1 |
Grant schrieb: |
2 |
>>> ... What if I bought a low-price/low-capacity SSD drive for each |
3 |
>>> of these systems, installed the system essentials on them, and used my |
4 |
>>> existing high-capacity HD drives for data storage? Would each system |
5 |
>>> keep running if the HDs died? If so, I think that would offer as good |
6 |
>>> or better system reliability than RAID1. What do you think? |
7 |
>> You don't need to buy SSD "drives" - instead you could use CF cards and a |
8 |
>> cheap adaptor. These are commensurate in capacity & cost with USB flash |
9 |
>> drives (4gig, maybe 16gig?), but CF cards "talk EIDE" and you can get cheap |
10 |
>> pin-convertors allowing you to connect them to EIDE cables and treat them |
11 |
>> like a hard-drive. |
12 |
> |
13 |
> Aren't CF cards much slower than SSD drives and HD drives? |
14 |
> |
15 |
|
16 |
Yep, especially the cheap ones which do not support DMA, just PIO. But |
17 |
this is not necessarily a problem: After starting all services etc. |
18 |
there will be very few reads on stuff like /etc and /usr. Just make sure |
19 |
to put all directories to which you write (parts of /var like /var/log |
20 |
and the several tmp directories) on an HDD, NFS or tmpfs. Of course, |
21 |
this all depends on your usage patterns and how much RAM you have. |
22 |
|
23 |
If you really need to write to the CFDisk, make sure to buy one with DMA |
24 |
support (and no, the label "super fast" which is regularly found on |
25 |
these things does not necessarily mean that it supports DMA). |
26 |
|
27 |
One drawback of this configuration: You can never use swap - never! |
28 |
Neither on the HDD (there is a high chance that the system would crash |
29 |
when the HDD fails) nor on the (cheap) SSD/flash drive (the drive would |
30 |
wear down, removing any advantage you tried to gain). |
31 |
|
32 |
>> I know of these used in Asterisk based PABX systems & PoS tills with the |
33 |
>> expectation that they're more reliable than disks, and have read statements |
34 |
>> by people deploying quantities of such machines that they've never had a |
35 |
>> failure in years of use. |
36 |
> |
37 |
> I like the sound of that. |
38 |
|
39 |
Where I work, we have a System-on-a-Chip (SoC) NAS. Albeit being the |
40 |
second most powerful machine we have in our server room (quad core CPU, |
41 |
lots of RAM, three redundant power supplies and a good dozen HDDs), the |
42 |
OSS itself resides on a removable card not bigger than my thumb. |
43 |
|
44 |
> |
45 |
>> I don't know how that really compares to RAID 1 - if you use hardware RAID |
46 |
>> (and you can get hardware SATA controllers for £50 these days) then you can |
47 |
>> assign a hot-spare, and hot-swap a replacement drive with zero downtime. |
48 |
>> With hardware RAID you can still boot if one of the drives fails, but you do |
49 |
>> add the controller as a potential point-of-failure. |
50 |
> |
51 |
> Would the system keeping running if I used a CF or SSD for the system |
52 |
> install and the HD drive died? |
53 |
> |
54 |
> - Grant |
55 |
> |