1 |
On Tuesday 04 September 2007, Remy Blank wrote: |
2 |
> Alan McKinnon wrote: |
3 |
[snip] |
4 |
> > The only case I can think of that *requires* initramfs right now is |
5 |
> > booting off a raid device |
6 |
> |
7 |
> Strangely enough, I am currently booting from a software raid device, |
8 |
> so you don't need an initramfs for that either. |
9 |
|
10 |
You have software compiled in the kernel, not as a module the, right? |
11 |
|
12 |
I would imagine you wouldn't need an initrd for that |
13 |
|
14 |
> >> And from what I remember, you can't resize a mounted ext3 |
15 |
> >> partition, |
16 |
> > |
17 |
> > balls. ext2online and resize2fs have been resizing ext3 partitions |
18 |
> > for ages. You can extend a mounted partition with ease and in |
19 |
> > safety. |
20 |
> |
21 |
> Have you ever tried pulling the plug while a resize operation was in |
22 |
> progress? I guess I'll have to test this myself, as my data is |
23 |
> valuable enough to me that I won't just believe what I read. |
24 |
|
25 |
An enlarge operation tends to be quite safe in my experience. At worst |
26 |
you get some new inodes that might not be accounted for, something that |
27 |
fsck will handle easily. |
28 |
|
29 |
A reduce might be a different case altogether. BUT, it's not an |
30 |
especially different operation to a defrag on Windows, and I have yet |
31 |
to see a Windows admin debate whether he should defrag or not based on |
32 |
the possibility of losing power halfway through... |
33 |
|
34 |
I can't comment too much on problems with reducing ext2/3, but I do |
35 |
reduce reiser3.6 filesystems often, once when the battery died, and it |
36 |
wasn't a problem when powered up. There was no feedback in the logs to |
37 |
speak of, so I assumed that the journal did what it was designed to do. |
38 |
This was in the first stages of the operation - moving blocks to the |
39 |
start of the volume, and I honestly have never done this test in the |
40 |
later stages - when inodes are removed from the superblock |
41 |
|
42 |
[snip] |
43 |
|
44 |
> > What you can't do, and to my knowledge no regular fs can do, is to |
45 |
> > *reduce* a mounted partition |
46 |
> |
47 |
> But who would want to do that? I always need *more* space, not less |
48 |
> ;-) |
49 |
|
50 |
emerged openoffice lately? :-) |
51 |
|
52 |
It pretty much always fails if you have <5G in /var/tmp/portage. On a |
53 |
laptop, that's 8% of my total disk space just sitting there free |
54 |
waiting for the day I emerge openoffice again. Umounting /var to reduce |
55 |
it is such a huge pita that I made /var/tmp/portage a separate volume |
56 |
and now reduce it at will. |
57 |
|
58 |
But true enough, especially on server, you will enlarge volumes much |
59 |
more often the reduce them |
60 |
|
61 |
[snip] |
62 |
|
63 |
> Anything special if I put the LVM over a software raid? |
64 |
|
65 |
Not in my experience. The only difficulty I ever had was persuading |
66 |
RHEL4 to install / like that - anaconda either doesn't support it or |
67 |
the button to click to do it is hidden in one of the magic cupboards at |
68 |
Hogwarts. But that's not a problem because: |
69 |
|
70 |
1. this is gentoo |
71 |
2. anaconda is these days less brain dead than it used to be |
72 |
|
73 |
Performance wise, it does well. The LVM and mdamd layers do their work |
74 |
in a fraction of the time it takes to get the data on/off the disk |
75 |
platters. In fact, Linux software usually outperforms most of those |
76 |
stupid el-cheapo we-say-it's-hardware-raid-but-actually-isn't raid |
77 |
controllers in low end hardware |
78 |
|
79 |
alan |
80 |
|
81 |
|
82 |
-- |
83 |
Optimists say the glass is half full, |
84 |
Pessimists say the glass is half empty, |
85 |
Developers say wtf is the glass twice as big as it needs to be? |
86 |
|
87 |
Alan McKinnon |
88 |
alan at linuxholdings dot co dot za |
89 |
+27 82, double three seven, one nine three five |
90 |
-- |
91 |
gentoo-user@g.o mailing list |