Securing my data
My machine is now running with /home
on my RAID-1 mirror. When I
booted my machine with one of the drives turned off (by removing the
power cable) it didn’t make a fuss. Putting the “faulty” disk back
online was a simple matter of adding it back to the array. The RAID then
resynched and was back to normal status after 40 minutes. Pretty
cool!
I ran bonnie++ to test the read performance on my regular
/dev/sda
drive and on the /dev/md0
RAID mirror. The results from
my normal disk:
------Sequential Output------
-Per Chr- --Block-- -Rewrite-
K/sec %CP K/sec %CP K/sec %CP
33601 94 49338 15 17551 4
--Sequential Input- --Random-
-Per Chr- --Block-- --Seeks--
K/sec %CP K/sec %CP /sec %CP
16533 46 42633 6 184.3 0
------Sequential Create------
-Create-- --Read--- -Delete--
/sec %CP /sec %CP /sec %CP
22150 87 +++++ +++ 22595 99
--------Random Create--------
-Create-- --Read--- -Delete--
/sec %CP /sec %CP /sec %CP
21485 87 +++++ +++ 21210 99
compared to the results from my RAID mirror:
------Sequential Output------
-Per Chr- --Block-- -Rewrite-
K/sec %CP K/sec %CP K/sec %CP
32071 91 54533 16 24885 6
--Sequential Input- --Random-
-Per Chr- --Block-- --Seeks--
K/sec %CP K/sec %CP /sec %CP
22946 62 52236 6 382.1 0
------Sequential Create------
-Create-- --Read--- -Delete--
/sec %CP /sec %CP /sec %CP
26386 92 +++++ +++ 24380 99
--------Random Create--------
-Create-- --Read--- -Delete--
/sec %CP /sec %CP /sec %CP
27755 99 +++++ +++ 22607 99
The performance is better in all areas, except that the CPU utilization is a tad higher. The read performance went up from 42 MiB/s to 52 MiB/s, an increase of 25%. I expected an increase, but it could have been bigger considering that the read requests are balanced over the two drives. But then again, the main goal of the RAID was to make sure that my data will be kept safe, so the increase in performance is just an added bonus!
Even the write performance went up, from 49 MiB/s to 54 MiB/s. This is a bit strange, since every RAID Howto I’ve read explains how the write performance should drop using RAID-1. This is because each block is put on the bus twice, once for each disk. But who am I to complain? :-)
With my data backed by two disks I’m feeling fairly safe on that
front. Of course my computer could still be stolen, hit by lightning,
or the data could simply be deleted. To protect against the latter
I’ve installed dirvish to take backups of /home
to my normal
disk.
These backups are made daily, and rotated so that I have images for the last two weeks. The nice feature of dirvish is that the backups are live — they exist on my disk as a normal filesystem tree.
Normally it would require a hidious amount of space to keep two weeks worth of full backups, but since dirvish only uses space for new and changed files it should do just fine with my 80 GB disk. The trick is using hard links for files which haven’t changed between backups — hardlinks take up almost no space,or rather, they take up inodes, but ReiserFS (which is the filesystem I use) allocates inodes dynamically as needed so I wont suddenly run out of them.
So with my data spread over no less than three disks I can sleep with ease at night :-)
Lars Geisler:
The write performance speed-up might be due to the simple fact that the new disks are faster than the old on :-)
24 February 2005, 11:35 amMartin:
Oh yes, of course. The old is a 80 GB Seagate Barracuda 7200.7 SATA disk and the new are 120 GB Seagate Barracuda 7200.7 SATA NCQ disks, both of which have similar specifications. So perhaps it is the NCQ that makes the difference? I don’t know.
24 February 2005, 1:33 pmThomas Mølhave:
Well, I know your IP-address, so let’s see how long your puny disks will have any meaningfull data. I you suddenly find your disk filled with random primes, you willl know I was there!
My kungfu is better than yours shimo!
24 February 2005, 11:26 pmMartin Reise:
I think your new disks are just faster then your old.
The newer disks have more cache, I think.
The higher CPU utilization is caused by slow bus-system.
25 February 2005, 3:32 amMartin:
No, all three have 8 MiB of cache so it’s not that. It would have been an obvious explaination though.
25 February 2005, 6:57 pmMartin:
Ehh? :-) I’m in Aalborg at the moment, behind my fathers firewall… so just come ‘n’ get me!
25 February 2005, 6:58 pm