1 |
William Kenworthy wrote: |
2 |
> On Tue, 2012-08-07 at 23:18 -0500, Dale wrote: |
3 |
>> William Kenworthy wrote: |
4 |
>>> On Tue, 2012-08-07 at 21:19 -0500, Dale wrote: |
5 |
>>>> Paul Hartman wrote: |
6 |
>>>>> On Tue, Aug 7, 2012 at 5:26 PM, Dale <rdalek1967@×××××.com> wrote: |
7 |
> ... |
8 |
>>> Goggle have a well known document |
9 |
>>> (http://research.google.com/archive/disk_failures.pdf) where they |
10 |
>>> analysed hard drive failures for a very large number of drives ... the |
11 |
>>> basic upshot is that a very large portion of failures happen with no |
12 |
>>> pre-warning, so testing a drive like you are proposing not going to |
13 |
>>> prove a thing. |
14 |
>>> |
15 |
>>> They also found that smart (is quite dumb) and its tests were of little |
16 |
>>> use. |
17 |
>>> |
18 |
>>> And high temperatures and work loads were also not a reliable guide to |
19 |
>>> trends in failure rates, both of which which surprised me. |
20 |
>>> |
21 |
>>> Some of those bathtub curves that I was trained on when setting |
22 |
>>> maintenance schedules dont hold water here! |
23 |
>>> |
24 |
>>> This anaysis of the paper looks quite good if you want the lite view: |
25 |
>>> http://storagemojo.com/2007/02/19/googles-disk-failure-experience/ |
26 |
>>> |
27 |
>>> BillK |
28 |
>>> |
29 |
>>> |
30 |
>> Well, I am going by actual real experiences from other users of this |
31 |
>> model of drive. I don't know what google was testing but I would bet it |
32 |
>> is not the drive model I just bought. The users who bought this exact |
33 |
>> model drive report that most failures are either out of the box or |
34 |
>> within a few weeks to a month. I'm just going to try to increase my |
35 |
>> odds even if it is just a little bit. |
36 |
>> |
37 |
>> Smart may not always predict a failure but it is better than nothing at |
38 |
>> all. Would you rather have a tool that may predict a failure or no tool |
39 |
>> at all? Me, I'd rather have something that at least tries too. The one |
40 |
>> drive I had to go bad, Smart predicted it very well. It said I had |
41 |
>> about 24 hrs to get my stuff off. Sure enough, the next day, it |
42 |
>> wouldn't do anything but spin. Without Smart and its prediction, I'd |
43 |
>> have lost the data on the drive with no warning at all. |
44 |
>> |
45 |
>> A couple questions. What if while I am testing this drive, it dies? |
46 |
>> Does that prove that my testing benefited me then? |
47 |
>> |
48 |
>> Dale |
49 |
>> |
50 |
>> :-) :-) |
51 |
>> |
52 |
> Read the paper - its written by someone who buys drives in batches of |
53 |
> 100's+, not by a few guys posting on a forum somewhere who bought one |
54 |
> random drive, who probably didn't use anti-static techniques handling |
55 |
> the drive, and thumped it around in the boot of the car or got it via |
56 |
> the courier who was famous for delivering TV's by throwing them over the |
57 |
> fence. It is a bit of an eye opener - read it. |
58 |
> |
59 |
> My impression of models is that it is not really the model that has a |
60 |
> run of failures, but the batch so a different run of the same model will |
61 |
> have a different failure pattern. There are exceptions such as those |
62 |
> IBM 60G deathstar drives, but then again they fixed it and following |
63 |
> drives of the same model were fine. |
64 |
> |
65 |
> My own experience of smart is it tells you something, but what it seems |
66 |
> to say is not right (notice I am not saying it tells lies, but that the |
67 |
> data and interpretation don't make sense on the drives Ive had) |
68 |
> |
69 |
> Drive failure does seem to be a semi-random lottery, but I am seriously |
70 |
> doubtful that testing will do anything ... it has as much chance of |
71 |
> precipitating failure that wouldn't occur otherwise because you are |
72 |
> seriously hammering it, or weakening the drive so it will fail at some |
73 |
> random time, but perhaps weeks away rather than the years it otherwise |
74 |
> would, or nothing will happen except for wasted electrons. |
75 |
> |
76 |
> Then again, I am of the view that modern electronics is |
77 |
> designed/programmed to fail a few seconds past warranty expiry (why else |
78 |
> do most devices have timekeeping built in :) |
79 |
> |
80 |
> BillK |
81 |
> |
82 |
|
83 |
Actually, I read the paper a long time ago. May give it another look |
84 |
but I'm still going by what people have posted about this specific |
85 |
model. If they make them all the same, then testing to at least see if |
86 |
it is going to get past the initial stages is a good idea. I do think |
87 |
some failures were because of the BIOS and I stated that in my original |
88 |
post. Getting a DOA drive can happen but when there are mobos around |
89 |
that can't see large drives, then one has to consider it. The ones I |
90 |
worry about are the ones that worked for a few weeks or a month then |
91 |
died. They obviously don't have BIOS issues but some other problem. |
92 |
|
93 |
Still, all things considered, I'm going to test the drive. If it can |
94 |
pass the test then I will feel better about putting my data on it. As |
95 |
for the paper: |
96 |
|
97 |
root@fireball / # ls -al /home/dale/Desktop/disk_failures.pdf |
98 |
-rw-r--r-- 1 dale users 247492 May 21 17:58 |
99 |
/home/dale/Desktop/disk_failures.pdf |
100 |
root@fireball / # |
101 |
|
102 |
|
103 |
I read it back in May. It's still sitting on my desktop. I might also |
104 |
add, it is about 5 years old. Drives have changed since then. For one, |
105 |
they have gotten larger. We don't know what else may have changed either. |
106 |
|
107 |
Dale |
108 |
|
109 |
:-) :-) |
110 |
|
111 |
-- |
112 |
I am only responsible for what I said ... Not for what you understood or how you interpreted my words! |