligent transportation systems;
multi-player games; distributed agile manufacturing; and many
others.
The realization of this new era of networked applications and
massively pervasive computing is not without significant obstacles
and technical challenges. Future networked applications will be
deployed in environments with significant variability and heterogeneity
in communication infrastructure and end-system resources. We can
envision an environment in which a potentially unbounded number
of intelligent sensors, actuators, and computation and storage
devices are connected across a heterogeneous (wired & wireless)
communication infrastructure. There is significant variability
in the capabilities and computational resources of client nodes,
from full-featured systems down to handheld mobile devices carried
by individuals. This gap will continue to widen as a significant
fraction of Internet clients will consist of high-volume handheld
devices and smart phones.
There is also significant variability in the quality of network
connectivity; the communication network infrastructure is a confederation
of many different technologies, each with unique capabilities.
The gap between the least and most capable in terms of bandwidth,
loss, and latency will grow with scale, requiring adaptation.
The system must also provide an appropriate data quality to each
user, when common data sets are accessed. Finally, future networked
systems must be able to adapt to rapidly changing application
demands. Applications may have different computational and data
delivery requirements, and these requirements may change over
time, rendering the overall system highly dynamic. The distributed
runtime environment for such systems must include middleware services
to support changing functional, timeliness, dependability and
security application requirements. The availability of a global
mobile infrastructure introduces further complications such as
name space explosion and the need for efficiently locating and
accessing instances of replicated resources.
Perhaps the central issue for the network core is quality of service:
"providing multiple diverse service levels, including guaranteed
performance levels, on a shared infrastructure, along with all
the resource management, provisioning, and engineering tools"
that is required. A particular challenge is that the core of the
network will continue to grow in capacity whereas the edges will
continue to be bandwidth constrained, potentially interconnecting
an enormous number of (wired and wireless) embedded devices and
client nodes.
2. For Networking Researchers....
Areas of networking research, potentially ready for experimentation
and deployment, spans a wide range: inter-domain and intra-domain
routing protocols and policies including multicast routing, QoS
resource provisioning and allocation, passive and active measurement
and analysis techniques/tools, aggregation of distributed monitoring
& probe points, scalable security architectures, and network intrusion
detection systems.
Numerous classes of network research problems require an experimental
infrastructure critical for implementation, deployment and evaluation
of next generation networking technologies. I will categorize
them into several broad groups, including:
3. Assuming that universities' connections to an advanced network
infrastructure evolve into a commodity service (to be purchased
from various commercial providers, gigaPoPs, or Abilene):
There are three fundamental obstacles to using existing experimental
or commercial networks in support of computing communications
research:
1) While we have observed a tremendous growth in the available
or planned bandwidth in the core networks, edges continue to be
bandwidth constrained. Campus-wide networking infrastructure continues
to be the bottleneck for getting access to the high-speed networking
infrastructure such as the vBNS.
2) On-campus support for deployment and maintenance of experimental
or commercial technologies is a very human-intensive effort, which
requires significant support from individual institutions.
3) Deployment and evaluation of most experimental network research
(such as those mentioned in Q2) on commercial networks is very
difficult or perhaps impossible. Commercial ISPs obviously tend
to be very conservative with respect to deployment of experimental
research technologies. The I2/UCAID/Abiliene are a direct outgrowth
of the research universities' frustration with the commercialization
of the Internet. However, the use of experimental networks (which
are intended primarily for data intensive applications) by the
networking research community is yet to flourish.
I would like to see an advanced network infrastructure program
complemented by multidisciplinary research programs which bring
network research in closer synergy with application research.
NSF can play a significant role in building a stronger user community
that works closely with the networking research community. NSFNET
is certainly a testament to how modest federal investment can
pay off enormously.
4. Assuming that NSF/ANIR desires to encourage increased commitments
from universities as a prerequisite for future network-related
funding:
Let me offer an answer based on my interaction with Merit Network,
which provides network connectivity to educational institutions
and libraries across Michigan. Merit has been involved in helping
its members institutions to develop their high performance connection
proposals. One of the main barriers was getting matching fund
commitments from university administrations. It would be helpful
for the researchers to promote the need for an advanced network
infrastructure to their university administration.
In response to the second part of the question, I should note
that the central administration at the University of Michigan
allocated seed funding to foster deployment of applications on
the vBNS network. Similar efforts matched by NSF funding can be
helpful in ensuring the deployment of new middleware and applications.
Furthermore, as mentioned above, an NSF-sponsored program that
brings together network researchers and application domain (user)
communities is crucial for the success of an advanced network
infrastructure.
Sid Karin
1. Assuming increased funds supporting distributed high performance
computing applications and networking middleware:
A. What is your vision of where the computer science community
(including network research) could be in five-years, particularly
in terms of distributed applications and computing across high
performance networks?
Five years from now, people who use the high performance networks
will be divided into two sub-groups, network researchers and researchers
who use the high performance network. The first group's intellectual
curiosity will be to study the actual workings of the high performance
network, i.e. routers, switches, backbone, bandwidth, and speeds
of packet transfers. The second group will merely be using the
high performance network to satisfy their intellectual curiosity
that focuses on disciplines including but not limited to, computer
science, astronomy, biology, marine science and chemistry. The
experiments that both groups will be working on will change over
the next five years; some of which we cannot even begin to imagine,
today.
B. What enabling technologies or infrastructure support would
you require to achieve this vision?
In five years, the network will possess many differences including
greater bandwidth and features that do not currently exist. These
features might include 'quality of service' technologies, reservations
for bandwidth, guaranteed bandwidth, and end-to-end security.
2. For Networking Researchers:
A. What network research is ready for deployment and experimentation
in the context of an emerging, advanced network infrastructure?
N/A
B. What classes of network research problems require an advanced
research infrastructure for experimentation, as opposed to testing
on the commercial Internet?
N/A
C. What policies or strategies might be used to truly effect a
paradigm shift in the nature of the advanced infrastructure?
The most significant policy that would truly effect a paradigm
shift in an advanced infrastructure would be a change in the pricing
model. In a usage-pricing model, the industry would move from
a fixed bandwidth scenario to a usage model. In this model, researchers
and industry would be charged for usage, and would guarantee or
reserve bandwidth time, comparable to the allocation of time on
supercomputers. A very important item to consider is the future
impact that wireless growth would have on usage. Every pager,
cell phone, car, and plane could become a node on the infrastructure,
and increase usage on the Internet. Imbedded processors and wireless
devices would not only increase the demand on the network, but
also impose new requirements on the system.
3. Assuming that universities' connections to an advanced network
infrastructure evolve into a commodity service (to be purchased
from various commercial providers, gigaPoPs, or Abilene):
A. Can the needs of the computing research community be effectively
addressed using commercial R&E services?
Yes, for the people who are not doing research on the network,
but no, for people who research networking.
B. 1) What are the most important end-to-end infrastructure issues,
e.g., constraints upon a user's ability to access manipulate remote
resources (such as computational resources, instruments and storage
media) from his desktop? 2) And is there any on-campus support
available to assist PIs with their wide-area-networking requirements?
1) The most important issues that need to be addressed would be
guaranteed and reserved bandwidth, as well as, bandwidth usage
coordinated with resource usage. For example, if an astronomer
had a limited amount of time reserved on a supercomputer and a
telescope in order to do his/her observations, and the area he
was to observe would be dependent upon certain observable data,
bandwidth availability would be crucial. With guaranteed/reserved
bandwidth, the astronomer could analyze incoming data in order
to accurately choose his/her coordinates, and continue his/her
experimentation in real-time. A lapse in communication due to
lack of bandwidth could cause the researcher to waste time on
irrelevant and perhaps useless data in regards to their specific
research. 2) Yes, there should be on-campus support available
for PIs.
C. What responsibility (if any) should NSF have in addressing
these problems/issues?
Ultimately, it is the NSF's responsibility to support the research
that the private sector will not. In regards to the user community,
i.e. industry and researchers, the NSF may not play a major role,
but the commercial world does not support network researchers.
In other words, network researchers worry about performance while
the private sector worries about price over performance. Therefore,
for network researchers, the NSF is the main source to look towards
for funding for this type of research.
4. Assuming that NSF/ANIR desires to encourage increased commitments
from universities as a prerequisite for future network-related
funding:
A. What methods might computing researchers use to engage their
universities, collaborators, or funding institutions as partners
in enhancing campus and/or related network infrastructure and
network support?
Commodity data communications should be treated by funding and
funded institutions, (the government and universities,) in the
exact same manner as voice communications. Currently, and in the
future, experimental systems will continue to require special
treatment.
B. What can be done to encourage development of next generation
distributed applications and middleware and the emergence of a
strong user community for advanced networks?
Incentives, such as awards and support, will be needed for people
who develop applications and middleware in order to encourage
next generation development. The awards could be in the form of
funded programs through the NSF with support coming from expert,
dedicated staff to assist the developers and the applications
people.
H.T. Kung
COMPUTING IN THE COMING DECADE
Processor, display and storage technologies will continue to advance
at a rapid pace. For example, a pager-size device will have a
plug-in module to perform today's largest desktop applications.
A personalized mobile device will satisfy most of an individual's
communications and computing needs. A video recording of a person's
entire lifetime will fit in a hand-held device.
In the meantime, network bandwidths will expand drastically. Wavelength
division multiplexing will allow fibers to expand hundreds-fold
in capacity, facilitating national backbones that operate at one
trillion bits per second, or beyond. Campus networks equipped
with gigabit links will have abundant bandwidth for their users.
For metropolitan areas, cables and high-speed digital transmission
technologies will deliver 100s of megabits per second.
Many distributed applications will emerge. They will depend on
constant connectivity to large networks. Transactions in electronic
commerce, involving multiple servers over the network, will be
routine. Network agents or robots will carry out searches of multiple
databases across wide-area networks. By default, computation,
data storage, and users will be separated by large distances.
Current client-server, remote procedure call, and web protocol
models represent only the very beginning of the work in this direction.
OPPORTUNITIES AND CHALLENGES IN COMPUTER SCIENCE
These expected changes in computing will demand new computing
models and the associated programming support. We will need to:
We don't really know how these new models will look, but we can
expect that they will be very different from the present ones,
and will have huge impacts on the operation of our society.
WORKLOAD DATA NEEDED IN NETWORKING RESEARCH
System-level networking research is important but has been difficult.
In the past, networking research at universities was mostly based
on analysis and simulation.
The situation is changing due to technology advances. Universities
can now set up experimental network testbeds easily using inexpensive
PC-based routers, switches, and network interface cards, as well
as source codes available in the open domain. For example, at
Harvard, in one month, we put together such a 10/100Mbps network
testbed involving about 20 hosts/routers/delay boxes.
However, we need workload data reflecting traffic on real-world
networks. For example, it would be useful to have data on (1)
the number of active flows on the Internet backbone and their
characteristics, (2) traffic source/destination patterns, and
(3) multicast usages. Part of this information is available, but
much of the important information is not. We need partnership
between industry, academia and government funding agents, to ensure
that, when new data is needed for research purposes, there will
be a way to acquire it.
IMPORTANCE OF OPEN SOURCE
One of critical components in encouraging development of next
generation distributed applications and middleware is a national
R&E infrastructure where high-quality and state-of-the-art software
sources are available in the open domain. The BSD source was a
clear example of this. We will need to nurture an environment
where continuous growth of quality open sources can be ensured.
Bill Lennon
1. It's all engineering compromise driven by application demands.
Distributed computing (or network) architectures are a function
of the current volume production of a technology moderated by
the laws of "universal pork." There is a continually recurring
cycle. High end computing castles (siting moderated etc....) gradually
evolved into data storage centers with computers at every village.
The village computers began to cluster and move back into the
castles leaving behind "client/server" access serfs. These later
sites evolved into more powerful local clusters which further
stored a portion of the data, caching the needed copies of the
rest -- the metaphor is starting to break down...
Visionaries succeed in articulating a vision. They promulgate
the vision widely and engage the R&E community in implementing
its components. Success is achieved when the implementations are
"ubiquitous" -- high volume production and use with easily mastered
human interfaces.
Political decisions will effect resource siting. Independently
of that, the resources that are clustered at advanced computing
sites will reflect the current technology about to be deployed
in 'volume' production. In five years the computer science research
community will be using the tools produced in volume as a result
of today's visions, implementing the current vision and formulating
the next. The implementations will display the continued cycles
of concentration and distribution during the evolution.
Central to this flow of computing and storage between "castles"
and "villages" are the network roadways -- the bigger and faster
they are, the more rapid the cyclic evolution. Central to deployment
of the networks is the application driven business case required
to justify funding the infrastructure. It will necessarily lag
the computer science R&E but both communities will benefit from
promulgating needs and capabilities into each others' domains.
2. Wider deployment of trials stressing security, QoS and human
interfaces are essential to attract the needed investments for
wider deployment of high speed infrastructure.
Research involving technologies which potentially threaten existing
infrastructure will require segregated resources. Examples are
new light modulation schemes or signally protocols. Segregated
resources are also required for the development of migration strategies
for, say legacy dense WDM.
3. The need can be address by a mix of production (commodity)
facilities and unique research resources. Relying on commodity
resources alone will doom us to "incremental" change. The strategic
needs of the country demand there be an environment facilitating
"revolutionary" rather than "evolutionary" change.
4. Finally, it is always wise to befriend Igor, the castle gate
keeper and the only one who knows all of the secret passages and
the locations of the deadfalls. Installing fiber optics in the
dungeons without that cooperation will guarantee crushed fiber.
The nation, the university, even the department have a strategic need to shorten the "time to high volume market" for good ideas. Every university Computer Science department should have an active role in planning the evolution of the university and regional infrastructures. An out of touch university will miss opportunities to introduce important paradigm shifts for the community and an unawares community stand to be disconnected from the rest of the country.
Tracie Monk
1. Assuming increased funds supporting distributed high performance
computing applications and networking middleware:
- What is your vision of where the computer science community
(including network research) could be in five-years, particularly
in
terms of distributed applications and computing across high
performance networks?
In an ideal environment, we would have OC3 to the desktop for
those
researchers requiring high-performance network connectivity.
Computing/networking infrastructures (hardware, software, and
support
services) would support collaboratory environments for researchers
enabling real-time audio, visual and data communications between
remote parties. This is distinctive of advanced computing research
which would require the availability of high bandwidths during
select
periods.
- What enabling technologies or infrastructure support would you
require to achieve this vision?
QOS and differentiated services are important -- as are the
development of appropriate pricing models supportive of their
deployment. Shifting the campus infrastructures toward a merging
of
voice and data services and upgrading of infrastructure and support
services for the research staff may also be critical.
2. For Networking Researchers:
- What network research is ready for deployment and experimentation
in
the context of an emerging, advanced network infrastructure?
Several QOS schemes are available for testing and deployment,
however,
in the absence of appropriate settlement and pricing models and
acceptable alternatives for accounting and billing, it is unlikely
that any technical approaches will be deployed in the commercial
Internet.
- What classes of network research problems require an advanced
research infrastructure for experimentation, as opposed to testing
on
the commercial Internet?
Numerous emerging protocols and technologies require early deployment
and experimentation on advanced research infrastructures. However,
increasingly these research infrastructures fail to provide the
real-world environments necessary for rigorous testing and evaluation
of new technologies -- particularly those requiring testing in
high
traffic, multi-backbones, multi-vendor environments.
- What policies or strategies might be used to truly effect a
paradigm
shift in the nature of the advanced infrastructure?
One possible avenue might be to encourage development of a neutral
experimental network access point where both commercial ISPs and
research networks could exchange traffic and collaboratively test
new
technologies and protocols in a relatively controlled setting.
This
would require a partnership of government, commercial and research
interests.
3. Assuming that universities' connections to an advanced network
infrastructure evolve into a commodity service (to be purchased
from
various commercial providers, gigaPoPs, or Abilene):
- Can the needs of the computing research community be effectively
addressed using commercial R&E services?
It depends on the definition of 'computing research community'.
For
most users, reliable, efficient Internet service is sufficient
for
their current needs. If modes of interacting with other researchers'
change and reliance on distributed resources continues to grow,
then
significant upgrades to campus infrastructures will be needed.
For
researchers requiring high bandwidths to support their real-time
and/or bursty network requirements, commodization of R&E networking
could be very destructive in the absence of QOS support and acceptable
settlements/accounting structures.
- What are the most important end-to-end infrastructure issues,
e.g.,
constraints upon a user's ability to access or manipulate remote
resources (such as computational resources, instruments and storage
media) from his desktop? And, is there any on-campus support available
to assist PIs with their wide-area-networking requirements?
Transit bandwidth is not a major issue for most users. The
limitations are those present in the local infrastructures (physical
and service related infrastructures). Wide-area networking typically
does not garner as much the attention from campus networking personnel
as local priorities, e.g., connecting dorms.
- What responsibility (if any) should NSF have in addressing these
problems/issues?
Education is fundamental. Identification of alternative strategies
(such as coupling of voice and data networks), as well as limited
financial assistance may be beneficial. Reimbursement of network
expenses as acceptable indirect costs for universities is important.
4. Assuming that NSF/ANIR desires to encourage increased commitments
from universities as a prerequisite for future network-related
funding
- What methods might computing researchers use to engage their
universities, collaborators, or funding institutions as partners
in
enhancing campus and/or related network infrastructure and network
support?
Making NSF priorities and goal explicit through language in agency
solicitations (across disciplines and divisions) would help.
- What can be done to encourage development of next generation
distributed applications and middleware and the emergence of a
strong
user community for advanced networks?
Encouragement of cross disciplinary collaborations and
industry/academia collaborations -- particularly in sectors where
near-term opportunities for commercial use are present, e.g.,
healthcare.
Craig Partridge
1. Assuming increased funds supporting distributed high performance
computing applications and networking middleware:
I'm gloomy about the "high performance" issue and optimistic about
middleware.
On high performance computing, I think the speed of light delay
has won. The relative difference in delays between local and remote
operation means that I don't think most people have a hope of
doing much high performance distributed computing over an area
larger than a few hundred miles in diameter. (There are exceptional
cases where the problem can be structured to work well over longer
distances, but they are rare).
On middleware. I think there's a general concensus that today's
middleware is terrible. It is fragile (does not comfortably adapt
to the vicissitudes of packet data networks) and it is inflexible
(RPC is a terribly limiting interface). And there's popular acceptance
of the idea (now ten or more years old) that we need a new paradigm
that makes code first class objects (e.g. JAVA). Now Java isn't
perfect, but the notion that we can ship code anywhere means that
we can reprogram components to interact and that's a big win.
So, my hope is that, with the right algorithms and constructs,
we could start to see a move to robust, mobile and flexible suite
of middleware in five to ten years.
2. For Networking Researchers:
We've got lots of stuff that could be deployed. Various schemes
for traffic management. New ideas about routing protocols. Schemes
for transmission over large numbers of parallel links.
Most of these ideas require an infrastructure independent of the
commercial Internet, because the ideas involve changing router
and switch software and, inevitably, will mean more frequent router
and switch crashes (and network outages) while the testing is
done.
The infrastructure, however, does not have to be advanced. It
can, for instance, be low speed.
3. Assuming that universities' connections to an advanced network
infrastructure evolve into a commodity service:
Most of the time, yes. The key issue is if someone needs to experiment
at the network level, such that they need to be able to break
and reconfigure the network. That requires separate facilities.
It depends. My experience is that application developers want
a rock solid service, that is reliable and highly available. Networking
folks want things they can break... Rock solid service is expensive...
I'm not sure.
4. Assuming that NSF/ANIR desires to encourage increased commitments
from universities as a pre-requisite for future network-related
funding:
I don't have any strong ideas here.
George C. Polyzos
1. Assuming increased funds supporting distributed high performance
computing applications and networking middleware:
Quality-of-Service support is one of the next big things for the
Internet. It will take a concerted effort to really get there.
Wireless access at multi-Mb/s and support for mobility is also
an important goal. Even though the broadband wireless access part
could be seen as an isolated network technology issue, support
for mobility and QoS in this context would require an appropriate
fixed infrastructure.
2. For Networking Researchers....
QoS approaches.
QoS support. Large-scale mobility support and QoS support for
wireless.
3. Assuming that universities' connections to an advanced network
infrastructure evolve into a commodity service (to be purchased
from various commercial providers, gigaPoPs, or Abilene):
A wide-area, large scale infrastructure that could support experimentation
in the areas of QoS support (in the stricter sense and in the
DiffServ sense) with minimal administrative overhead would be
a big push towards the next breakthrough in Internet technology
and practices.
4. Assuming that NSF/ANIR desires to encourage increased commitments
from universities as a prerequisite for future network-related
funding:
Lawrence A. Rowe
1. Assuming increased funds supporting distributed high performance
computing applications and networking middleware:
As a user I should transparent access to general computing resources
(e.g., storage, computing cycles, etc.) and specific services
(e.g., animation rendering, databases, etc.). The point is that
I should be able to sit at my desktop and just use resources without
having to know exactly where or how they are provided. I should
also be able to interact with colleagues using widely deployed
video conferencing software and hardware. It should be possible
to easily bring together a group of people to discuss, view, and
modify documents and discuss, view, and operate physical equipment.
Beyond the obvious high-speed connections and computing resources
(e.g., parallel cycle providers, storage, etc.), two required
major software middleware and applications areas are: 1) multicast
streaming media applications and 2) parallel processing applications.
Research on the hardware infrastructure (e.g., networks, storage
systems, parallel processors, etc.) will not produce the desired
outcome unless a substantial investment is made in software and
applications. Some of this investment can be made to specific
research groups developing or using specific applications (e.g.,
a climate modeling system), but investments must also be made
to deploy shared applications and services (e.g., video conferencing,
animation rendering, video compression, etc.).
2. For Networking Researchers....
One problem I see is that Internet2 initiatives, such as CENIC,
require a level of production service that is incompatible with
experimentation. For example, it has been difficult to get CENIC
to deploy multicast services - in part due to lack of knowledge/understanding
on the part of network administrators and in part due to real
problems with multicast protocols. We need a way to run these
protocols so that experimentation using high-speed connections
can be performed without impacting the production operation of
the basic networks.
3. Assuming that universities' connections to an advanced network
infrastructure evolve into a commodity service (to be purchased
from various commercial providers, gigaPoPs, or Abilene):
Yes, but only if you are willing to pay the money required to
support and operate the various experimental services. It may
require operating separate virtual private networks over the same
links that have guaranteed services (e.g., bandwidth allocations
and latency guarantees).
4. Assuming that NSF/ANIR desires to encourage increased commitments
from universities as a prerequisite for future network-related
funding:
Build/acquire applications and expertise to support people in
the community deploying applications. For example, a reliable
high-quality multicast conferencing application and service should
be available to every researcher at his or her desktop. Few researchers
today have the required hardware (e.g., microphones and cameras)
and software (e.g., Internet MBone tools, or H.323 compatible
system) deployed. It is difficult to select the appropriate hardware
and software and provide the support required to install it. The
research community should develop hardware configurations/bundles
and provide software and support.
Another application/service might be on-line training and educational
material. We should use the tools we develop/advocate to improve
our own use of the technology.
Tatsuya Suda
1. Assuming increased funds supporting distributed high performance
computing applications and networking middleware:
I am not really sure what this question is asking, but here is
my take on this question. I hope I am not way off from what's
being asked here.
We will start seeing applications which are more personal and
more customized to meet individuals' varying needs than those
we currently discuss. Researchers often talk about applications
such as distance learning, crisis management, virtual reality,
etc., which I think are important and can justify federal funding,
but are rather limited in terms of commercial market. Applications
such as video games and interactive movies (i.e., applications
which are appealing to ones' hobby and personal interests) will
probably become more dominant over the Internet.
Computer Science community will need to be diversified in order
to be able to develop above applications. For instance, we probably
need to look into social science aspects such as how human uses
computers for fun, how network applications can change human behaviors,
etc.
enabling technologies
social science (e.g., understanding human behaviors, how technology
changes human behaviors)
economy (e.g., game theoretic models of human behaviors and of
network resource usage)
man machine interface (e.g., easy to use interface for 5 year
old kids and 60 year old grand ma.)
multimedia networking within home and within classrooms for kids
to have run fun applications
2. For Networking Researchers....
multicast, QoS, IPv6, video related applications
hardware related network research (e.g, wireless transmission
systems, new high speed router hardware + software)
resource allocation/management/scheduling
middleware (QoS supporting middleware, object oriented frameworks)
Force network researchers (e.g., univ. professors) and network
infrastructure managers (e.g., CIOs) work together. They don't
currently understand each other's needs.
3. Assuming that universities' connections to an advanced network
infrastructure evolve into a commodity service (to be purchased
from various commercial providers, gigaPoPs, or Abilene):
What's R and E?
I doubt so. Cost ($$) can be an issue. Also, sometime "sharing
a link" may become an issue for some experiments. For some real
time experimental applications (e.g., controlling telescope remotely),
we may need to set aside a huge chunk of network resources (for
the duration of experimentation), and I don't think commercial
services can meet this need.
End system issues such as efficient disk storage, fast processing
of protocols, new CPU architecture/instruction-sets for fast protocol
processing, etc. Often networks can deliver a large number of
bits to end systems, but end systems cannot handle them.
One word. Increase funding :-)
Well, stimulate interdisciplinary research. Stimulate collaboration
between network researchers and network managers. Provide access
to a testbed to a number of researchers.
4. Assuming that NSF/ANIR desires to encourage increased commitments
from universities as a prerequisite for future network-related
funding
No suggestion here. I have to think about this.....
Application development, tool building, and infrastructure building
are sometime considered "not exciting" and used against in tenure
decisions. Educate university researchers and university mangements
so that they are engage in such research without fear of not getting
tenure.
Pull in researchers from other research areas, educate them about
network infrastructures, have them work with network researchers,
so that new applications emerge.
John Wroclawski
1. Assuming increased funds supporting distributed high performance computing applications and networking middleware:
- What is your vision of where the computer science community (including network research) could be in five years, particularly in terms of distributed applications and computing across high performance networks.
- What enabling technologies or infrastructure support would you require to achieve this vision?
Traditionally, networking infrastructure projects have focused on increasing speed. There are two reasons: speed is a simple, quantifiable objective, and speed has been seen as the enabler for broad new classes of applications.
In my opinion it has become increasingly clear that a) speed is not the hard problem, and b) the commercial sector is both able and motivated to develop new technology as needed in this area.
Instead, the challenging problems are to be found in three areas:
- Scale - ability to support billions or trillions of devices
- Conformance to human needs - security, policy and autoconfiguration
- Cost - order-of-magnitude reductions in network interface and
access technology cost.
Progress on these problems will lead to a significant shift in the basic role of computer networks. Today's networks are primarily thought of as providing connections between separate, autonomous computers. Within five years, networks will need to accomodate a much larger number of much more specialized, less costly devices. Physical-world sensors and simple information retrieval appliances will augment and numerically dominate today's PC's, workstations and servers. Ubiquitous wireless access will accomodate both mobile professionals and the communication needs of rural areas. Applications will become vastly more fluid, dynamically composing themselves around the resources available and relevant to the specific circumstance - for example the set of sensors near an emerging storm, or the collection of databases and natural language front-ends needed to support speech-based access to specialized medical or scientific knowledge.
Systems of this sort will result in compelling new applications of benefit to both the research community and the larger society. Interest is growing in examples such as these derived from the recent PITAC report:
- Low-cost wireless sensors can provide detailed real-time information
on air and water conditions, dramatically improving our ability
to conduct environmental research, monitor the environment for
pollution, and respond to man-made disasters.
- Personal "guardian angels" - small networked deviced backed
by automated data monitoring and analysis - could monitor the
health and safety of at-risk individuals such as firemen, home
health care patients, and researchers in remote areas.
The three challenges listed above are of course interrelated. A network of millions or billions of appliances, sensors or smart devices will not be practical if the infrastructure protocols cannot scale to that size, but it will equally not be practical if it cannot effectively configure and manage itself to meet users' technical or policy requirements, or if the cost of network interfaces has not dropped by an order of magnitude from current levels. To create the paradigm shift, we must make progress on all three fronts at once.
With this shift in vision, the infrastructure needed to support networking researchers expands. Three key needs, beyond the traditional ones of reasonable speed access and programmable infrastructure components on which to test new algorithms, are:
- Access to non-traditional client platforms, including mobile
and embedded systems hardware and inexpensive wired and wireless
networking technologies. The researcher interested in algorithms
and protocols for sensor-nets, information appliances and the
like would benefit greatly from a community-supported, documented
toolkit of low-level hardware and software for cheap, easily reproducable
networked devices. Examples might incorporate Bluetooth or emerging
"home networking" platforms.
- Access to high-performance WAN wireless technologies. Partnerships
with cellular and satellite communications providers are difficult
for individual researchers to arrange, but are vital to understanding,
anticipating, and benefiting from commercial investment in cutting-edge
access technology.
- Access to truly high-performance network modeling and simulation
capabilities. Unfortunately the development of such capabilities
is a research task in itself - we do not currently have any way
of simulating billion-device networks with any degree of assurance.
Possible approaches include more effective distributed and parallel
simulation algorithms and the use of higher-level "abstract" simulation
algorithms that remain accurate while discarding unnecessary details,
along with the obvious one of access to faster computers.
Finally, turning briefly to a more traditional area of infrastructure support, rapid advances in the technology of low-cost optical *access* networks may occur in the next five years. Where previous optical network R&E has focused on high-speed backbones, today's research is beginning to focus on driving the *cost* of all-optical networks down to the point where they are useful as last-mile access technology. The goal is to achieve order-of-magnitude cost reduction by using inexpensive passive optical components wherever possible in the backbone to end-user path. This has two effects.
- It opens up the possibility that very high speed (2.5-10Gb)
network connections will become available - and affordable - for
end systems.
- It creates a significant new challenge for network management
and administration technology, because of the increased use of
simple, passive components in the cloud. Today, network management,
policy control, monitoring, and administration are done *within
the network*, at boundary routers and switches. Passive, all-optical
distribution networks, and particularly transparent optical desktop
to backbone connections, leave that approach unworkable. A substantial
rethinking of the management and control problem will be required.
For these reasons, I suggest that leading-edge availability of all-optical access infrastructure may have high payoff for both the high-performance computing and the networking research communities.
Lixia Zhang
1. Assuming increased funds supporting distributed high performance
computing applications and networking middleware:
(Due to time constraints I am only attaching a brief response
at this time, and most of it is trying to further clarify the
concepts and issues; I will try to elaborate further over the
weekend and at the workshop)
To seek answers to this question I would like to first have a
clear understanding of the terminologies: what are included in
the "distributed applications"? Past experience shows that different
research communities often interpret these words differently.
For example, is Web, or Web search, or Web caching, considered
distributed applications? Or what intended here is more specific,
such as scientific experiments accessing remote supercomputers?
2. For Networking Researchers:
Should I interpret this "advanced network infrastructure" as a
long haul network testbed with dedicated resources? If so, I am
not aware of any, in terms of network research as opposed to high
performance computing research.
I do believe that there is an urgent need for conducting research
in networking infrastructure; my use of the word "infrastructure"
does not include the raw bandwidth resources. Research effort
is needed to address the following large-scale infrastructure
issues: scalability, robustness (against failures as well as malicious
attack), mobility, and security. Research effort is also needed
to develop a broad range of *service* infrastructure that enables
the development of more advanced distributed applications, hence
enables the Internet to evolve towards being the telecommunication
infrastructure for the next century.
3. Assuming that universities' connections to an advanced network
infrastructure evolve into a commodity service (to be purchased
from various commercial providers, gigaPoPs, or Abilene):
I am eager to learn the answer from the computing community.
4. Assuming that NSF/ANIR desires to encourage increased commitments
from universities as a prerequisite for future network-related
funding
for more information: info @ caida.org
last update:
this page: