Bart Smaalders wrote: > Sarah Jelinek wrote: > >> I didn't assume the central issue was corruption. I assumed the central >> issue was the need to write to the contents file for every package >> add/remove and that this was a time consuming process, basically a >> performance issue. Peter's analysis does indicate that this does take >> ~1/3 of the install time. > > > Simply put, installing a 10K package on my machine requires 25MB + 10K > of IO. > > Watch an install sometime; the whole thing starts off pretty quickly, > but package by package, slows down as the contents file gets larger and > larger. >
I think Peter demonstrated pretty clearly that he could reduce that problem very greatly for the installation case merely by re-ordering the package operations to halve the I/O. But installatopm performance doesn't have to depend strongly on packaging performance or writing to the contents file. That's how we've implemented it today, but not what I expect we'd do for the future. If you remove that scenario from the equation, packaging performance impacts due to the contents file implementation are a lot less interesting. Dave
