In the previous sections, we have mainly concentrated on the results for SIRS-3. Formally, SIRS-3 is the last subproject to be completed as part of the much larger SIRS project. In this section, we reconsider the SIRS project as a whole and see what has been achieved with respect to the original plans and where further improvements are needed. In particular, we take a look at the dissemination of the results and future plans
One of the original goals of SIRS was to come to a solution for offloading popular FTP sites, notably the NLUUG software archive, which acts as a mirror for many software packages. The brute-force approach by effectively improving network access, storage capacity, and possibly processing power, will generally work (and which has now been successfully applied to the NLUUG site). However, such an approach is not immediately interesting from a scientific and long-term perspective.
The development efforts put into the SIRS projects amount to almost six person-years of work. I estimate that the accompanying research amounts to another four person-years. In addition, there are approximately two person-years of development efforts done researchers that have lead to software that is now integrated into SIRS (notably GIDS and security software), leading to a grand total of 12 person-years. Research and development concerning the Globe location service (at least another eight person-years) have not been included in this calculation.
The scientific output related to SIRS (and again excluding the Globe location service) is quite large: we have published approximately seven papers in conferences and journals. At least two papers are planned for production this year. The combination of the research results and development efforts have improved the visibility of the Globe project as a whole.
In conclusion, the SIRS project from a scientific and development point of view seems to be reasonably successful. However, when evaluating the usability of the results and the impact the project has, there is clearly room for improvement. In the following two sections, we concentrate on two important issues related to making SIRS successful: the dissemination of the results and the future work.
One of the issues for SIRS that has not been dealt with in a satisfactory manner, is the dissemination of the results. It is taking quite some effort to push SIRS/GDN to usage by users other than those that directly or indirectly linked to the Globe project. Let us first take a closer look at what we have done with respect to ``marketing'' SIRS/GDN.
We decided to make the source code for SIRS/GDN available at the end of the SIRS-2 project. At that point, it was possible to set up a simple GDN system that we felt was stable enough for experimentation. It took us quite some effort to reach this point, mainly because the amount of ``baseline'' software (object server, name server, client) was quite substantial. Since December 2001, we have made a new release of SIRS/GDN every 1-2 months.
All the code and the gog have been available from the FTP site at the VU. Also, we have made the package available through SourceForge since April 2001. The total number of downloads and groups that have voluntarily started SIRS/GDN is effectively restricted to our own group. We have had mail from a few developers, but it is obvious that the Globe community is essentially only the group at the VU along with a small circle of friends.
During the course of the SIRS-3 project, it was felt that we should more pro-actively attempt to disseminate the software to interested parties. It was decided to do so by means of a number of demonstrators. At present, GDN is hosting software for Linux (2.4 kernels and the 7.2 updates), Amoeba, and Minix. We are also hosting various Web sites. Experiments have begun to host a large portion of the SourceForge database, replicating it across San Diego (CAIDA), Redwood City (Vixie Enterprises), and Amsterdam (VU). These experiments still need to be completed and require additional development work at the VU.
SIRS/GDN is now running on multiple sites across the Internet. There are various ``local'' sites hosted in The Netherlands; the main international sites are (in alphabetical order):
Meanwhile, we are continuing our efforts to host content. For example, attempts are being made to see whether SIRS/GDN can act as a replicated mirror of www.faqs.org. Also, we have contacted people that may be able to provide us access to Internet2 nodes to expand the SourceForge experiment across more nodes.
Being a group of researchers, we are keen on publishing scientific papers in which SIRS/GDN can play a role. As mentioned above, we now have a handful of papers (and accompanying presentations) on SIRS/GDN. A next presentation is planned for SANE 2002, including an accompanying paper describing the system from the perspective of its users; a paper describing SIRS/GDN is also planned for submission to Wiley's Software Practice & Experience.
Despite these efforts so far, it turns out that disseminating SIRS/GDN to the community requires considerable effort. Although Globe, by now, has established a good reputation, more effort is clearly needed to bring it to the public. It is not obvious what the best strategy is that we can follow. However, we have learned from the current efforts that pro-actively advertising SIRS/GDN is better than just making the package available.
The strategy we are currently following (seeking for sites that are willing to host SIRS/GDN, and letting us manage those nodes) appears to be the best we can do when it comes to demonstrating that SIRS/GDN is for real. We know from colleagues that organizing wide-area experiments often requires a considerable amount of time. However, the results that can be reported at conferences and such tend to help people get convinced that something serious is happening.
In addition, we want to continue our search for hosting content. Unlike the SourceForge experiment in which we focus on large amounts of data, it is presumably more effective for the dissemination of results to concentrate on many different data sources. In this sense, turning SIRS/GDN into a ``cheap'' content delivery network is perhaps the best way to draw attention. Again, we expect that it may take a relatively long time before SIRS/GDN will prove its value to a larger community, but we see no other way to reach this point.
Although SIRS-3 is formally finished, the work on GDN continues. First, as already indicated, we will actively seek for content that can be hosted by GDN. In doing so, we hope to attract attention from content providers seeking for a solution for distributing their content.
Second, and related to hosting real content, we plan to conduct a series of experiments with SIRS/GDN so that we can convince third parties of the technological advances our software has to offer. In addition, experimentation should reveal important faults (bugs), making it possible to improve the robustness of our software. It is already encouraging to see that the performance of our system is comparable to that of standard-configured Apache-based systems.
However, we expect that SIRS/GDN will attract more attention as we continue to enhance its functionality based on our future research. At present, our group is actively conducting research in the following areas:
At present, we have one full-time senior researcher looking into the dynamic replication of Web documents. So far, this research has provided us insight in how and when to evaluate access traces in order to decide what a best replication strategy is. Starting in August 2002, two PhD students will be added to the project. One student will concentrate on the selection of the best location to place a replica. The other student will take a look at dynamically switching the consistency protocol so as to optimize network resources and access delays. A separate scientific programmer will be hired to implement the results. Embedding these results in SIRS/GDN is an obvious choice.
We are currently looking at management issues related to the Globe location service. One postdoc is considering the problem how we can automatically bring up an entire, worldwide-spanning tree without initially having to physically distribute that tree across the Internet as well. This research will certainly influence the (semi-)automatic distribution of GDN across multiple sites.
In addition, a PhD student has recently started to concentrate on more general systems management issues related to distributed systems such as GDN. Again, the results of this result are planned to be embedded in SIRS/GDN, partly to demonstrate the feasibility, but also to make it easier to deploy SIRS/GDN.
In conclusion, we feel that we have just made a start with SIRS/GDN when it comes it its deployment. However, we are continuing the research that has been set out as part of SIRS/GDN, exemplified by the two subprojects mentioned above. Recognizing that deploying SIRS/GDN may be a matter of much time, it is important that our research is set out along the lines of such a deployment.