[nmglug] production server question follow-up

[Brendan]

On 10/09/2009 09:32 AM, Andres Paglayan wrote:
> now I just ordered the hardware, since the chassis supports 8 disks, and
> the mobo 6,
> so I purchased 8 disks just to have two identical spares sitting there
> unplugged.

Two port SATA controllers are $20.  Pick one up and get those spares active.

> This means that I'll have 6 disks running on the machine,
> the server does constant little writes, and probably 50 times more
> little reads.
> mostly from the DB

Since you have more disks I'll renew my suggestion that you use one set 
for DB and another for other IO modes.

> Since I don't see myself changing or adding disks during the life of the
> server,
> I will not do LVM on top.

That's too bad.  LVM snapshots are freaking awesome.

> three mirrored groups of two, and then stripping, raid 10 (3/6
> available, 2 or 3 can fail)

Uh, parser error.  Are you saying raid10(raid1,raid1,raid1)?  Or are you 
saying raid0(raid1,raid1,raid1)?  The former means you can lose between 
2 and 4 drives and keep going.  The latter means, in theory, 2 drives 
dying could lose data.  Remember that you can do RAID1 with more than 2 
disks.  A 3 disk mirror means you have 3 copies of each drive.  It's a 
little brute force, but it works.

> raid 6 (4/6 available, 2 can fail)

Yep.

> My sensation for what I read is that 10 will be way faster,

Writes will often be faster.  Random IO might be faster.  Throughput 
will be slower.

> and seems to be more in compliance with what DB vendors suggest,

Remember that the more drives you have in a system, the sooner one of 
them will fail.  Get those 2 spares hot.  And if at all possible, do 
some load testing with the different drive layouts.  Good luck,

-Brendan