On 2010-09-20, at 10:11 AM, Cuong Hoang Bui wrote:
> Hi,
>
> I'm wondering about the storage backend for chunk servers. I have 2 options
> 1. Using RAID for underly storage backend for chunk servers.
> 2. Not use RAID, Just separate disks attached directly to OS.
>
> With RAID (maybe RAID 6) option, the solution is very good, but take
> money for RAID card.
>
> With no RAID, I wonder about tolerance (disk failure). If I have 8 disks
> attached directly to OS, for example I mount as /mnt/sdb, /mnt/sdc,..,
> .mnt/sdh. If one of them failed, what happens to data, also operations
> that's performing on this disk. Please give me some advice.
>
I guess without a hardware RAID card, to ensure no data loss due to a single disk failure, you could run a separate chunk server process per disk on the given machine, and then set the goal for the number of copies of files in the MFS store to be something more than one. This is likely a messy and complex configuration, it would likely be better to have two machines with 4 disks in each, for example, or to use a logical volume manager, multiple device, or software raid, depending on what your OS supports. as that might have slightly better performance than the network chatter for multiple instances of chunk servers on the same box.
Travis
|
