Version 1 (modified by Vic Thomas, 9 years ago) (diff)


The Experimenter Workflow Tools and Services Working Group met on 17 November 2009 at the GEC held in Salt Lake City, Utah.

Vic Thomas, the GPO systems engineer supporting this WG, introduced the new working group chairs (Prof. Jeannie Albrecht of Williams College and Prof. James Griffioen of the U. of Kentucky) and thanked them for agreeing to chair the WG. Vic went over the charter of the working group, documents produced to date by the WG, and WG plans for Spiral 2.

In Spiral 2 of GENI the WG will focus on identifying tools and services needs of the experiments planned for this spiral. The WG meeting agenda was designed to understand the needs of the experimenters and assess how well these needs are being met or will be met by tool and control framework developers.

The first part of the meeting agenda included four talks by researchers planning on doing experiments on GENI or building services that live within a GENI slice and hence look a lot like experiments. The WG chairs, Jeannie and Jim, had asked the speakers to provide a concise summary of their experiments and then talk about their "experimentation requirements". In particular, they were asked to address four possible dimensions of the "experiment space":

  • Duration: Is the intended experiment short-lived (e.g., requiring more interactive features) or long-lived (e.g., requiring (error) event notification rather than interactive support)?
  • Layering: What layer of abstraction does the experiment want to control? (e.g., does it need to control the topology and link characteristics? or does it need to completely replace IP with a new packet format and router processing? does it need to control/install in-network transport services e.g., retransmission or encryption, or ... service? or does it need to control placement of application-level (e.g., P2P) services? etc)?
  • Domain: What type(s) of resources does the experiment need? (e.g., wireless, wired, mobile, sensor, storage, ..., or some combination)
  • Scale: How many resources comprise the experiment? (e.g., very large experiments will need tools that help manage large sets of resources collectively rather than individually; whereas small experiments may want tools that interact with each resource individually).

Felix Wu described his experiment "Davis Social Links" that plans on hosting Facebook applications on GENI. Needs he identified included support for 24x7 operations, ability to quickly grow or shrink a slice to adapt to quickly changing resource requirements of the applications, the ability to load custom OS images onto nodes and the ability to replace the IP-based Layer 3 protocols with his own Layer 3 protocol and routing mechanisms.

Emmanuel Cecchet described the data intensive applications being developed by his group. These applications involve high bandwidth sensors such as weather radars. He pointed out the need for large amounts of storage with high-bandwidth access from the testbed. Cloud storage may not be suitable for his applications because of limitations of bandwidth to/from the storage.

Sonia Fahmy described the shared measurement system she is developing for GENI. This is an active measurement system and she needs the ability to install custom probes in or near routers. These probes are intended to inject and measure traffic close to the routers and core of the network.

Finally, Beth Plale described her service to collect and record provenance information for GENI experiments: Their lineage or historical trace that can capture experiment conditions, time ordering, and relationships within the experiment and across the application and infrastructure layer. The expectation is much of this information will be inferred and will not be explicitly provided by the experimenter. Some of this information is available to the control framework and it would be helpful to be able to get at this information.

These four talks were followed by panel discussion moderated by Jeannie Albrecht. The panelists included representatives of each of the four control frameworks: Ivan Seskar (ORBIT), Rob Ricci (ProtoGENI), Larry Peterson (PlanetLab) and Jeff Chase (ORCA). The panelists were asked to address specific needs identified by experimenters. Some needs, such as supporting 24x7 operations, was deemed by the panel to be outside the scope of the control framework; experiments but be prepared to deal with failures of resources used by their experiments. Other needs such as loading of custom OS images are already supported by some control frameworks. The WG chairs felt this was the start of a useful dialog and there needs to be more discussion on specific things the control framework developers can do to make it easier for experimenters or those developing tools for experimenters.

The meeting agenda and slides presented at the meeting are at