1 |
On Sunday 31 July 2011 15:17:16 Joshua Murphy wrote: |
2 |
|
3 |
> There probably is a fair chunk of difference in maximum speed the disk |
4 |
> can work at on each end (I've even seen around a 20MB/s difference on |
5 |
> several 160GB drives I've dealt with), but outside of some older |
6 |
> drives that've been heavily abused in their lives, I'm not sure I've |
7 |
> seen a sata drive that I've used my usual drive test (MHDD on a |
8 |
> Hiren's bootable USB) on register below around 60MB/s on the slow end, |
9 |
> and USB2's *theoretical* limit is 480Mb/s (60MB/s) ... real-world |
10 |
> implementations rarely reach, let alone top, around 40MB/s, so disk |
11 |
> speed variation across the disk is an unlikely source of the slowdown. |
12 |
|
13 |
Sounds entirely reasonable, and I wasn't really trying to blame the slowness |
14 |
on that variation - just mentioning it in passing. |
15 |
|
16 |
> More likely, it's the fact that parted has to start from the end, and |
17 |
> work its way backwards, reading, writing, and verifying in separate |
18 |
> rotations of the disk with no benefit from the drive's ability to |
19 |
> stream a larger block into cache, since the whole process is backwards |
20 |
> compared to the streaming read most drives are optimized for. |
21 |
|
22 |
Perhaps I'm naive here, but I should have thought an intelligent disk |
23 |
copying algorithm would be able to account for that, at least in part. Maybe |
24 |
that's why it ran the speed tests at the beginning. |
25 |
|
26 |
> Of course, this is all off the cuff conjecture on my part, including my |
27 |
> assumptions about how parted approaches the whole task... mixed with a |
28 |
> bit of anecdotal evidence on my end... but, makes for amusing |
29 |
> conversation and contemplation, if nothing more substantial. |
30 |
|
31 |
Indeed. |
32 |
|
33 |
> I will point out that the newer advanced format WD 500GB blue's I've |
34 |
> worked recently with pulled a consistent 120-110MB/s speed from end to |
35 |
> end... when their older 320s usually peaked at around 85 or so. |
36 |
|
37 |
Well, I haven't run any proper tests, but watching gkrellm during an |
38 |
occasional large transfer I don't remember seeing more than half that lower |
39 |
figure. These are two Samsung Spinpoint F3 1TB disks in md-raid with LVM-2, |
40 |
and I haven't fiddled with any of their settings. |
41 |
|
42 |
-- |
43 |
Rgds |
44 |
Peter Linux Counter 5290, 1994-04-23 |