1 |
begin quote |
2 |
On Wed, 13 Oct 2004 00:31:34 -0700 |
3 |
"Robin H. Johnson" <robbat2@g.o> wrote: |
4 |
|
5 |
> The one thing that your (previously known) method does bring out is |
6 |
> that reducing the I/O required really helps. |
7 |
|
8 |
Agreed, preferrably we should be able to distribute a bin-ball of the |
9 |
files that we can rsync out, since if a file doesn't move inside it (as |
10 |
is the case with filesystems) they stay on the same place and gets good |
11 |
replication in rsync. |
12 |
|
13 |
|
14 |
> Pack it into a loopback reiserfs instead, way better performance. |
15 |
actually not. it kills performance to put a journalled filesystem |
16 |
in a loopback system onto a data-journalling filesystem. |
17 |
|
18 |
I care more about data and fragmentation than I care about performance, |
19 |
that was the sole reason I first spiralled it off like this. The tree is |
20 |
constrained. |
21 |
|
22 |
As for the reiser protectionist idea, *cough* I don't have good |
23 |
experiences with reiser, tailpacking, and performance. |
24 |
|
25 |
|
26 |
> For an even bigger boost, put the loop file into tmpfs or use some |
27 |
> other direct memory scheme. |
28 |
|
29 |
Kills reliability. And performance isn't everything. (could just as |
30 |
well increase write times on the hosting fs, increase it even more on |
31 |
the loopback, then simply "cat filesystem.img > /dev/null" before |
32 |
operations. I don't. ) |
33 |
|
34 |
|
35 |
Overall, I'm not after complete performance. I walk the middle road, |
36 |
between performance and reliability. |
37 |
|
38 |
|
39 |
//Spider |
40 |
|
41 |
|
42 |
-- |
43 |
begin .signature |
44 |
Tortured users / Laughing in pain |
45 |
See Microsoft KB Article Q265230 for more information. |
46 |
end |