Tycoon gLite Integration

This page is intended to help the communication between Tycoon Grid users and developers working with the Tycoon gLite integration. Design documents and current status of the efforts will be posted continously here.

Roadmap

1. Synchronize Source: Creation of Wiki. Put latest version in Mercurial (2 weeks)

2. Set up Testbed: Install latest version and set up automated tests (2 weeks)

3. SLC 4.0: Auctioneers and VMs should be ported to run on Scientific Linux Cern 4. The delivery of this depends on whether a Xen port is necessary (2-6 weeks)

4. Custom VM images: Add support for gLite WN and CE images that can easily be reused when deploying a Tycoon gLite cluster (4 weeks)

5. Milestone Release 0.5.4: Tag the Tycoon version in mercurial that has support for gLite as described above (July 10-24)

Status

KTH - Updated initial BalticGrid code to work with newest version of Tycoon (best response redistribution agent still not working with the latest code).
- Synching up mercurial version.
- Installation Manual Written.
- gLite SL VM image created to be used when creating Worker Nodes and Compute Elements.
CERN - Presentation given during the OGF/EGEE '07 User Forum (Manchester).
- Latest KTH version of integration tested.
- CERN/gLite Computing Element and Worker Node images ready.
- OS Farm: repository and generator for OS images that can be used in Xen Virtual Machines (http://cern.ch/osfarm).
- Doing scalability tests. Scalability tests done.
- Complete new implementation of Tycoon-gLite integration with GUI and auto-configuration of the nodes. This implementation is not a modification of Tycoon therefore we don't need to modify it with every new version.
- Presentation given during the EGEE '07 Conference (Budapest) within the Business track.
- Poster presented during the Distributed Computing Workshop (London, UK).
HPL- Wiki set up and initial set of presentations posted.
- Next step to install the latest version from KTH at HPL and set up some auto tests

Presentations

Mailing List

All mail regarding the Tycoon Glite integration should be sent to the Tycoon Developer's Mailing List.

Source Code Modifications

All patches and requests for Mercurial commit rights should be sent to thomas.e.sandholm(at)hp.com and the Tycoon Developer's Mailing List should be cc:d.

Bug Tickets

All tickets concerning bugs and feature requests should be added to the gLite Milestone.

FAQ

Q1: Who is the 'VO manager'?
A1: It is a physical person managing a cluster where VOs can run jobs. The VO manager is responsible for creating the WN and CE virtual machines. Future versions will also automate the dynamic funding of these VMs using a best response agent.

Q2: Does JDL have to be parsed by the CE?
A2: No, the current version is completely agnostic to job descriptions. It is up to the VO manager to configure the virtual machines to host all the jobs in the VO. Future versions may allow some boosting of performance on a job level too.

Q3: Why does the CE need to be virtualized?
A3: They don't strictly have to but it simplifies the dynamic creation of them by the VO manager.

Q4: We have the agreement with a VO that it will receive X CPUs at our site no matter what. Can Tycoon make such reservations or will it be able to make them in the future?
A4: Currently there is no inherent support in Tycoon to give such guarantees, although it is research in progress. It is however easy to set up such guarantees in a controlled environment where only a fixed set of VO WNs can be created with a limited or fixed funding level on each physical machine in the cluster. If a fixed set of VOs with fixed relative funding levels is used you can get the same guarantees you would get in a non-virtulized dedicated gLite cluster.

Q5: Is the public Tycoon network used or only some private Tycoon based network?
A5: Initially semi-private networks are likely to be used to offer better performance guarantees (see A4), but long term both WNs and CEs should be able to reside anywhere in the Tycoon network.

Attachments