If possible, could you please go into a bit more detail regarding numerical instability? Is the instability a general hazard with the ellipsoid method, or parti

If possible, could you please go into a bit more detail regarding numerical instability? Is the instability a general hazard with the ellipsoid method, or particular to to the Gr\"{o}tschel et al algorithms? I've heard of the odd occasion where there is a common belief that an old algorithm does not perform well in practice, based on older experimental studies, but it turns out that the algorithm is actually quite effective, but perhaps only when combined with a few other good implementation decisions.