Re: [exim] [OT] RAID

Marc Perkel wrote:
>
> Sander Smeenk wrote:
>> Quoting Nigel Metheringham (nigel.metheringham@???):
>>
>>
>>> I have to say I am getting less convinced by RAID 1 on systems.
>>>
>> RAID at all for me, actually. I've seen it crumble down to a grinding
>> halt several times (using software raid, mdadm, that is). While it was
>> designed so that a disk could fail without the service being
>> interrupted, in my experience a disk failing will still make the system
>> unusable. Either because the kernel goes haywire trying to adress the
>> non-working device or mdadm making decisions causing the entire set to
>> go offline.
>>
>> Especially mdadm setups where you combine RAID-0 and RAID-1 sets to
>> achieve RAID-10 tend to break when there's problems.
>>
>> Still, recovery is easier as the set can rebuild or data can be
>> retrieved from just one disk from a set...
>>
>>
>>
>
> I have an unusual system myself but if I were setting up something
> ordinary what I would do is run 2 drives with no raid. Then every hour
> or so run rsync to copy all your important data between the drives. You
> rick losing an hour's worth of email in the case of hard drive failure
> but you gain speed and simplicity.

You can do much better than that...

IF both drives are 'local', just use an unseen router or shadow
transport to duplicate as you go.

'near real time' and no need to do all the checking that rsync and
sputniks need to do to know *what* to copy, ergo lighter system load at
the end of a long day and shorter exposure window to diff / loss.

Still looking for an easier way to make Dovecot do the same w/r updating
IMAP folder changes. There is a GMIRROR split-RAID trick that serves,
but it is OS-specific and a bit of a PITA to set up and admin.


>
> Also - I recommend really big SATA II drives. Buy a 750G or 1T drive. On
> the lower numbered tracks these drives are a lot faster. Most people
> don't know this but the outer tracks are about 2,5 times as fast as the
> inner tracks. So an oversized drive will increase performance.
>

Focused on the wrong weak link, and for two reasons:


- Multiple smaller drives are far faster as more spindles also = more
*independent* head positioners, RAID or otherwise, and positioning time
is nearly always a bigger factor than r/w IO rate.

- Too many eggs in one basket.

Regards,

Bill Hacker