Prof. Giuseppe Lipari and I have found an upper bound to the lateness of real-time tasks scheduled with global EDF on multiprocessors. We got the following paper published on this topic.
Unfortunately a bug has been found in one of the proofs (credits: UmaMaheswari C. Devi, coauthor of a paper on the same topic, presented at the same conference).
We prepared a backup Technical Report, containing only a subset of the sane proofs of the published paper. This subset leads to a higher bound with respect to the one in the published paper.
We are also working on a new Technical Report containing a slightly different set of proofs, which should not be affected by that error. Contact me for details.
Finally, you can click here to download an extended version of the original paper in which we highlighted the point where the bug is.