Based on various googled comments I have selected 'deadline' as the elevator for the disks comprising my md arrays, with no further tuning yet ... not so stellar :(
Basically concurrent reads (even just 2, even worse with 1 read + 1 write) don't work too well. Example: RAID1: I bulk-move some 90GB of files onto the mirror, which takes a while. Meanwhile I want to do an ls on another directory on the same mirror. The ls output fits on 20 lines but takes in excess of 10 seconds. RAID5: That's where the 90GB above are coming from. Concurrently my music player wants to load a song every 3-5 Minutes from the same array. Takes 5-10 seconds. During the transfer I also tried to open a samba share on another disk which was idle at the time - still took about 3 seconds, as samba itself is on the mirror. The bulk transfer (one mv) seems to utterly starve everything else. CPU load is non-existent, all of 2GB RAM is used for cache, swap is unused. I realize there's no way to tell the kernel which traffic is interactive and which is bulk, but I also don't see why deadline should starve so much. Any pointers? Thanks, Christian - To unsubscribe from this list: send the line "unsubscribe linux-raid" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
