mixing backup with app serving on same box

mixing backup with app serving on same box

[Louis-David Mitterrand]

Hi,

I recently ordered a shiny Dell R510 2U server with 12x3.5" drives.
This machine will primarily be a backup host for nightly incremental
rsync runs of a dozen other machines.

Since this is a well-spec'ed box with dual CPUs, generous bandwidth and
32G RAM I'm wondering if it could also do some medium web serving
(mod_perl + postgres).

For the main backup partition I plan in using a mdadm raid6 on 8x3T
disks and raid6 4x1T disks for the app side.

So my questions are:

- is it realistic to have mix rsync processes and web serving on the
  same host?

- am I naive in thinking I can get away with it by using distinct disks
  sets for each task?

- all 12 disks being served by the same H200 controller, will the real
  bottleneck be at the controller level?

- any other consideration I should take into account?

Thanks,

Re: mixing backup with app serving on same box

[David Brown]

On 24/08/2011 13:49, Louis-David Mitterrand wrote:
> Hi,
>
> I recently ordered a shiny Dell R510 2U server with 12x3.5" drives.
> This machine will primarily be a backup host for nightly incremental
> rsync runs of a dozen other machines.
>
> Since this is a well-spec'ed box with dual CPUs, generous bandwidth and
> 32G RAM I'm wondering if it could also do some medium web serving
> (mod_perl + postgres).
>
> For the main backup partition I plan in using a mdadm raid6 on 8x3T
> disks and raid6 4x1T disks for the app side.
>
> So my questions are:
>
> - is it realistic to have mix rsync processes and web serving on the
>    same host?
>
> - am I naive in thinking I can get away with it by using distinct disks
>    sets for each task?
>
> - all 12 disks being served by the same H200 controller, will the real
>    bottleneck be at the controller level?
>
> - any other consideration I should take into account?
>
> Thanks,


Backups are useful for several purposes.  The most common case for 
needing a backup is user error - someone deletes the wrong file, and 
needs a copy.  For that purpose, a backup close at hand is very useful.

The second important use case is for disaster recovery - imagine a fire 
or theft, or the OS going bananas and eating all your data, or a 
break-in on your webserver.  For those purposes, a backup on the main 
machine is useless.

Personally, I prefer two backup sets - one on-site (on the same machine, 
or a nearby machine) for the first use, and one off-site for 
emergencies.  So a backup scheme on this server is good for half your 
backup needs.

However, I'd make some more effort into separating the backup 
functionality and the webserver functionality.  I'd strongly recommend 
you look at openvz - it gives you lightweight virtualisation so that you 
can keep functions like this separate without the resource costs of a 
full virtualisation solution.

Re: mixing backup with app serving on same box

[Gordon Henderson]

On Wed, 24 Aug 2011, Louis-David Mitterrand wrote:

> Hi,
>
> I recently ordered a shiny Dell R510 2U server with 12x3.5" drives.
> This machine will primarily be a backup host for nightly incremental
> rsync runs of a dozen other machines.
>
> Since this is a well-spec'ed box with dual CPUs, generous bandwidth and
> 32G RAM I'm wondering if it could also do some medium web serving
> (mod_perl + postgres).

Medium?

OK, so this is probably not the best place to be asking this stuff, but a 
lot will depend on what you mean by medium, however I know of busy 
ecommerce sites that are served on much lessser hardware than you have.

> For the main backup partition I plan in using a mdadm raid6 on 8x3T
> disks and raid6 4x1T disks for the app side.
>
> So my questions are:
>
> - is it realistic to have mix rsync processes and web serving on the
>  same host?

If the backup is mainly at night and the serving side is mainly during the 
day, (but watch for an international audience) then why not?

> - am I naive in thinking I can get away with it by using distinct disks
>  sets for each task?

Seems like an OK idea to me..

> - all 12 disks being served by the same H200 controller, will the real
>  bottleneck be at the controller level?

Possibly..

> - any other consideration I should take into account?

The cost of providing separate servers...

However, FWIW, I do something similar with my own hosted LAMP servers - as 
well as an offsite backup of critical data, they each rsync to a mirror 
overnight and as most of my physical servers are now 'hosts' for virtual 
servers using LXC, then in-case of a total hardware failure, I can 
actually bring-up the virtual servers on new hardware more or less right 
away (minus the data delta since the backup of-course).

Make sure you use a big chunk-size on the RAID-6, and make sure the 
underlying filesystem can make use of it too, if possible. (The -Ostride 
options for ext3/4 for exmaple) do take time building it and benchmarking 
it for your chosen purpose before going live...

Gordon