CentOS - Aug 2013 - LVM RAID0 and SSD discards/TRIM

I'm trying to work out the kinks of a proprietary, old, and clunky 
application that runs on CentOS. One of its main problems is that it 
writes image sequences extremely non-linearly and in several passes, 
using many CPUs, so the sequences get very fragmented.

The obvious solution to this seems to be to use SSDs for its output, and 
some scripts that will pick up and copy our the sequences in proper 
order once it's done. I have two 512GB SSDs, and I've used LVM to set up
a RAID0 between them.

I've got that part running, but since I'm on CentOS 5.8 (which is what 
this application officially supports), I don't have a kernel with SSD 
discard support, and after a few days (I told you, this application is 
write intensive), things get very slow.

Using hdparm to secure erase the drives and recreating the LVM RAID0 
gets things back to speed again, but that's obviously not ideal.

So, from what I understand, if I can get this thing running on CentOS 
6.4, I'll get kernel discard support, and discard support in LVM when 
running a RAID0. I'm using ext4.

Is that correct? Will this solve my problem? I want to confirm that 
discard support works on a RAID0 of SSDs using LVM and ext4 before I 
start working on getting this legacy application to run on a newer CentOS.

-- 
Joakim Ziegler  -  Supervisor de postproducci?n  -  Terminal
joakim at terminalmx.com   -   044 55 2971 8514   -   5264 0864

On 19.08.2013 20:28, Joakim Ziegler wrote:
> I'm trying to work out the kinks of a proprietary, old, and clunky
> application that runs on CentOS. One of its main problems is that it
> writes image sequences extremely non-linearly and in several passes,
> using many CPUs, so the sequences get very fragmented.
>
> The obvious solution to this seems to be to use SSDs for its output, and
> some scripts that will pick up and copy our the sequences in proper
> order once it's done. I have two 512GB SSDs, and I've used LVM to
set up
> a RAID0 between them.
>
> I've got that part running, but since I'm on CentOS 5.8 (which is
what
> this application officially supports), I don't have a kernel with SSD
> discard support, and after a few days (I told you, this application is
> write intensive), things get very slow.
>
> Using hdparm to secure erase the drives and recreating the LVM RAID0
> gets things back to speed again, but that's obviously not ideal.
>
> So, from what I understand, if I can get this thing running on CentOS
> 6.4, I'll get kernel discard support, and discard support in LVM when
> running a RAID0. I'm using ext4.
>
> Is that correct? Will this solve my problem? I want to confirm that
> discard support works on a RAID0 of SSDs using LVM and ext4 before I
> start working on getting this legacy application to run on a newer CentOS.
>
What kind of SSD are you using? We use Intel 520's here and don't really
see these kind of slowdowns.

Regards,
   Dennis