I will discuss this at the next AFSUG. Briefly, larger chunks are more
desireable for about 90% of cases. Larger chunk sizes will improve
warm cache performance. An implementation deficiency in RX in AFS <
3.2 made large chunk sizes undesireable when slow networks were
involved; this has been fixed in 3.2. For the time being, you should
avoid large chunk sizes if you have network links slower than 56 Kbps
in a commonly used path.
AFS *used* to fetch whole files, once upon a time, but it turned out
to be hard to manage the cache efficiently in that case. The other
issue with very large chunk sizes is that they are counterproductive
for certain applications that process only small parts of very large
files.
