Take the 2-minute tour ×
Server Fault is a question and answer site for system and network administrators. It's 100% free, no registration required.

I am working on a server build project for work and have inherited the following items that I need to figure out the best way to handle and configure.

The primary aims of this box will be to provide storage services and file sharing for a number of clients through NFS, possibly SAMBA if we cannot get the Windows clients to connect to NFS, and maybe function as a DNLA server for some TV's or media boxes that we have from the stored content on the server. We would also like to use part of this storage as possible space for Linux KVMs.

We would also like to utilize the following items within the Red Hat/Cent OS/Scientific Linux world:

  • LVMs for expandable or changeable space
  • LUKS for encryption of the data
  • And a RAID level of some sort - but we have found confusing information about this that is listed below and we are not sure if it is accurate
    • Linux's LVM supports RAID 0 and RAID 1
    • Linux's md supports RAID 0, RAID 1, RAID 4, RAID 5, RAID 6, and all nestings.
    • The I/O Crest card supports RAID 0, RAID 1, RAID 5, and RAID 10

The challenge and questions that we have include the following:

  • What RAID level would be the best for our applications that we intended to use?
  • Would it be better to have the I/O Crest card handle the RAID or to let Linux perform the RAID in software?
  • We initially were considering RAID 6 for fault tolerance but are not sure if that was an appropriate choice and if it would even work considering that many of the items listed do not support it?
  • Is there a better way to do this?
  • What are the trade-offs or considerations that need to be made in terms of speed and in the event of a recovery?
share|improve this question
    
You need to figure out how much usable space you need first. Then we could talk ;-) –  Belmin Fernandez Oct 10 '11 at 5:14
    
I would say as much usable space as we could get with considerations for some level of fault tolerance to exist. :) –  John Oct 10 '11 at 5:14

4 Answers 4

up vote 7 down vote accepted

•What RAID level would be the best for our applications that we intended to use?

You give not even half the information needed to decide it. I would say right now to go with Raid 6 given that I am not sure how you handle backup / restore.

•Would it be better to have the I/O Crest card handle the RAID or to let Linux perform the RAID in software?

Depends. Yes, only sensible answer. GIven the CREST card does not support Raid 6 it is likely a low end or outdated Raid controller.

If it is this one: http://techbucketblog.com/www/2010/09/23/io-crest-sata-raid-pcie-card/ - gratulations, you did not buy a raid controller but a card with a raid driver. Dump it, (as in dont use it) and go for MD. This thing has no processsor on the card anyway to offload riad calculations - nothing is lost.

•We initially were considering RAID 6 for fault tolerance but are not sure if that was an appropriate choice and if it would even work considering that many of the items listed do not support it?

Only one needs to support it. The card not supporting it soundsl bad to me - md does, so you are safe. You can use D.

•Is there a better way to do this?

Considered using something that supports ZFS? Would also handle the long term grow better.

What makes me cringe is:

SAMBA if we cannot get the Windows clients to connect to NFS,

I would not try to force windows clients on an inferior protocol without good reason - you buy tons of maintenance issues and likely loss of functonality. Samba is a good server and it works without any configuration on the clietns, this is a hugh time saver. Also long term you are more likely to support network level deduplication this way when Windows Server 8 comes out - something NFS totally fails at to my knowledge.

share|improve this answer

Type : Your primary concern

RAID-10: i/O throughput, rather than efficient capacity use

RAID-5: i/O isn't a matter, nice I/o would suffice, efficient capacity use. Risky when having lots of disks due to severe performance impact on a disk' failure and high possibility of data loss.

RAID-6: data integrity and safety (very poor i/O)

Also, you can combine some levels together, for e. g., three RAID-5 (or two RAID-6) can be stripe-grouped with LVM-2 or Linux SoftRAID + LVM-2. You just need to realize how many disks you're going to spend for redundancy (meta-data) and data-store and see what do competing levels offer.

share|improve this answer
    
Forgotten to say: keep in mind chances of data loss as well. –  poige Oct 10 '11 at 5:45
    
and also bear in mind rebuild times –  Sirex Oct 10 '11 at 9:18
    
-1: Raid 10 for I/O throughput and I/O not relevant to RAID 5??? –  symcbean Oct 10 '11 at 10:48
2  
i/O stresses on Output, whilst I/o on Input. Details are important, -1er. –  poige Oct 10 '11 at 12:08

Hardware RAID 10, if it were my choice, because i

In most cases RAID 10 provides better throughput and latency than all other RAID levels except RAID 0

and nested RAIDs in common is excellent in terms of redundancy and data-loss recovery capability

share|improve this answer

"for a number of clients" - what number of clients? What patterns of file access? How much space do you need? What are your constraints (availability, performance, budget, physical number of disks)?

share|improve this answer

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.