Re: Aix Mirroring Vs Informix Mirroring
>From: uunet!GAN03X!spitz (Richard Spitz) >Subject: Re: Aix Mirroring Vs Informix Mirroring >Date: Wed, 19 May 1993 09:08:57 GMT >X-Informix-List-Id: <news.3374> > >Jonathan Leffler (obelix.informix.com!johnl@emory.mathcs.emory.edu) wrote: >: In general, operating system level mirroring and hardware mirroring are >: more efficient than Informix mirroring, if only because they can schedule >: writes to both the primary and the mirror disk concurrently, whereas >: Informix has to write to the primary and then write to the mirror. > >: When you don't have lower level mirroring, use Informix mirroring, but use >: a lower level mirroring system if you have a choice. These rules apply to >: version 4 or 5 -- they may need to be reconsidered with version 6, though I >: doubt it. >This confuses me a little. We are using I-Online 4.1 with rootdbs mirrored >by Informix. Since Online handles raw partitions and does not use the >UNIX file system for accessing the data, how can operating system level >mirroring be used for Online? > >Maybe I'm having a terrible misconception here, please explain. The problem is that some raw devices are really partially cooked by the O/S. A raw device is something which shows up in /dev as a character special file. On my machine, I have the following raw disk devices: crw-r----- 1 root 17, 24 Nov 4 1992 /dev/rsd3a crw-r----- 1 root 17, 25 Nov 4 1992 /dev/rsd3b crw-r----- 1 root 17, 26 Nov 4 1992 /dev/rsd3c crw-rw---- 1 informix 17, 27 May 17 14:53 /dev/rsd3d crw-rw---- 1 informix 17, 28 Nov 4 1992 /dev/rsd3e crw-rw---- 1 informix 17, 29 Nov 4 1992 /dev/rsd3f crw-r----- 1 root 17, 30 Nov 4 1992 /dev/rsd3g crw-r----- 1 root 17, 31 Nov 4 1992 /dev/rsd3h Three of the partitions are used by OnLine systems, the others are mounted file system or unused. These disks are conventional raw devices, but the devices represent entry points in the kernel which know that when I write to byte address 102400 on /dev/rsd3d, that translates to a particular track, head and sector on the single physical disk represented by the set of devices above. So, unless I used /dev/rsd3c (which does represent the whole disk), I go through a mapping layer in the kernel which translates my write requests into the correct location on the disk. On machines with logical volume managers and the like extend the mapping work done by the kernel. I could see similar entries to the above, but an extra layer of software would sit between the OnLine system and the actual disks, and would organise the layout of the disks and mirroring, and so on. As far as OnLine is concerned, it behaves the same as an ordinary raw disk. However, the O/S knows what it has to do to handle the job properly. This is used, for example, in HP-UX Version 9, where the LVM can make 10 GB of RAID disk appear like a bunch of five 2 GB raw partitions. Similar effects can be achieved by AIX (and Sequent, and ...). I set up an OnLine system on such a machine earlier this month, and the 10 GB of RAID disk was itself mirrored by another lot of 10 GB of disk attached to another machine in another room and connected by fibre-optic cable to the primary machine. The secondary machine didn't access the second bank of disks unless it was brought up in disaster recover mode. The O/S handled the mirroring (or maybe it was the hardware -- I didn't enquire too much because I didn't need to know). So, it depends on your O/S, but if the O/S provides mirroring, the "raw" devices will not be as raw as they are on traditional systems. Yours, Jonathan Leffler (johnl@obelix.informix.com) #include <disclaimer.h>