[TriLUG] SSD-based NFS servers for production - ready for primetime?

[Ron Kelley]

Thanks for the reply Jeff!

We have two NFS servers with 8x 1TB RAID drives in RAID 5 on the Areca controllers.  As I mentioned before, they have been very reliable serving NFS clients over the past few years.  They were put in place mainly as a data repository for SVN, home directories, server backups, etc (typical slow-speed data requirements).   Recently, however, we started using them for everything including VM datastores, and they have struggled to keep up with the demand.  As usual, we tend to over-use technology in the infrastructure since it works so well (just add a **few** more VMs, etc).

As for the software RAID questions, I have used RAID-5 on Linux before with great success.  While RAID10 does out-shine RAID5 in performance, even software RAID5 can push much more traffic than our GigE interfaces can handle.  Plus, in a 6-disk array, I only burn the capacity of 1 drive instead of 3.

As for RAM, I have always been under the impression Linux NFS servers with hardware RAID controllers don't need much RAM to work efficiently since most/all the disk I/O is managed by the controller.  Perhaps this is not true?  Right now, I have 4G of RAM in both NFS servers and could easily add more.  But, will I gain any client read/write performance by doing so?

Thanks for the great questions,

-----------------------------
Ron Kelley
rkelleyrtp at gmail.com

On Jan 15, 2012, at 2:26 PM, Jeff Schornick wrote:

> Have you considered rethinking your RAID layout?
> 
> I'm no storage expert, but red flags start to pop up when I read this
> combination of statements:
> 
>> has pushed the NFS servers to the point of I/O starvation.
>> As for the configuration, I will put them in a 6-drive RAID-5 array using Linux software RAID.
>> we need about 1.5T of usable NFS space to store VMDKs and other high-sensitive items.
> 
> What does your I/O profile look like?
> 
> Given the sensitivity of the data, I/O starvation, and relatively
> small storage requirement, I would strongly consider a RAID10 profile.
> It's more straightforward, more resilient, performs better in a
> write-heavy environment, and should be tunable to perform just a s
> well as RAID5 on reads.
> 
> There are fancier things out there, but when limiting the choices to
> software raid, you have to work pretty hard to steer me away from
> RAID10.
> 
> Coming from another direction, given the low price of memory these
> days, would your NFS servers benefit from spending the money on more
> memory?  Once again, depending on your I/O profile, more caching could
> be a cheap and effective upgrade.
> 
> Yet another area to consider... how many NFS servers are we talking,
> and how is the data distributed?
> 
>  - Jeff
> -- 
> This message was sent to: Ron Kelley <rkelleyrtp at gmail.com>
> To unsubscribe, send a blank message to trilug-leave at trilug.org from that address.
> TriLUG mailing list : http://www.trilug.org/mailman/listinfo/trilug
> Unsubscribe or edit options on the web	: http://www.trilug.org/mailman/options/trilug/rkelleyrtp%40gmail.com
> TriLUG FAQ          : http://www.trilug.org/wiki/Frequently_Asked_Questions