I will be running postgres with about 100gb of data. Their will be few quite big insert transactions (e.g 1000 updates every 30 sec or so) from max. 10 processes. This will be running in snapshot isolation mode.
Then their will be alot of read only transactions from a various number of processes.
I want to secure the data against loss of power or disk errors, I don't care about fire, crashing planes or leaking nuclear plants. I will only need the WAL until the db flushes changed pages to disk.
I have 6 sas disks and 12g memory. How should I split the disks?
I was thinking of putting ths os and the data on 4 disks (RAID10) and the log on 2 disk (RAID1).
Is this a good idea or stupid? WAL shall never be slowed down, but it will be only using about 1 gig of the disk. I've got the feeling of wasting 2 disks. Would it have an negative effect if I keep all 6 disks as one RAID 10?
I've got a battery backed controller, but would the write cache be clever enough to always cache the WAL? I don't think so... :-(