I have several suggestions to gain better performance and less surprise
for some applications using /proc. As we know, top shows all process on
all nodes and other monitoring utilities have each instance show all the
Shutdown is slow because some services are searching all processes on
The proposal is that /proc show local information for non pid files
(mostly what it does now) and readdir on /proc only shows processes
locally while lookup on /proc/<pid> is clusterwide (so debuggers etc.
work no matter where processes are running).
I contend that with this change most things will just work and "top"
will work as is (but will only be showing local information and local
processes). ps already has to change, so we need to discuss what it
should look like and how it should get it's information.
An additional proposal is that we create a /cproc. Only clusterwide
info would be in there (pids and sysvipc and ??). Readdir would show
all the non-kernel-daemon or non-system processes on all nodes (may have
a little definition problem here). ps would generally use this
filesystem to search for processes owned by a given user or terminal,
etc. but withoug the countless=20
kernel threads and prcoesses, it would run much more quickly.
Get latest updates about Open Source Projects, Conferences and News.