On 03/03/2014 06:55 PM, Sage Weil wrote: > On Mon, 3 Mar 2014, Andrey Korolyov wrote: >> Hello, >> >> On Mon, Mar 3, 2014 at 2:41 PM, Pieter Koorts <pieter@xxxxxxxxxxxx> wrote: >>> Hi >>> >>> Does the disk encryption have a major impact on performance for a busy(ish) >>> cluster? >>> >>> What are the thoughts of having the encryption enabled for all disks by >>> default? >> >> Encryption means stricter requirements to handle a power failure, because >> container contents may be lost entirely as easy as regular filesystem may >> get corrupt on same event. Therefore enforced sync policy along with the >> additional CPU resources consumption and (very possibly) battery for disk >> controller requirement describes all the difference. > > Hi Andrey, > > You're talking about dm-crypt, right? How does that affect safety? I > assumed that it passes IOs directly up and down the stack without > reordering or buffering. > Hi, Yes, my point is primarily about dm-crypt containers, HDD cache matters there. Or, in most terrible case, for any FS with default mount options when volume laid on it as a regular file (though nobody will do this for Ceph because of related performance impact). My experience represents the actual results for 'benchmarking' this three years ago but I think not much changed there. May be it will be interesting to collect fault statistics over different block sizes for crypto containers and raw storage with default device cache settings. _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
 |