Open Source Documents: CAREER Grant - Transparent Bandwidth Conservation Techniques - Review Results

Info - Why am I doing this?

Comment - Striegel

The following reviews are taken verbatim from Fastlane and put in the not so great verbatim format for the Wiki. Apologies in advance if you have to scroll to read the reviewer comments.

Initial Information

Your proposal to the CAREER FY 2004 competition has been received by the National Science Foundation (NSF), and was assigned to the program that you requested upon submission. If necessary, proposals may be reassigned to other NSF programs. Information on any changes in the program assignment of your proposal, the date of final program approval, as well as the identification of overdue reports and other factors that might be delaying the proposal process, will be available on this page and should enable you to track the progress of your proposal.

Deadline for submission of CAREER proposals is July 22, 23, or 24, 2003, depending on the Directorate/Office to which you submitted your proposal. Administrative review of proposals for completeness and conformance to NSF guidelines, as well as appropriate program assignment, is carried out in the weeks immediately following these submission deadlines. Proposals that do not meet required guidelines will be returned without review.

All compliant CAREER proposals will be evaluated by external peer review. Please note that the review process and specific dates of proposal processing may vary from program to program within this general timeframe. The Program Officer to whom your proposal has been assigned is your best source of detailed information on the processing of your proposal and will assist you with any additional information that you require.

NSF is striving to be able to tell applicants whether their proposals have been declined or recommended for funding within six months of the submission deadline. Notification of the nature of final actions on CAREER proposals will be made electronically to the institution's sponsored projects office in the case of awards, and to applicants in the case of declinations.

Abstracts of CAREER awards for all competitions (past and current) can be accessed from the CAREER Web page at http://www.nsf.gov/home/crssprgm/career/press.htm.

Context Statement

NATIONAL SCIENCE FOUNDATION

Division of Computer and Network Systems CISE Directorate

General Information for Applicants

CAREER FY 2004

The Computer and Network Systems Division received one hundred twenty-two CAREER proposals for the FY2004 competition. The proposals were grouped into five research areas with each area reviewed by a separate panel of experts. About 28% of the CAREER proposals will be awarded.

Each proposal was reviewed by a panel and supplemented when necessary by ad hoc reviews when additional expertise for a particular proposal was needed. The panel based its evaluation of the proposal on the intellectual merit of the proposed activity and its relevance to the CAREER program objectives and goals, the broader impacts, and the integration of research and education. At the end of the panel, the panelists reached a consensus as to a recommendation to NSF. The recommendation of the panel was based on substantive comments from both individual reviews and a collective assessment of the proposal by the panel.

All of the reviews for your proposal are available electronically on FastLane? . In reading them, please keep in mind that the reviewers are addressing their comments primarily to NSF, not to the principal investigator. Although many reviewers do provide helpful information, they sometimes make remarks without giving detailed references or providing specific suggestions for improvement. Some reviews may contain non-substantial, irrelevant or erroneous statements that the program office did not use; such comments are so marked.

Decisions about particular proposals are often very difficult, and factors other than the reviewers' comments and rating enter into the decision. The overall merit of the proposed research activities, the significance and likely impact of the proposed research activities, the principal investigators' competence and the broader impact of the proposed activities are critical considerations. Maintaining an appropriate balance among sub fields, the availability of other funding, and the total amount of funds available to the program are also important decision factors.

Information about reconsideration of declined proposals is found in the NSF's Grant Policy Manual, which should be available at your institution, usually at the office that formally submitted your proposal and on the web at http://www.nsf.gov/pubsys/ods/getpub.cfm?nsf02151.

Panel Summary

Proposal Number: 0347392

Panel Summary: The panel believes that the proposed stealth multicast approach is novel and has some merits. It might lead to a gradual adoption of IP multicast.

The PI should be aware of some of the caveats associated with his approach. In particular, while stealth multicast will result in bandwidth savings for ISPs, it will also incur nonnegligible processing (and storage) overhead at edge routers. The PI needs to consider more carefully the scalability issues and find ways to ameliorate them.

Since the PI envisions stealth multicast as a step towards wide deployment of IP multicast, he needs to be aware of the real reasons that have so far prevented ISPs from "turning on" the IP multicast capability in their routers. The real beneficiaries of IP multicast are end users, not ISPs (ISPs prefer to sell more bandwidth to end users). It is recommended that the PI accounts for the pricing aspects in his research.

Some panelists raised concerns about the potential bandwidth gain that can be achieved. So far, multicast traffic constitutes a small portion of the overall traffic volume. Achieving 10 to 15% reduction in bandwidth usage of multicast may not be significant, but the panel acknowledges that the situation could change in the future.

Overall, the PI articulated a good research plan, with good integration of education and outreach activities.

Panel Recommendation: Competitive

Review 1

Rating: Very Good

What is the intellectual merit of the proposed activity? 



What are the broader impacts of the proposed activity? 


Summary Statement 

In this proposal the PI has tried to compromise between multicast and caching-based approaches for bandwidth conservation in the Internet. He has proposed a new and novel approach called stealth multicasting that can be adopted in the Internet domains to provide 'localized multicast' within the domain. Variations of this approach can be made to extend the scope of this approach to end host or inter-domain cases. 

Stealth multicasting is a very novel idea and the PI has provided adequate motivation for the approach. This approach, unlike multicasting, does have an incentive for the Internet service providers. 

The PI has given sufficient thought on the approach and has done a preliminary study for estimating the benefits of the approach. The implementation issues and several possible extensions of the scheme have been addressed. 

The PI has very high potential as evident from the novelty of the idea and the extent to which the ideas have been thought in depth. 

This is a very well-written proposal. 

The proposal has one of the strongest educational components. The PI seems to be serious about education. He has also published papers on educational issues.

Review 2

Rating: Very Good

What is the intellectual merit of the proposed activity? 

The PI plans to combine IP multicast with data cache approaches to improve bandwidth efficiency in the Internet. 
By improving the bandwidth efficiency, it is expected that QoS improves in the network too. 
The solution is applicable to a single domain. 
Inter-domain is also studied to assess the advantages of the proposed technique in a practical real environment. 


What are the broader impacts of the proposed activity? 

Initial simulation-based results indicate potentially significant advantages (improved bandwidth efficiency). 
If generalized to various scenarios, and various applications, the proposed approach could yield improved network performance. 
Open-source prototypes will be developed and made available based on Linux and COTS hardware. 


Summary Statement 

The proposal is well written. 
The problem is clearly stated. 
The proposed solution is innovative, and deserves further analysis. 
Initial results obtained by the PI are encouraging. 
The proposed plan for execution of the project seems reasonable.

Review 3

Rating: Good

What is the intellectual merit of the proposed activity? 

The PI proposes a new paradigm of doing multicasting in existing Internet while not relying on the availability of multicasting protocols in the routers. 

What are the broader impacts of the proposed activity? 

As the PI claims the proposed research will offer a catalyst for the deployment of IP multicast and hence exploit the considerable body of research devoted to multicast. 

Summary Statement 

This is an interesting and important research area. The proposal is clearly written in objectives, issues, approaches, and integration of research and education plan. The PI has thoroughly listed research issues and has demonstrated some preliminary research results.

Review 4

Rating: Good

What is the intellectual merit of the proposed activity? 

This proposal centers on the concept of Stealth Multicast, a proposed new approach to implementing multicast in the internet. The idea is that a network service provider could identify copies of the same packet being sent to different destinations and replace them with a single multicast packet forwarded toward the same destinations. The copies of these multicast packets would be converted back to unicast before they leave the network service provider's domain, making the conversion to multicast and the reconversion back to unicast transparent to the end user. 

I think the PI seriously misunderstands the reasons that multicast has yet to be deployed in the internet. The simple fact is that network service providers have no positive incentive to deploy multicast while they have a very strong disincentive. The main beneficiary of a multicast service is the sender. A multicast sender can send large volumes of information through a relatively small bandwidth internet connection. This is terrible for the network service provider, who is paid, not based on the amount of data delivered, but based on the bandwidth of customers' network connections. Given the current pricing models, it is to the network service provider's advantage that customers continue to use unicast, since multicast use can only lead to a decrease in revenue. The trouble with current internet multicast is that it does not lend itself to a charging model in which a network service provider is compensated on the basis of the number of packets delivered. To enable a shift to such a model, the multicast mechanisms must be extended to enable reliable end-to-end accounting of multicast sessions and there must be agreements on how to share the revenue generated by multicast sessions among the various network service providers supporting them. Given these issues, I think there is little possibility for stealth multicast leading to wide-spread acceptance of network multicast. 

However, stealth multicast might still have some appeal to a network service provider that happens to carry large volumes of traffic that is naturally multicast in nature. I think the PI deludes himself that there is a large volume of such traffic. I expect the most significant source of this type of traffic is internet radio stations. It seems unlikely that the total traffic from this type of sources is even 1% of the total traffic on the internet, giving a network service provider very little incentive to install the very complex mechanisms that would almost certainly be required to implement stealth multicast. While this might conceivably change in the future, it's wildly unlikely that more than 10% of internet traffic will ever be multicast in nature (even assuming universal deployment of IP multicast). So a network provider that deploys stealth multicast cannot realistically expect a very large cost savings. 

What are the broader impacts of the proposed activity? 

Education of graduate students. 

Summary Statement 

This proposal centers on the concept of stealth multicasting, a proposed mechanisms for using multicast as a bandwidth-saving mechanism within a service provider's network domain.

Review 5

Rating: Good

What is the intellectual merit of the proposed activity? 

The proposed stealth multicast concept is an interesting approach, but may only be needed on the short term. Essentially, stealth 
multicast is a form of intra-domain multicast, whereby only edge routers of a domain are involved (ingress routers act as merging points and egress routers act as splitting points). 
The approach is sub-optimal, as the merging/splitting points are done on a per-domain basis (whereas, true IP multicast takes the whole network into account in one shot). Also, the stealth multicast approach is not fully network-transparent as claimed in the proposal, 
since edge routers need to be aware of it and need to do extra processing of packets. 

Since multicast traffic still represents a tiny portion of the overall network traffic, I do not continue to rely on ALP for the short term (which involves building multicast-aware applications), 
and eventually migrate to a true network-layer multicast. The PI needs to make a stronger argument in support of his proposed stealth multicast approach. 

Another major concern is the coordination between stealth multicast and "inter-domain peering". The latter is intended to extend the multicast capabilities beyond a single domain, and allow 
adjacent domains to coordinate their multicast support. That is a reasonble strategy, but one that needs to be carefully treated. Specifically, when trying to create an end-to-end multicast 
solution out of intra-domain multicast, one needs to make sure that loops are not created (proposal does not address this point). 

The proposed work on virtual group management is too sketchy to judge its merits. For example, the PI should outline the type of "tuneable algorithms" that will be used to predict the performance. The same thing for the inter-domain peering techniques; the PI does a good job 
in identifying the issues involved in such peering (QoS impact, security, protocol development), but gives no specifics as to the approaches that will be employed. The PI, nonetheless, provides a convincing proof-of-concept version of his ideas for the stealth multicast (the MYDEKI), which gives me some faith in his ability to pursue the proposed work. 

The proposed research on "exploiting conservation" is interesting, but the claims may seem too ambitious. Part of this work is somewhat similar to packet-header compression, which has been 
addressed in the literature. Although the PI's proposed research in this area is related to eliminating the duplicate payloads but keeping the distinct headers, he could leverage some of the existing work on packet compression. The ideas related to enhancing the 
protocol stack (and making it more amenable to "bandwidth conservation") are appealing. In general, the concept of application-assisted detection schemes for aligning redundant data 
is a direction worth pursuing. 

One minor comment: it is not clear to me why packets will become smaller as we start using bandwidth conservation techniques. Common logic would say it is the other way around. 

PI's qualifications: Based on his prior work, the PI is qualified to conduct the proposed research (although his publication record is weak and lacks strong journal and conference articles). The available facilities are reasonble for the proposed effort. 

Budget is reasonable. 



What are the broader impacts of the proposed activity? 


The proposed research may accelerate the deployment of IP multicast in the Internet. This can result in significant bandwidth savings for ISPs, which may translate into cost reduction for users of multicast-oriented applications (e.g., teleconferencing). 

The project will involve the training of undergraduate and graduate students. Software generated from the project may be used by other researchers. 

The PI's educational plan is sound. His prior experience in class development (refs. [24] to [26]) puts him in a strong position to integrate his research agenda in course material and classroom instruction. 



Summary Statement 

The PI's research agenda is focused on developing new approaches for multicast that are transparent to both the application and the network. IP multicast has not yet been widely deployed in the Internet, due to scalability and performance concerns. Application-layer multicast (ALM) is a reasonble alternative, but is hurrily dissmissed by the PI because it requires application support (i.e., the application has to be modified to support multicast). Accordingly, 
the PI proposes the concept of "stealth multicast", which the PI claims to be transparent to both the IP network and the end user. To extend the capability of "stealth multicast" beyond a single domain, the PI proposes to investigate "inter-domain peering" techniques. He also proposes techniques for bandwidth reduction (e.g., network stack development).

Response - Program Manager

• Striegel-FY03-Questions.doc: Response doc - program manager