From the perspective of computer architecture, the author provides what look like -- on their face -- good arguments. They quote from Aeter Suleman, who says t

From the perspective of computer architecture, the author provides what look like -- on their face -- good arguments. They quote from Aeter Suleman, who says things like: "they throw off hardware prefetching", or "they reduce the benefit of out-of-order execution". These may have been true on machines of the past, but these are no longer true on modern systems. With the advent of trace caches and runahead execution [1], linked lists are really no longer as painful as they once were. (Indeed, even back in the day, Alpha had low-cost "explicit" runahead-like semantics, where the programmer could specify other work to do while waiting for DRAM; this was usable to accelerate linked-list traversal.) [1] http://users.ece.cmu.edu/~omutlu/pub/mutlu_hpca03.pdf http://users.ece.cmu.edu/~omutlu/pub/mutlu_hpca03.pdf (disclosure: I worked closely with Onur at one point; his Ph.D thesis introduced runahead. I may be excessively biased in favor of that technique :-))