1 |
On Thu, Nov 27, 2008 at 5:26 AM, Stroller |
2 |
<stroller@××××××××××××××××××.uk> wrote: |
3 |
> |
4 |
> On 27 Nov 2008, at 02:08, Volker Armin Hemmann wrote: |
5 |
>> |
6 |
>> On Donnerstag 27 November 2008, Grant wrote: |
7 |
>>> |
8 |
>>> I'm considering buying a solid-state drive to improve I/O performance |
9 |
>>> and even reduce noise. Has anyone tried this? I was considering |
10 |
>>> getting the lowest capacity I can find and putting most of the system |
11 |
>>> on it. There is a roundup on tomshardware.com and it sounds like some |
12 |
>>> are very much better than others. SLC sounds vastly superior compared |
13 |
>>> to MLC, but also much more expensive. |
14 |
>>> |
15 |
>> |
16 |
>> http://valhenson.livejournal.com/25228.html |
17 |
>> |
18 |
>> I would rethink that after reading that post. |
19 |
> |
20 |
> From TFA: |
21 |
> |
22 |
> Postscript: Yes, this analysis is based on anecdotal evidence and |
23 |
> personal experience, but I can't afford the time to do real research |
24 |
> unless someone pays me to. If you know someone who will, send me |
25 |
> email! |
26 |
> |
27 |
> I've read a number of other reports, also based on anecdotal evidence and |
28 |
> personal experience, from a number of people who have very happily been |
29 |
> using flash as root volumes for years. Their opinions disagree with TFA. |
30 |
> |
31 |
> Typically the reports I've read have been from people using CFcards - 4gig |
32 |
> is now unbelievably cheap, and CFcards talk EIDE with only a small, cheap |
33 |
> physical adaptor - on MythTV frontends & low-overhead servers. CFcards look |
34 |
> ideal for these purposes because they're quiet - you want to minimise noise |
35 |
> when playing back video in the living room, for instance. |
36 |
> |
37 |
> I think the last anecdote I read on this subject was written by Trubox |
38 |
> (Truebox?) on the Openmoko-community list a month or two ago. They sell |
39 |
> Aserisk systems to small business (in my area, as it happens) and I would |
40 |
> imagine that typically the system sits in the corner of an office and is |
41 |
> untouched for years at a time. I would imagine that have plenty of installed |
42 |
> systems throughout the UK (otherwise they'd be going hungry). They report a |
43 |
> very low failure rate, as did someone else on the MythTV-users list who also |
44 |
> bases a commercial offering on flash-based hardware. |
45 |
> |
46 |
> Whilst I would probably, myself, install a second flash drive myself & |
47 |
> back-up (to a stage 4?) periodically, and avoid disk-writes when logging, I |
48 |
> get the strong impression that there's little to be scared of using flash |
49 |
> memory. |
50 |
> |
51 |
> Everything I read that says flash - and particularly its wear-levelling - is |
52 |
> unsuitable for this purpose makes sense to me, but it doesn't jibe with the |
53 |
> real-world experiences of those who ARE using flash VERY happily. |
54 |
> |
55 |
> I've yet to see empirical evidence on the longevity of flash for this |
56 |
> purpose, but I'd advise anyone considering it - anyone thinking flash |
57 |
> unsuitable - to search the mailing lists I've mentioned. The Trubox post |
58 |
> should be easy to find, and the subject comes up on MythTV-users every few |
59 |
> months. |
60 |
> |
61 |
> Stroller. |
62 |
|
63 |
The catch, though, is that I'd guess commercial offerings of MythTV |
64 |
boxes like that would be updated infrequently and that the actual |
65 |
recording storage, and likely logs and other frequent write files, is |
66 |
done on a normal disk. It doesn't seem logical for the average Gentoo |
67 |
user that follows the 'update often' mentality, and someone looking to |
68 |
milk the very top in the way of speed out of their system through disk |
69 |
throughput is very likely a 'ricer' in other respects. When you start |
70 |
using it for frequent writes (like your average system with everyday |
71 |
use and frequent upgrades) you start getting a little closer to the |
72 |
line on write cycles for small-sized MLC... but SLC is going to, by my |
73 |
guess, outlast its speed benefits by far (much like the old 800MB |
74 |
harddrives I have around that have far outlasted their size benefits |
75 |
from their day). Looking at SLC from a $$/GB standpoint you'll find |
76 |
they're horrendous, but from a $$/performance... at the very least |
77 |
Intel's X25-E starts to look a lot more reasonable for its cost (it's |
78 |
easily enterprise grade and mops the floor with just about anything |
79 |
else that holds data through a reboot)... which is around $760 for the |
80 |
32MB model. |
81 |
|
82 |
On the topic of using CF for the job... just looking at |
83 |
http://www.robgalbraith.com/bins/multi_page.asp?cid=6007, which |
84 |
doesn't take CF -> SATA adaptors into account, the highest read speed |
85 |
across the board is about 50MB/s, which is half of what the |
86 |
Velociraptor averages (and 1/5 of its burst read). |
87 |
|
88 |
-- |
89 |
Poison [BLX] |
90 |
Joshua M. Murphy |