| 1 |
On Fri, Jun 21, 2013 at 1:40 PM, Mark Knecht <markknecht@×××××.com> wrote: |
| 2 |
> One place where I wanted to double check your thinking. My thought |
| 3 |
> is that a RAID1 will _NEVER_ outperform the hdparm -tT read speeds as |
| 4 |
> it has to read from three drives and make sure they are all good |
| 5 |
> before returning data to the user. |
| 6 |
|
| 7 |
That isn't correct. In theory it could be done that way, but every |
| 8 |
raid1 implementation I've heard of makes writes to all drives |
| 9 |
(obviously), but reads from only a single drive (assuming it is |
| 10 |
correct). That means that read latency is greatly reduced since they |
| 11 |
can be split across two drives which effectively means two heads per |
| 12 |
"platter." Also, raid1 typically does not include checksumming, so if |
| 13 |
there is a discrepancy between the drives there is no way to know |
| 14 |
which one is right. With raid5 at least you can always correct |
| 15 |
discrepancies if you have all the disks (though as Duncan pointed out |
| 16 |
in practice this only happens if you do an explicit scrub on mdadm). |
| 17 |
With btrfs every block is checksummed and so as long as there is one |
| 18 |
good (err, consistent) copy somewhere it will be used. |
| 19 |
|
| 20 |
Rich |
Archives