summaryrefslogtreecommitdiff
path: root/open_issues/resource_management_problems/io_accounting.mdwn
diff options
context:
space:
mode:
Diffstat (limited to 'open_issues/resource_management_problems/io_accounting.mdwn')
-rw-r--r--open_issues/resource_management_problems/io_accounting.mdwn49
1 files changed, 49 insertions, 0 deletions
diff --git a/open_issues/resource_management_problems/io_accounting.mdwn b/open_issues/resource_management_problems/io_accounting.mdwn
new file mode 100644
index 00000000..113b965a
--- /dev/null
+++ b/open_issues/resource_management_problems/io_accounting.mdwn
@@ -0,0 +1,49 @@
+[[!meta copyright="Copyright © 2011 Free Software Foundation, Inc."]]
+
+[[!meta license="""[[!toggle id="license" text="GFDL 1.2+"]][[!toggleable
+id="license" text="Permission is granted to copy, distribute and/or modify this
+document under the terms of the GNU Free Documentation License, Version 1.2 or
+any later version published by the Free Software Foundation; with no Invariant
+Sections, no Front-Cover Texts, and no Back-Cover Texts. A copy of the license
+is included in the section entitled [[GNU Free Documentation
+License|/fdl]]."]]"""]]
+
+IRC, freenode, #hurd, 2011-07-22
+
+ <braunr> an interesting question i've had in mind for a few weeks now is
+ I/O accounting
+ <braunr> what *is* I/O on a microkernel based system ?
+ <braunr> can any cross address space transfer be classified as I/O ?
+
+IRC, freenode, #hurd, 2011-07-29
+
+ < braunr> how does the hurd account I/O ?
+ < youpi> I don't think it does
+ < youpi> not an easy task, actually
+ < youpi> since gnumach has no idea about it
+ < braunr> yes
+ < braunr> another centralization issue
+ < braunr> does network access count as I/O on linux ?
+ < youpi> no
+ < braunr> not even nfs ?
+ < youpi> else you'd get 100% for servers :)
+ < braunr> right
+ < youpi> nfs goes through vfs first
+ < braunr> i'll rephrase my question
+ < youpi> I'd need to check but I believe it can check nfs
+ < braunr> does I/O accounting occur at the vfs level or block layer ?
+ < youpi> I don't know, but I beleive vfs
+ < youpi> (at least that's how I'd do it)
+ < braunr> i don't have any more nfs box to test that :/
+ < braunr> personally i'd do it at the block layer :)
+ < youpi> well, both
+ < youpi> so e2fsck can show up too
+ < braunr> yes
+ < youpi> it's just a matter of ref counting
+ < youpi> apparently nfs doesn't account
+ < youpi> find . -printf "" doesn't show up in waitio
+ < braunr> good
+ < youpi> well, depends on the point of view
+ < youpi> as a user, you'd like to know whether your processes are stuck on
+ i/o (be it disk or net)
+ < braunr> this implies clearly defining what io is