Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
On 27.09.2007 16:27, Rustedt, Florian wrote: > MD-Raid mirror on the disks per server, then LVM ontop and then DRBD on the > resulting partititions? Or LVM -> RAID -> DRBD? At least if you plan to use proto C and care about performance, I consider it essential to have a battery backed hardware controller and to not use soft raid. It's much faster to commit a changed block to controller cache than to physical media. > And what is the right setup for the partitions? > > The ideal-case would be, that both systems are able to boot the same data > from scratch, but i saw in the list and google, that most systems do NOT > use DRBD for the root-partition? Why not? Isn't it stable enough? Well, first you'd need at least an initrd for it, second I don't think it's a good idea to boot two identical systems. The systems would need to get their IP addresses and config from DHCP. > So what is the best way to partition tihs beast? And why? > > And is basically LVM a good choice or does it have a remarkable impact on > performance, so it would be better, to keep rid of it? It has a measurable impact. It depends on type of load and the hardware, but I consider it negligible. About 4-6% for PostgreSQL OLTP load on top of HW Raid 10 - to give you an idea. -- Regards, H.D.