On 05/27/2014 02:03 PM, Stefan G. Weichinger wrote: > I think I have some IO-topic going on ... very likely some mismatch of > block sizes ... the hw-raid, then LVM, then the snapshot on top of > that ... and a filesystem with properties as target ... oh my. Chosing > noop as IO-scheduler helps a bit but maybe I have to roll back and > rebuild one of the HW-RAID-Arrays with a different blocksize. Stefan
Hi Stefan, block size / stripe size mismatches only really penalise random io, if you are trying to use dd and have slow speeds this would suggest something else is awry. I don't know the c600 rad chip personally, but in trying to google it it appears to be a motherboard based raid device? is it a real raid or fakeraid? I'm a little confused over your setup to help. i'm sorry if there is duplication but it would be useful to have all info in one hit rather than trying to piece it together from all your messages. 1. please can you list your hardware raid config. I'm looking for the physical disk sizes, the virtual disks and their raid types. do you have cache enabled on the raid card, is there a background scrub or anything like that running? do you have active seek/ prefetch configured ? parity size being 50% of total size is just odd to me - but i guess these are mirrors ? but it says raid-level3 --- just odd, most setups use raid0(not raid) raid1(mirror) raid5(parity stripe) raid6 (double parity stripe) or combinations, like 50.. raid3 is allocating a single disk to parity but is very rarely used. 2. how many other devices are actively doing IO? do you have any other raid cards/io cards of note that might be clashign on the board. 3. do you have active I/O when doing your performance tests? if you have several virtual machines running depending on what they are doing they will crucify your access. 4. are you using any type of CGroups ? 5. i'm also confused over your LVM config. please can you send the output of "vgs" "pvs" and "lvs -a -o +devices" 6. please also send the output of "mount" 7. do you have atop or iotop that you can use to monitor performance - specifically we are looking for disk ios per device and disk latency per device. both before and during you are trying to run your backup. this should give us a better idea of where the problems lay.
