1 |
Jason Booth <jbooth@××××××××××××××××.net> posted |
2 |
200610221115.22864.jbooth@××××××××××××××××.net, excerpted below, on Sun, |
3 |
22 Oct 2006 11:15:22 -0600: |
4 |
|
5 |
> On Sunday 22 October 2006 06:16, Richard Freeman wrote: |
6 |
>> I'd just make SWAPDEVICE and LOOPDEV command-line parameters and then |
7 |
>> call the script 4 times. |
8 |
> or drop a for loop into it... |
9 |
> |
10 |
> I don't know much about raid, but if it's treated in /dev as a single device, |
11 |
> you may just be able to replace it and go. |
12 |
|
13 |
The neat thing about swap is that the kernel stripes it on its own -- no |
14 |
raid drivers needed. You just mount the raid swap partitions and set all |
15 |
the ones you want striped to the same priority (I use pri=1), and the |
16 |
kernel will do the rest on its own. If they are all the same size, great, |
17 |
if not, it'll stripe them until the smallest one is gone then it'll stripe |
18 |
the remainder, again all automatically. |
19 |
|
20 |
(Striping means it writes a few bytes, maybe the standard half-kb |
21 |
block tho in the case of swap tho I'm not sure, to the first device, then |
22 |
the next block to the second, the third block to the third, etc. Because |
23 |
bus speed is far faster than physical disk write speed, with four disks by |
24 |
the time you've sent the data to the fourth one, the first is pretty much |
25 |
done actually writing it to disk and ready for more again, so the data is |
26 |
written out and read in at bus speed rather than at bus speed until the |
27 |
cache on the drive fills up or empties, then at drive speed. The caveat |
28 |
with pure striping, aka raid-0, is that while it's much faster, it's not |
29 |
redundant at all, the "r" in "raid" isn't! Thus if one disk goes out, you |
30 |
lost what amounts to everything, tho of course a good data recovery place |
31 |
can still recover say 3 out of every 4 blocks if it was a four-way stripe. |
32 |
However, this isn't a problem as long as you don't need five-nines uptime |
33 |
or the like -- if you want a bad drive to crash the system anyway, so you |
34 |
know about it and can recover the non-raid-0 non-swap data on the other |
35 |
drives due to the redundancy of the other raid formats.) |
36 |
|
37 |
> May be overly paranoid, but writing encrypted data multiple times could help |
38 |
> someone to guess what certain file is and make an attack on the encryption |
39 |
> easier. I use ext2 for my encrypted loops so there's no journal as well. |
40 |
> Although the power fails sometimes, and can be a pain to fsck, i haven't lost |
41 |
> anything yet. |
42 |
|
43 |
If one were using a non-striped raid, say raid-1 (mirrored), or raid-6 |
44 |
(striped minus two, which are parity, so a 4-way is 2-way striped plus two |
45 |
parity, raid-6 allows you to lose any two of the drives), the data would |
46 |
be redundant, but not in pure striped. A redundant raid form swap might |
47 |
be used where uptime is critical and hot-swap drives are used, so the |
48 |
system could continue running after a drive crashed, while it was |
49 |
hot-swapped out. However, that's the big costly operation way of doing |
50 |
things, not a hobbyist's way of doing things unless you are Mark |
51 |
Shuttleworth or something, and going down to replace the drive is expected |
52 |
here anyway, after which the swap could be reconfigured, so no big deal. |
53 |
|
54 |
-- |
55 |
Duncan - List replies preferred. No HTML msgs. |
56 |
"Every nonfree program has a lord, a master -- |
57 |
and if you use the program, he is your master." Richard Stallman |
58 |
|
59 |
-- |
60 |
gentoo-amd64@g.o mailing list |