1 |
Thomas Rösner <Thomas.Roesner@××××××××××××××.de> posted |
2 |
45AC0F0A.5050408@××××××××××××××.de, excerpted below, on Tue, 16 Jan 2007 |
3 |
00:32:26 +0100: |
4 |
|
5 |
> Duncan wrote: |
6 |
>> As for disk access, the average guy with a single hard drive will |
7 |
>> certainly find that the bottleneck in an unlimited jobs scenario such as |
8 |
>> the above. I'm running a four disk SATA based RAID array, RAID-6 (so |
9 |
>> two-way-striped, with two-way recovery as well) on my main system, |
10 |
> |
11 |
> Uhm, if you'd use RAID-1-1, you wouldn't have to calculate those |
12 |
> checksums... |
13 |
|
14 |
True, but computing checksums is relatively trivial compared to the I/O |
15 |
overhead, and reading (which doesn't need checksums unless the |
16 |
RAID array is operating in degraded mode) would be half-speed relative to a |
17 |
4-disk RAID-6 (so two-way data striped). Given the trivial checksumming, |
18 |
Writing would be comparable to a three-disk RAID-1, and is slower than a |
19 |
two-disk RAID-1, as for every data stripe there's two parity stripes to |
20 |
update, but there's not a lot of data actually written anyway during |
21 |
emerges, since all the work is done on tmpfs and only the final write is |
22 |
to the disk. When compiling the kernel, I'm working directly on disk but |
23 |
that mount is a four-way striped RAID-0, so I get its speed and no |
24 |
checksumming overhead there. (/boot is RAID-1 so GRUB can read it, so |
25 |
when the kernel is actually installed, it's written to 4-way RAID-1, but |
26 |
that's only one writing the completed kernel, so no big deal there.) |
27 |
|
28 |
If I was running less than four disks, the choice would be a bit tougher, |
29 |
between a 3-disk RAID-5 with two-way striped read speed but |
30 |
two-way-mirrored (plus checksumming) write speed, while losing the |
31 |
ability to recover from a two-drive failure, a 3-disk RAID-1, keeping the |
32 |
two-drive recovery but at a severe cost in speed, or a 2-disk |
33 |
RAID-1/mirrored. In that instance I expect the 2-disk RAID-1/mirrored |
34 |
would win as it would be too difficult to justify either of the 3-disk |
35 |
options. |
36 |
|
37 |
> Btw, may I come to your place when UT2007 is out? ;-) |
38 |
|
39 |
As a gamer, you'd probably not appreciate my choice of video card -- an |
40 |
older but freedomware driver Radeon 9250. I won't do |
41 |
proprietaryware/slaveryware. |
42 |
|
43 |
Luckily I'm not much of a gamer, but the rest of the hardware specs are |
44 |
complimented by dual 400x300 mm (21" diag) monitors, normally running @ |
45 |
1600x1200 stacked for 1600x2400. I used to run 2048x1536 stacked for |
46 |
2048x3072, but while it worked, text was blurry as that was below the |
47 |
monitors' native pixel pitch (set for 1600x1200), and the 85 Hz refresh |
48 |
doesn't hurt (I was limited to 60 Hz with the higher resolution, |
49 |
acceptable with the light text on dark backgrounds I prefer, but not |
50 |
ideal, or even tolerable with dark text on light backgrouns) either, so I |
51 |
finally caved in and downgraded my resolution. |
52 |
|
53 |
-- |
54 |
Duncan - List replies preferred. No HTML msgs. |
55 |
"Every nonfree program has a lord, a master -- |
56 |
and if you use the program, he is your master." Richard Stallman |
57 |
|
58 |
-- |
59 |
gentoo-amd64@g.o mailing list |