summaryrefslogtreecommitdiff
path: root/open_issues/tty_activitiy_vs_disk_io.mdwn
blob: 26382d56d737dbe1728da0c42ecba237abe6ea73 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
[[!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]]."]]"""]]

[[!tag open_issue_hurd]]

IRC, freenode, #hurd, 2011-07-25

    < youpi> Mmm, typing something on the mach console triggers a write on the
      disk
    < youpi> because the /dev/console node gets updated
    < youpi> I don't really see why
    < youpi> (yes, just typing at the bash prompt, not even running something)
    < youpi> typing during the sleep command (i.e. mere tty echo) doesn't
      trigger it, however
    < youpi> running bash's echo does trigger it
    < braunr> during sleep, the glibc stream functions handle I/O, while with
      bash, its readline takes care of it, right ?
    < youpi>  /bin/echo too
    < youpi> during sleep it's the tty process which handles I/O
    < braunr> the write may be due to a write time update on the inode
    < braunr> modification* time
    < youpi> probably yes, but how so?
    < youpi> ext2fs is only supposed to pass the thing to the console
      translator
    < braunr> not sure
    < youpi> actually, ext2fs even isn't supposed to come into play when it's
      about typing at the bash prompt
    < youpi> once it's opened, isn't the port for /dev/console supposed to be
      directly to the translator there?
    < braunr> i think so
    < youpi> (s/tty/term/ in what I said)
    < braunr> well, it's certain
    < youpi> so I don't see how ext2fs can be triggered to write an atime or
      mtime
    < braunr> what does rpctrace say ?
    < youpi> io_read_request and io_write_request
    < youpi> braunr: it doesn't happen at the login prompt
    < youpi> interestingly, atime is always 3-4 secs earlier than ctime & mtime
    < youpi> doesn't happen with dash
    < braunr> we should implement relatime and experiment with it
    < braunr> it shouldn't be hard
    < youpi> well, there's noatime already
    < youpi> but my point is that this update shouldn't happen
    < youpi> and I believe it's the source of the i_file_acl e2fsck warning
    < braunr> i wasn't saying that concerning this problem, it was just a
      separate idea (noatime is more problematic than relatime)
    < braunr> and i agree, it shouldn't happen :)
    < youpi> ok, it's set_node_times which gets called

IRC, freenode, #hurd, 2011-07-27

    < antrik> BTW, I'm not sure it's still relevant; but the reason accessing
      translators such as the console modifies the underlying node is that most
      stat information is generally passed through
    < antrik> (in some cases it might be unintentional though, simply using the
      default implementation from trivfs carelessly...)
    < youpi> I know
    < youpi> I've seen that in the code
    < antrik> OK
    < youpi> it is still relevant: I still find it useless to write it on the
      disk
    < youpi> though w uses it to show idle time over reboot
    < braunr> is it useful to keep the information across reboots ?
    < youpi> for some value of "useful" for w
    < braunr> i wonder what would break if this was entierly kept in memory
    < youpi> nothing, probably
    < youpi> note that it doesn't overload ext2fs so much, it just adds a write
      every ~5s
    < youpi> (at worse, i.e. when keeping showing text, for instance)
    < braunr> indeed, the behaviour seems the same on linux
    < antrik> ah... that explains why the disk doesn't spin down while IRC is
      active... always wondered about that :-)
    < youpi> that's not very power-saving, yes
    < youpi> well, we might want to put /dev on ram someday