Again, inspiring the uninspired

The GUADEC folks keep harassing me for a photo and a bio and a urine sample. It is a neverending tirade from them. They just keep wanting more and more but never give anything. Blah, blah, blah! Anyhow, I did not have any self-portraits on hand--who does?--so I had to recruit Joey and my pinata, not coincidently named Lil' Joey, for a quick shoot in the office.

Lil' Joey!

Another list from Tomboy.

First names that I have had: Robert.

Venezuela's Chavez finally figures out to which ideology he subscribes. In the process, he demonstrates that he has no understanding whatsoever of economics. Or of history, for that matter.

Lil' Joey!

Yesterday, I hacked up a kernel patch and corresponding pmap(1) support for per-VMA RSS statistics. Today, I added per-VMA anonymous statistics.

This allows us to see exactly how much of a given mapping is physically in-core (for example, which pages of this 289MB file mapping are we actually using?) and how much of a mapping is anonymous (for example, which of these shared pages did we COW?). We have always had these statistics on a process-wide level, but never per-mapping. Our developers definitely have a need to see a per-VMA break down of total size, RSS, and anonymous versus file-backed pages.

The patch is a bit of a hack and the accounting is pretty naive, but for curiosity in user-space, it works.

    $ ./pmap 16449
    16449: gvim
    Start         Size       RSS      Anon  Perm  Mapping
    08048000     2000K     1076K        0K  r-xp  /usr/X11R6/bin/gvim
    0823c000      204K       64K       60K  rw-p  /usr/X11R6/bin/gvim
    0826f000     2552K      260K      260K  rw-p  [ anon ]

Good accounting or not, this definitely highlights how badly we need optimal function reordering, e.g. GROPE.

SUSE users can just grab a kernel package.

List of things that my second ex-wife took when she left me: The oven mitts, all of our fabric starch, the wooden leg we had on on the mantle, my pride.