It doesn't look like swap space is the issue. When the command was still running, despite having many cores running R, only 2-3 of them were using any cpu according to htop. I am able to get the program to run in reasonable time (a few minutes a slice) when I cut out extra covariates, leaving only my two main DVs and their interaction. But as soon as I add another covariate, the run time jumps to about 2 hours a slice. Completely specifying the model I wanted puts runtime at 4-8 hours a slice. That model isn't too complex though: 1+group*(age+sex+covariate)
Working with a very patient sysadmin, I've separately updated AFNI and R, with no improvement. After updating snow, afex, and phia within R, the program now quits with model test failure, when it previously ran (albeit slowly). Is there a way I can force R to print the warning messages during the model test before aborting? Is there any specific version of any of these packages that I need to be using?