Are you ditching just the cache, or the data structure as well?  Bug 1867
isn't rife with details, but it could be that the cache itself doesn't
really buy us much over just reading the catalog files and creating the
objects directly from that, rather than from the pickle.

So if the cache is making that little difference, I'd ditch it.  It might
be nice to find out what it's like on a crappier machine, though.  Again,
the bug doesn't have many instances of timings, and I've no idea what the
original filer's machine was like, but 7s seems reasonable, even if it'd be
nice to get it faster.

Note that "pkg list" (not -a) should be much faster, as it now only has to
do a readdir() on a single directory, rather than stat()ing files zillions
of files, but it does need to read the catalog in order to get the
upgradable column filled in.  Though I'll note that using the dtrace
command in bug 1899 shows that we're not really doing fewer opens or stats
than we were at the time, but brk() is getting called significantly more.

I'll note that on my work desktop, "pkg list" takes 12s, and "pkg list -af"
takes about 17.  (I have the dev and extra repos, plus whatever's on cyber,
and a smattering of other packages.) On my machine at home, which is
noticeably slower, I only have the dev repo and a handful of other things,
and I get about 5.5s and 9s, respectively, so cyber must be adding quite a
load.

Danek
_______________________________________________
pkg-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/pkg-discuss

Reply via email to