Douglas Allan Tutty wrote:
On Fri, Apr 06, 2007 at 09:39:28AM +0800, Bob wrote:Interesting, the layout I was considering with 2 drives was to have the system (including swap and /tmp) on a RAID0 array with the resultant speed boost that entails, and have /home on a RAID1 array with the protection that offers. If a drive goes down the system crashes and I have to reinstall, but that's not hard and I've still got my personal data, plus I've been enjoying what should have been a noticeably more responsive system in the meantime.How is having the binaries on raid0 going to make the system niticeably more repsonsive? Raid1 does round-robin reading anyway. To make things faster, mount /usr noatime so no reads have to happen every time something is read.
You mean no *writes* have to happen every time something is read, and the faster the system can access its binaries the faster it feels (to a point) as the faster things load, particularly new programs, like the first time you start Open Office, if the binaries are on a fast drive they load faster. An experiment I want to run is to get a 4200 rpm SATA laptop drive and 10,000 RPM SATA Raptor of similar size and compare the responsiveness of the same install on the same hardware on different drives and then figure out what mountpoints should go on a CF card in a DMA capable IDE (or SATA) adapter to achieve similar speed, my hunch is I should put at least /var and /home on the harddrive and the rest on the CF card.
I'd like to use a ram disk fs (unionfs?) like a bootable cd where things get linked back to the media and you can replace a file in the running system, by replacing it in ramdisk, but then have a commit feature to update the CF card.
Unless you're short on ram, put /tmp on tmpfs. If you are short on ram, add more, then put /tmp on tmpfs.
I haven't tried tmpfs, but I thought that the standard place to put ISOs your building was in /tmp so if you have a DL DVDRW it should be 9GB, what else is tmp used for.
If the system crashes because of a drive failure, and your /home is on software raid1, what keeps the system crash from not crashing the software raid1, and the filesystem on top of it?
Nothing other than the unlikeliness of it, I'd have thought the RAID software was quite robust, certainly I've had powercuts to my file server and the RAID software grumbles, does some checking and comes up, then I do a consistency check of the file system (JFS on RAID5, no LVM ATM, it'll be interesting when I come to add another drive) and mount as usual.
RAID is not a substitute for backup anyway, it's meant to be a tool to increase availability by adding redundancy, in this case I could slap in a new drive, do a reinstall, rebuild my RAID1 array, and I'm back up without having to do all that tedious mucking about with DDS tapes (or DVDRs, Rsync, Gmail, whatever you use) but even if your data is hosed you can restore from backup (you *do* backup right) and your fine.
Actually I don't mean to sound quite so sanctimonious about it, my backup frequency is dictated by how much it going to piss me off if I loose everything science the last one, when I get the the point of losing sleep over it I dig out my DDS drive and SCSI card, drop them in and do a backup, (I really need a bigger box) then I sleep well for a couple of months.