On Mon, Jun 11, 2012 at 8:05 PM, Nick Holland <n...@holland-consulting.net>wrote:
> On 06/11/12 19:25, Jens A. Griepentrog wrote: > > Dear Mailing Listeners, > > > > Let me know, please, whether it makes sense to modify disk geometry > > for solid state disks? > > no > > > Which meaning have the default values of cylinders, > > heads, and sectors for these devices? > > roughly the exact same thing it has meant for IDE, SATA, and SCSI disks > since..well...about 20 years or so...not a thing. > > All modern drives, and really anything made in probably the last 20 > years (i.e., anything worth putting on an OpenBSD machine) use > translation...the "geometry" and "reality" are unrelated in any > recognizable way. > Like many such generalizations, ignoring the details can cause catastrophic failures. If your systems are virtualized, particularly virtualized on NetApps which use 4096 byte block drives on the back end, that "translation layer" can be overwhelmed. In particular, the use of the classic "msdos compatibility" and the 63 blocks of 512 bytes typically assigned for MBR and parttition can have a disastrous impact, which is tied to an old, old standard for boot loaders and partition information. Whether or not OpenBSD uses such an alignment structure, ignoring it by aying "oh, we just translate" and "we've ignored that for decads" can cause catastrophic slowdowns of the NetApp when the buffer on the NetApp used for translation overflows and the NetApp goes into single CPU mode. The white paper on the problem is here: http://www.citrix.com/site/resources/dynamic/partnerDocs/BestPracticesforFileSystemAlignmentinVirtualEnvironments.pdf The burden is not as catastrophic on a local drive with a sane local controller, but that re-alignment is still an unnecessary performance hit that should be avoided in any high performance system.
