Re: Edu mainserver - raid
- To: Giorgio Pioda <email@example.com>
- Cc: firstname.lastname@example.org
- Subject: Re: Edu mainserver - raid
- From: Steven Chamberlain <email@example.com>
- Date: Mon, 10 Jun 2013 22:50:53 +0100
- Message-id: <51B64A3D.firstname.lastname@example.org>
- In-reply-to: <20130531101650.GB5288@macchianera.pioderia.lan>
- References: <20130531095946.GA5288@macchianera.pioderia.lan> <20130531100607.GJ9549@diskless.uio.no> <20130531101650.GB5288@macchianera.pioderia.lan>
On 31/05/13 11:16, Giorgio Pioda wrote:
> I have only 4 disk slots, unfortunately... And for this setup a
> minimum of 5 is fixed, I think.
> I don't remember where and why, but time ago I've read that raid5 is not
> the best choice for NFS mounts. [...]
That's probably because NFS does synchronous writes, and serving
multiple users could mean lots of random reads, so the array could
become limited by IOPS.
The type of disk will determine that most (SSD > 10kRPM SAS > 5.4kRPM
SATA). And if the RAID controller has a write cache (BBWC) that may
help, but I've found at least the HP Smart Array P410 to be quite poor.
Total IOPS are affected by RAID configuration something like:
* 4 disks in RAID 5 => get 1x IOPS, N-1 = 3x capacity
* RAID 1+0 (mirrored pairs) => get 2x IOPS, but less space (N/2 = 2x)
For big files the throughput should be about the same either way.
The best idea though is definitely to try it. Boot a live system and
test with iozone or bonnie++ with several GiB of data. But predicting
IOPS/throughput a deployed system will need is hard...