[CentOS] settings up cheap a NAS / SAN server, is it possible?

nate centos at linuxpowered.net
Wed Jul 2 13:40:44 UTC 2008

Sorin Srbu wrote:

> The guy who initially asked, IIRC, wanted some 3-4TB storage. This can be
> accomplished easily with a regular mid/maxi-size tower and a handful of
> 1TB-SATA drives. Even the midsize oldish Compucase-case I have at home can
> fit
> four 3,5"-drives in the hd-cage and another four in the 5,25"-bays. Suppose
> you fill that case with 1TB-drives and you have 8TB available. My argument
> is
> that there is no explicit need for the hardware you mention, despite how
> sexy
> it sounds. 8-)

Unfortunately it's common for most people that don't have experience
with storage to confuse raw storage space with I/O performance(IOPS).
It's not uncommon for someone to need say 4TB of space but need 50 or
even 75 disks to use for that space because they need the IOPS. In my
experience the average I/O size seems to be sub 20kB, which means a
10k RPM disk can only sustain about 1.9MBytes/second (15kB I/O size),
before latency starts becoming a serious issue.

Of course if you have specialized applications you can probably
push the I/O size much higher, I'm talking for just generic off
the shelf type tasks, such as file serving, Oracle/MySQL databases,

I came in late to the conversation I did see the original posts
around wanting a cheap system and saw several good answers, I
myself have used Openfiler and it's pretty good(for iSCSI, it seems
flakey for NAS). Those types of systems really cannot be upgraded
online safely. The poster I responded to asked about how do you
upgrade, so I responded. The obvious answer for a "cheap" system
is to upgrade the live box and reboot. There aren't many ways to
upgrade such a system. Maybe if you have a cheap fiberchannel
storage array(ala Infortrend) you can have two NAS front ends
connected to the same back end storage, minimizing downtime
and risk you can upgrade on system and flip people over to it,
if it screws up you can flip back fairly quickly(though still
not an online operation).

I recall a kernel update to Openfiler that caused random kernel
panics. Fortunately I was able to go back to the earlier kernel.
(this was running their "stable" distribution)


More information about the CentOS mailing list