i'd really appreciate comments on a new disk setup proposal for my file server.
goals
-> 8 GB for OS data
-> 2 TB for user data
-> lower the probability of OS data loss
-> lower the probability of user data loss
bonus
1) less power consumption is better, ceteris paribus
2) greater performance is better, ceteris paribus
3) lower cost options are better, ceteris paribus
---
geom mirror 1: OS
(qty 3) lexar 8 GB 60 MB/s UDMA CF (
link)
(qty 3) SATA to CF adapters (
link)
geom mirror 2: user data
(qty 3) hitatchi ultrastar 2 TB 7200 RPM disks (
link)
---
my impression is that the failure rates and the power consumption of CF are both drastically lower than hard disks. i also perceive that CF is suitable for the OS of a file server.
is this true?
yes, i want three disks in the user data mirror. that's because if one fails (which is likely), i'll have time to send it in and receive a replacement while retaining redundancy. however, i doubt i need three CF cards, because i think their failure rates are very low.
is this true?
any other comments or suggestions are welcome. thanks!
total cost: $831
:o
Instead of having three of each type of storage connected at all time, you could have one spare for each. So if one of the running ones fail, you can connect a brand new one immediately.
Would a small solid state disk be faster and more resilient than flash cards?
I just 'verified' that wikipedia's ssd article has something about it:
For low-end applications, a USB flash drive may be obtainable for anywhere from $10 to $100 or so, depending on capacity, or a CompactFlash card may be paired with a CF-to-IDE or CF-to-SATA converter at a similar cost. Either of these requires that write-cycle endurance issues be managed, either by not storing frequently written files on the drive, or by using a flash file system. Standard CompactFlash cards usually have write speeds of 7 to 15 MB/s while the more expensive upmarket cards claim speeds of up to 40 MB/s.
Found the low end 8 GB Kingston SSD that perfoms as follows:
http://www.newegg.com/Product/Product.aspx?It … 6820139427
Sequential Access - Read: up to 90MB/s
Sequential Access - Write: up to 30MB/s
MTBF - 1,000,000 hours
Notice that there's even a time before failure estimation, I looked up a few CF cards without any such information. It's something to consider. I know your CF card listed here claims 60 MB/s but it's not both read and write, just the read speed from the looks of it.
I don't know about the difference between power consumption between CF and SSD though.
> Instead of having three of each type of storage connected at all time, you could have one spare for each. So if one of the running ones fail, you can connect a brand new one immediately.
yeah, but rebuild time is still a few hours, and that would require a reboot.
> Found the low end 8 GB Kingston SSD that perfoms as follows
cool! i've never seen such a small capacity, inexpensive SSD.
> Notice that there's even a time before failure estimation, I looked up a few CF cards without any such information. It's something to consider.
MTBF probably isn't listed for flash cards because flash cards aren't expected to be used in a continuous fashion.
as far as i can tell, CF and SSD will have identical write cycle endurance, given they are both the same type of flash memory (e.g., SLC NAND).
from what i've read, SSD (like the one you posted) is just a controller and a flash chip (the same type of NAND memory as in the CF).
> I know your CF card listed here claims 60 MB/s but it's not both read and write, just the read speed from the looks of it.
sandisk says it's 60 MB/s for both read and write.
---
would you guys be willing to run a server's OS from SSD? under what conditions?
> would you guys be willing to run a server's OS from SSD? under what conditions?
I used to run my router with just a CF card. I did nothing special because of the SSD.
i'll just stick with the 250 GB disks for the OS. they cost $40 each.
i'll get the hitatchi enterprise 2 TB disks for user data.
> Instead of having three of each type of storage connected at all time, you could have one spare for each. So if one of the running ones fail, you can connect a brand new one immediately.
what's the advantage to this? just decreasing wear on the spare?
options (time as degraded):
* upon disk failure, return the disk and wait for a replacement with a degraded array. (2 weeks)
* upon disk failure, immediately order a disk to be delivered overnight. (24 hours)
* keep a spare disk on hand for each mirror. (6 hours)
* keep a hot spare online for each mirror. (2 hours)
* keep a third disk on each mirror. (no time)
> what's the advantage to this? just decreasing wear on the spare?
Pretty much.