Is there any way to say which is "better"? I'm trying to set up our server here. 4 scsi disks of equal capacity, should do raid5. But the question is - how to set up filesystems on them? We'll need several addtitional filesystems in additional to standard ones. There are 2 options. a) create set of partitions on each, one partition for each /usr, /var, /home, etc., and create one raid device for each filesystem. I.e. sd[abcd]2 => md2 => /usr, sd[abcd]5 => md3 => /var etc. b) create one large partition, create raid array on it, and use something like LVM to "partition" it. Some essential (system) partitions (/usr, /var) may be created on separate raid arrays (as in option 2) in order to be less dependant on LVM (I still can't trust it fully). I know LVM's variant gives me more flexibility in layout, esp. having in mind that we'll use Oracle on raw partitions heavily, and that SCSI disks may have only 15 partitions max (this is terrible small). But the question is - will 2 raid arrais created on the same set of disks work "better" than one, if accesses concurrently? I.e. having 2 concurrent processes that are both I/O bound but are working with different filesystems, which layout of underliyin devices is preferrable - 2 raid arrays or one large divided into 2 pieces somehow? In principle, the end factor is disks anyway, but will raid code block other process while servicing requests from first process in this case too? Disk drives can do TCQ and can reorder/optimize requests somehow, I don't know if this helps here or not. May raid benefit from SMP machine in this context? I think not, since it should ensure consistency by doing some internal locking anyway. Well, for 2 arrays, 2 CPUs may be better than one (not having in mind other processes), but one array seems to be for one CPU anyway, correct? :) And another question, probably an off-topic here, is - raid5 + lvm combo - how it perform? Is it worth it to try? My tests shows that LVM may significantly slow disk access down for raw devices, and this is quite unacceptable for us. But I can't imagine how it will work once set up for real life load. The problem is that it's really difficult to set up a test machine: the whole process takes several 10s of hours. I've already set up multiple raid5 devices as in variant a) above, it seems to work almost ok but system CPU usage is relatively high. Thank you. /mjt - To unsubscribe from this list: send the line "unsubscribe linux-raid" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
