Highlights

AssessGrid is software!

Methods and algorithms elaborated in the projects are put into action in software for Risk Management. The project released its first prototype in fall 2007.  The second prototype was delivered during summer 2008. Final result of the project was available in February 2009.

As it is intended for reuse in other projects, the project also provides separately the Negotiation Manager for WS-Agreement protocol.

All the source code is released as Open Source under Apache V2 license.


Safty first in Server Rooms

Risk-aware Broker Service

For the second prototype, the AssessGrid development team focused on enhancing the broker to provide workflow jobs. This requireed also an enhancement of the end-user interface (web portal) to specify and negotiate workflows. On the provider layer, the implementation of the SLA negotiation protocol has been modified in order to better adapt to real market environments. Furthermore, the risk management has been improved in order to achieve a profit-oriented SLA provisioning.
The changes "Risk-aware Broker Service" introduces for stakeholders compared with the first prototype are detailed below.

On the end-user's side:
The portal now allows to define and submit workflow jobs. An editor for specification of workflow applications has been added. Users can interactively and graphically define workflows (adding, modifying or suppressing workflows tasks) by using this editor. Once they are created, workflow application can be submitted and negotiated.

The enhanced Web portal also includes new features:

  • Single jobs can be submitted to be executed “as soon as possible”
  • User credential delegation is now managed using MyProxy
  • Monitoring of job execution and states has been enhanced
  • Usability and look and feel of the application have been improved.

A dedicated page presents in detail the web portail of the second AssessGrid prototype.

For the broker:
An SLA request for a workflow job can be combined into a single SLA request containing a detailed job description for each sub-task and guaranteed maximum probability of failure, earliest start / latest finish time for the entire workflow. The broker negotiates independently with a number of providers in order to fulfil each sub-task requirement and associates these agreements with the end-user’s SLA with the broker.
While SLA requests for sub-tasks can be combined to one SLA offer for the end-user, the execution of SLA sub-tasks with several Grid providers can be negotiated in parallel. If a sub-tasks fails, the broker negotiates a second execution of this task and - if necessary - of affected sub-tasks, thus providing fault tolerance mechanisms.
Failure estimation has also been improved. An independent 3rd party reliability estimate is provided for provider generated probability of failure estimations. These reliability estimates are based on historical SLA records. The provider’s past risk assessments are statistically compared with the corresponding SLA outcomes in order to produce a measure that identifies providers who are systematically overestimating or underestimating the probability of failure for the SLA. The approach uses a weighted average to ensure that more recent SLAs have a greater influence. Finally, this version of the sofware provides automatic updates of the broker’s historical SLA data on completion of SLA jobs between broker and provider, in order that up-to-date reliability assessments can be performed.

From the provider perspective:
A quote machanism has been integrated in the SLA negotiation users can ask for a non-binding quote (including price and probability of failure that can be guaranteed) before creating an agreement. The NegotationManager component's web site contains details on the modified protocol.
A pool of spare resources has been defined in the resource management system. It offers more flexibility and enhanced failure management.
If not all SLAs can be fulfilled, providers can rely on a profit-oriented schedule in which the (in expectation) most profitable jobs are preferred.
Providers are now able to negotiate SLAs with other providers in order to outsource SLA bound jobs which cannot be scheduled due to resource failures that prevent SLA compliant execution of these jobs. Kernel-level checkpointing is also performed for all SLA bound jobs; jobs affected by a resource outage are migrated (internally or externally) and resumed from the latest checkpoint.

Web portal

Changes and improvements of the web portal are described in detail in a separate page.

Downloads

Source code

All the source code of Assessgrid applications can be downloaded in several tar files.