wiki:ExptSvcsWGGEC3Minutes

Version 1 (modified by Vic Thomas, 11 years ago) (diff)

--

Minutes of the GENI Experimenter Workflow and Services

Working Group Meeting

30 October 2008

Recorded by Vic Thomas (GPO System Engineer)

The GENI Experiment Services and Workflow Working Group met as part of the GEC-3 conference hosted by HP Labs, Palo Alto, California.

Meeting agenda:

  • Welcome and meeting objectives -- Jeff Chase, Duke University (WG chair)
  • Lifecycle of a GENI Experiment -- Vic Thomas, GPO (System engineer supporting the WG)
  • Project talk: GENI Experiment Control using GUSH-- Jeannie Albrecht, Williams College
  • Lightning talk: Slice Control and Programmability in Wireless Sensor Networks -- Hongwei Zhang, Ohio State University
  • Lightning talk: Orbit Management, Control & Measurement Framework -- Max Ott, NICTA
  • Project talk: Raven - A Provisioning Service for GENI -- John Hartman, University of Arizona
  • Project talk: Instrumentation Tools for a GENI Prototype -- Jim Griffioen, University of Kentucky
  • Lightning talk: Digital Object Architecture -- Giridhar Manepalli, CNRI
  • Project talk: Data Plane Measurements -- Deniz Gurkan, University of Houston
  • Next steps for the WG -- Vic Thomas, GPO

Jeff Chase described the scope and activities of the working group. He showed where this WG fit into the overall GENI architecture and introduced the talks to be presented at this meeting. Jeff's presentation includes a number of questions to be addressed by this WG. The presentation also includes strawman GENI usage scenarios--- canonical usage scenarios that can be used for discussions on experimenter workflow and services. Time constraints prevented Jeff from presenting this material; he hopes these issues will be discussed on the WG mailing list and in future meetings. Presentation: http://groups.geni.net/geni/attachment/wiki/presentations/JeffChase-ExptSvcsWG.ppt

Vic Thomas presented his view of the workflow for an experimenter using GENI. For each step of the workflow he identified tools and services needed to support that step. The objective of the presentation was to get feedback on whether this workflow was consistent with the thinking of the WG participants, to understand how proposed projects and ideas fit into this workflow, and to identify areas of disagreement. A person in the room pointed out that it would be useful to have a tool that schedules experiments based on resource needs and availability. Presentation URL: http://groups.geni.net/geni/attachment/wiki/presentations/ExptLifeCycle-GEC3-v2.ppt.

Jeannie Albrecht described her project that is developing abstractions and tools for addressing the challenges of managing multiple applications. The project is developing Gush, a distributed experiment management infrastructure based on her previous work called Plush. Gush supports multiple steps in the experimenter's workflow: Describing an experiment using application building blocks, acquiring resources to run the experiment, configuring the resources, starting the experiment, monitoring the experiment (including failure handling), and finally cleaning up the resources used. Gush is being built to integrate with the PlanetLab control framework. Presentation URL: http://groups.geni.net/geni/attachment/wiki/presentations/jeannie-gush-gec3.ppt.

Hongwei Zhang gave a lightning talk on experimenter tools available on Sensei, the wireless sensor network testbed from Ohio State. This testbed provides "slice control" services. These services allow slices to be moved, i.e. the set of sensor nodes that belong to a slice can be changed dynamically to track events of interest to the experiment. They also help experimenters maintain slice invariants and collect slice statistics. Presentation URL: http://groups.geni.net/geni/attachment/wiki/presentations/Zhang%20-%20SliceControl-Programmability.ppt.

Max Ott gave a lightning talk on experiment related measurements on Orbit. The experimenter describes the experiment using a Ruby-based language. The information to be collected is also specified using this language. The measurements are collected in a relational database. Orbit provides APIs for analyzing the data collected. Presentation URL: http://groups.geni.net/geni/attachment/wiki/presentations/OrbitMeasurementLayer-Max.ppt

John Hartman gave a project talk on Raven for managing software deployments for GENI experiments. Raven supports the deployment of software to programmable components and updating this software while the experiment runs. It builds on the Stork project and like Stork is targeted to the PlanetLab control framework. Raven supports the creation and distribution of packages and metadata files with configuration instructions. Raven will also provide tools for resource discovery and allocation. Raven and Gush are complementary projects: Raven addresses software management while Gush addresses slice management. Presentation URL: http://groups.geni.net/geni/attachment/wiki/presentations/Raven-hartman.pdf.

Jim Griffioen gave a project talk on the easy to use instrumentation tools they are building so GENI experimenters have a better understanding on the runtime behaviour of their experiments. They are modifying the Edulab web interface to include experiment monitoring pages. Examples of information that can be monitored include OS performance/load, traffic graphs and packet traces. Pictures and GUIs are used to specify the information to be collected. This project complements Paul Barford's (U. of Wisconsin) project that is looking at measurement data collection. Presentation URL: http://groups.geni.net/geni/attachment/wiki/presentations/GEC3_Kentucky.pdf.

Giridhar Manepalli gave a lightening talk on CNRI's Digital Object Architecture that includes a handle system, a digital object repository and a digital object registry. The handle system is a distributed and secure policy management system. It can potentially be used for managing GENI principal records. The registry and repository can potentially provide registration, discovery and storage services to GENI entities such as principal, component and slice registries, and code artifacts. Control frameworks interested in using the Digital Object system should contact Larry Lannom or Giridhar Manepalli of CNRI. Presentation URL: http://groups.geni.net/geni/attachment/wiki/presentations/GEC3-CNRI-LightningTalk.ppt.

Deniz Gurkan gave a project talk providing GENI experimenters with measurements at the optical layer. Examples of measurements include bit error rate, optical power, and SONET and Gigabit Ethernet performance monitoring. This measurement work is especially useful for non-IP networking experiments. The project will identify feasible optical layer measurement technologies for GENI and will design a framework that will allow experimenters to remotely configure and gather instrumentation data. Presentation URL: http://groups.geni.net/geni/attachment/wiki/presentations/GENI%20GEC3%20DGurkan%20D_Meas-final.ppt.

Finally, Vic Thomas listed the formal deliverable of this working group. He, as the GPO system engineer, will draft these documents but would appreciate active community involvement. At the very least community members should comment on the documents. Text contributions to the document will be greatly appreciated. WG discussions will be carried out on the WG mailing list.