- if (la_hold_total < LA_HOLD_TOTAL && la_hold_total < nmbclust / 64) { + if (la_hold_total < nmbclust / 64) {
I have disagreed with claudio about this aspect of the diff. The refresh attempt is the crucial problem to fix, because of what it has done to tcp with slowstart etc. If refresh works, we may not need to increase the hold mbufs count at all. Secondly, I don't like how this is coupled to the MD nmbclust value. Some architectures have that very large, others very small, for other reasons. Bringing it into play here isn't right.