Search

Why does this application require zero redundancy, i.e. no comparison of results across devices?

Unlike many of our research projects, FightAIDS@Home – Phase 2 does not require redundancy, where the same research task is sent to two devices and the results are compared for consistency. Instead, this project will be using various processing metrics during the computation of a research task to validate that the task is progressing without errors.

What is validation?

World Community Grid is a volunteer computing grid. This means that work is being sent to computing devices that are outside the control of World Community Grid. Most devices that perform this work are reliable. However, there are a few devices that are not reliable due to things such as users over-clocking their machines, memory errors, disk errors, CPU errors or viruses being present. This means that the results returned need to be validated to make sure that they represent the correct answer.

We perform three different types of validation at World Community Grid:
 

• Redundant Computations: In this type of validation, two copies of the work unit are sent to members devices. Once both results are returned, they are compared to ensure that the results are identical. If they are, then the result is accepted. If they are not identical, then additional copies are sent until several devices agree on what the result should be. This policy establishes a very high level of confidence in the reliability of the results. Mapping Cancer Markers and Uncovering Genome Mysteries are examples of projects that use this technique.
• Single Validation - Type 1: In this type of validation, only one copy of a work unit will be sent to a device if the device is "trusted", that is, if it has been participating long enough and returning good results. If the device is not trusted, then it will still be assigned the work unit, but a second copy will be sent to another device and the rules for redundant computation above apply. As a precaution, the research code computes certain items that allow us to quickly check on the server if the computation is likely to have finished correctly. Additionally, trusted devices are randomly sampled to have their results double-checked. These techniques provide a very high level of confidence in the reliability of the results. FightAIDS@Home and Outsmart Ebola Together are examples of projects that have used this technique.
• Single Validation - Type 2: This is similar to Single Validation - Type 1 except that due to the fact that different results are generated each time the work unit is run (due to the research techniques applied in the application), we send out many copies of each work unit. We currently do not have any research projects utilizing this technique.