[Gluster-users] Gluster infrastructure question

[Joe Julian]

Nux! <nux at li.nux.ro> wrote:
>On 09.12.2013 13:18, Heiko Krämer wrote:
>> 1)
>> I'm asking me, if I can delete the raid10 on each server and create
>> for each HDD a separate brick.
>> In this case have a volume 80 Bricks so 4 Server x 20 HDD's. Is there
>> any experience about the write throughput in a production system with
>> many of bricks like in this case? In addition i'll get double of HDD
>> capacity.
>
>I have found problems with bricks to be disruptive whereas replacing a 
>RAID member is quite trivial. I would recommend against dropping RAID.
>

Brick disruption has been addressed in 3.4.

>> 3)
>> Failover of a HDD is for a raid controller with HotSpare HDD not a
>big
>> deal. Glusterfs will rebuild automatically if a brick fails and there
>> are no data present, this action will perform a lot of network
>traffic
>> between the mirror bricks but it will handle it equal as the raid
>> controller right ?
>
>Gluster will not "rebuild automatically" a brick, you will need to 
>manually add/remove it.
Not exactly, but you will have to manually add an attribute and "heal...full" to re-mirror the replacement.

>Additionally, if a brick goes bad gluster won't do anything about it, 
>the affected volumes will just slow down or stop working at all.
>

Again, addressed in 3.4.

>Again, my advice is KEEP THE RAID and set up good monitoring of drives.
>

I'm not arguing for or against RAID. It's another tool in our tool box. I, personally, use JBOD. Our use case has a lot of different files being used by different clients. JBOD maximizes our use of cache.