wiki:PlasticSlices/Reports

Version 6 (modified by Josh Smift, 13 years ago) (diff)

--

As part of the Plastic Slices project, the GPO will publish three interim status reports, and a final report for presentation at GEC 11.

Interim report : 2011-05-09

The project is off to a good start, especially given the aggressive schedule for ramping up from scattered pieces to a full integrated mesoscale GENI infrastructure. Five campuses are now online and ready to go, along with the two backbone providers, and the three other campuses are close. We've begun creating some slices, although we've been slowed by some obstacles surrounding the complexity of one person running ten slices with ten separate OpenFlow controllers, etc. Monitoring data is starting to flow into the GMOC, with more campuses coming online this week, and we're on track to complete the first baseline by next Monday.

Links of interest:

Some graphs of interest within SNAPP:

  • All network traffic for slivers on reporting MyPLC plnodes:
    • Click the Browse tab at the top.
    • For Browse By, select "Location".
    • For Location, select "All".
    • For Type of Data, select "auto-16-plnode_sliver_network-15sec".
    • Click "Aggregate Graph".
  • All types of recent activity on a particular slice:
    • Click the Search tab at the top.
    • Into the text box, type a slice name (e.g. "plastic-102").
    • Click "All Graphs Side by Side".
    • Scroll down a bunch.
  • DPIDs on the GPO Lab Expedient server (tulum.gpolab.bbn.com):
    • Click the Browse tab at the top.
    • For Browse By, select "Location".
    • For Location, select "gpolab".
    • For gpolab, select "tulum.gpolab.bbn.com".
    • For Type of Data, select "auto-68-flowvisor_dpid_state-15 sec".
    • Click "All Graphs Side by Side".

Our next interim report will include results from the first four baselines.

Interim report : 2011-06-06

Things are going well so far. All eight campuses, and both backbone providers, are up and running, and the first four baselines were completed very close to as-intended. The obstacles along the way have proved informative as well; for example, we tried some different approaches to the problem of how to log the results of the experiments running in each slice. We'll share more details about that in the final report.

We've been using five artificial experiments, each with somewhat different traffic characteristics, each on two slices (one slice using each of the core VLANs). The first three baselines included large bursts of traffic (when each experiment sent about 1 GB of data more or less as fast as it could); the fourth included a steadier flow of constant traffic. We concluded that there weren't any non-artificial experiments ready to run in this context, so we're staying with the same artificial experiments for the next four baselines.

Comprehensive monitoring data -- about slices, slivers, servers, nodes, etc -- is flowing to the GMOC, and is visible online via the SNAPP UI (http://gmoc-dev.grnoc.iu.edu/api-demo/data/). We've used the GMOC's XMLRPC interface to download raw data and process it for easier slice-specific analyses of Plastic Slices experiments. We think that these type of web page displays may be useful to other GENI campuses and experimenters who want to create their own customized pages, and we'll be sharing the details in our final report.

Our next interim report will include results from the next four baselines, and more information about our plans for the final few weeks of the project.

Interim report : 2011-07-11

Things continue to go reasonably well. The baselines continue to encounter challenges, and the challenges continue to prove informative, turning up bugs in software (many of which have been fixed, or have fixes in the process of being released), flaws in various approaches to experiment management (which we continue to refine), and issues with operational procedures (which are being improved for future testing). We'll share many more details about all of those in the final report.

We've continued to use five artificial experiments, each with somewhat different traffic characteristics, each on two slices (one slice using each of the core VLANs). We've now written up one set of variations that include large bursts of traffic (when each experiment sends about 1 GB of data, more or less as fast as it could), and another set that sends a steady stream of constant traffic. We reviewed again whether there were any non-artificial experiments ready to run in this context, and while early adopters continue to make solid progress, we again concluded that there weren't any that were quite ready yet, so we stayed with the artificial experiments for the last four baselines.

We plan to re-run some of the later baseline tests again after GEC 11, because we've made software, configuration, and procedure changes during the course of the Plastic Slices project that should improve the results for the mesoscale infrastructure. We also expect to define some new tests, based on operational feedback from the existing sites, and on preparations for replacing artificial experiments with real ones. Some ideas we already have in mind:

  • More Emergency Stop tests (like Baseline 7).
  • More high-volume slice / sliver creation tests (like Baseline 8).
  • Testing the effect of dynamic ARP on OpenFlow slivers.

Our final report will include more details about those, plus many more details about the various things we've described in passing in our interim reports. We'll also be doing a presentation on the project and the results at GEC 11 in Denver.

Final report

(forthcoming)