[zfs-discuss] Re: Write Caching Without BBU
gordan.bobic at gmail.com
Sun Aug 18 14:24:08 EDT 2013
On 08/18/2013 07:00 PM, Niels de Carpentier wrote:
>> If the controller correctly handles flush commands (i.e. ignore them
>> when a BBU is present, but honor them otherwise), then it doesn't matter
>> if it reorders writes or not - the result will always be correct,
>> up-to-date and consistent.
> That's a very big if. I don't see any indication or reason that the write
> back cache behaves different with or without a BBU. They just state that
> there are no guarantees if you use WB without a BBU.
> The just say that if you want reliable data use a BBU or WT cache.
> The need for specific WB cache behaviour without a BBU is very rare and
> probably limited to a few hobbyists. Seems reasonable to me, BBU's are not
> that expensive if you care about your data.
Just to quantify that statement:
8-port LSI SAS HBA: ~ £140
BB cache module (a DIMM with a battery port): ~ £300
They are not inexpensive.
Regardless, I think the most important point this thread produced is
that ZFS effectively deprecates caching controllers.
To unsubscribe from this group and stop receiving emails from it, send an email to zfs-discuss+unsubscribe at zfsonlinux.org.
More information about the zfs-discuss