[wellylug] Computer System Vendor Wgtn

[Cliff Pratt]

Jethro Carr wrote:
> On Sat, 2007-04-21 at 10:35 +1200, Cliff Pratt wrote:
>>  > So you have to buy a spare or two, or just hope that it won't
>>> die at an inconvenient time. Software raid is much easier, replace the
>>> faulty parts, get the kernel drivers compiled for whatever changes
>>> you've made, and your system is back.
>>>
>>> Getting the raid running that quickly sure beats restoring from backups.
>>>
>> I don't believe that reloading a software RAID is quicker. Firstly, you 
>> have to install the Operating System on the first disk, then you have to 
>> partition and format the second disk (assuming RAID 1 for the moment), 
>> then you have to copy all the data to the second disk, then you have to 
>> build the RAID, which writes the superblocks on the disk. Then you have 
>> to reboot.
>>
>> Once you have a pristine software RAID, then and only then can you 
>> install stuff on it. And hope like hell it doesn't throw a superblock 
>> error five minutes after it is all installed.
>>
>> With a hardware RAID you just connect to the build server and load 
>> everything on, 15 - 20 minutes, tops. No superblock errors, failures are 
>>   notified to me if they do happen and you just pop a disk out and pop a 
>> new one in. Trying doing that with software RAID!
> 
> ? I don't know how you are making it such a complicated process.
> 
> 
> 1) Install OS - setup software raid partitions when running the
> partitioner.
> 
Because the above step 1) doesn't work reliably! (Using the RedHat and 
Debian installers, anyway). And if it fails you have to dd /dev/zero to 
the whole disk to clear out any stray superblocks, because if you don't 
the build *may* appear to work but will produce disk errors all over the 
place when you reboot!

I've learnt the hard way that the best and most reliable way to do it is 
my method above, and I still genuflect and cross myself when I walk past 
the servers.

Cheers,

Cliff