On Sat, Dec 12, 2009 at 04:29:50PM +0000, Peter Nikolic wrote:
Le 12/12/2009 08:44, Bob S a écrit : Thin it is time to re- visit the old way of doing things /de/hda ect this /dev/disc/by ID ect is nothing but a mine field over which there is very
On Saturday 12 Dec 2009 08:25:26 jdd-gmane wrote: little control .
The approach of using /dev/disk/by-id makes things more reliable. The device name stays idependent of the order the controllers get initialized. Or think about adding a new disk to an existing system. Your old device names stay independet if a controller or new disks get added.
I know there were people saying they don't have enough devices with the old system well maybe they need to completely Re- think their computing needs/System either that or we have a system whereby you can choose which system you use according to your system requirements big box too many drives hung off it "/dev/disc/by-ID " smaller system 2 or 3 drives hung off it "/dev/hdx " or if you really must insist on this darn scssi naming convention then "/dev/sdx" ect personally i can see no need for this darn silly calling an PATA or IDE device a Scsi device it just adds yet another level of un needed complication and confusion
This approach makes it much easier for the maintainers of the disk sub system. You gain one comman way to access disks independent of the particular interface (PATA,SATA,SCSI) in use. See the documentation to libata. For example http://www.kernel.org/pub/linux/kernel/people/jgarzik/libata.pdf
SATA devices should maybe /dev/sat but sdx humm i know this has been beaten up before but i see it as a very valid concern here we are trying to spread Linux and gain more users yet here we also go making it more and more difficult for new users to grasp if there is logic in there somewhere i cant see it .
There is a logic in libata and users new to Linux will very unlikely think about or know if a device was named /dev/hda or /dev/sdc in the past. The current approach makes things easier in the majority of use cases. Till now I've only found one use case where the old approach is an advantage. That's when you're running virtual systems and you like to have the guest OS disk device name being independent of the disk system used by the host. A person with deeper knowledge of the kernel should correct me if I'm wrong. Lars -- Lars Müller [ˈlaː(r)z ˈmʏlɐ] Samba Team SUSE Linux, Maxfeldstraße 5, 90409 Nürnberg, Germany