[Info-vax] Databases versus RMS

[Dirk Munk]

Arne Vajhøj wrote:
> On 4/17/2012 3:40 PM, JF Mezei wrote:

>> The reason I am askig is to know whether the "industry" has accepted
>> those risks of hiding physical disks behind storage arrays that may not
>> write right away and behind database engines that not only delay writes
>> and cache stuff on their own, but can also get corrupt because of a
>> power failure.
>
> You can get RAID controllers with huge caches, but those comes
> with battery backup.
>
> So those should be pretty well protected against power
> failure.
>
> Arne
>

Any decent raid controller, even a small PCI controller in a x86 system, 
should only cache writes (write back) when there is a valid battery backup.

Caching writes on a raid controller, and specially small log file 
writes, can be very beneficial for the database performance. The normal 
random database writes are being distributed over many disks, but the 
the log files are written sequentially on just a few disks. By caching 
the many small writes and replacing them by just a few large writes, 
this bottleneck is prevented.