[mdlug] New Server - Hardware Configuration

[Aaron Kulkis]

Robert Adkins II wrote:
>> Robert Adkins II wrote:
>>> The plan is to copy all of the data to the hot swap drives and... 
>>>
>>> Swap them out every day, every few days.
>>
>>
>> Bad idea.  Keep two set of drives permanently installed, set 
>> 2 fully mirroring set 1.
>> These can (and should) be hot-swappable, but none of these 
>> drives should be swapped unless the unit in question fails.
>>
> 
> 	The plan is to have two sets of drives permanently installed. There
> was never any other plan with regards to that. In my research on this, Linux
> and the majority of SATA RAID cards out there do not like to mix any RAID
> Level and Hot Swap on off the shelf SATA, I have played with that setup
> before and discovered with much consternation that it doesn't like to
> recognize a new drive added or removed and will potential crap out the
> kernel or just corrupt the drive. 
> 
>> a THIRD set of hardware addresses should be mapped to 
>> hot-swapable bays, and by turning mirroring on and off, a 
>> snapshot of the filesystem state can be obtained with no 
>> system downtime and only momentary application service downtime.
>>
> 
> 	While the shop will occassionally run 24 hours a day, the servers
> are only required for a typical range of 14 to 18 hours a day. That leaves
> me with between 6 to 10 hours to perform a first rsync and then a regular
> update of the changes should take considerably less time.
> 
>> [I used to work in the financial sector, where this stuff is 
>> important -- good backups are required by LAW, and this 
>> method works well, both reliably and efficiently.].
>>
>> Doing round-robin rotation of your disks is a good way to 
>> have all of your disks hit MTBF (mean-time-between-failure) 
>> in rapid succession -- which is NOT a good thing.  You want 
>> disk failures to be staggered in time, not cascading at you 
>> like an avalanche.
>>
> 
> 	I only plan on performing the round robin rotation on the back-up
> discs and if one of those happens to fail, it's really inexpensive to buy
> another these days. Really, I'm not concerned about anything beyond a
> totally hardware failure of the primary hard drives in the server. All of
> the file shares are ran via Samba and I have Samba configured to interpret
> any file delete action as a unix mv command with the resulting file being
> moved into a "Recycle Bin" which is only accessible to myself as R/W (via
> Samba) and RO to a handful of other users. 


Thanks for the clarified explanation.
That sounds satisfactory.