Current-Users archive
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]
Re: raidframe performance question
After posting lots of questions, and getting lots of expert help, I
finally got my new server up and running with raid! Since I asked
so many questions, it was suggested that I recap the process so that
someone else might be able to make it work without having to ask!
So ...
Here's what I did recently to get raidframe(4) up and running on my
new server.
0. This is really important! Read the man page for raidctl(8). It
is a long man page, but it is important to understand how things
work. And there are some actual examples towards the end, which
will only make sense if you've read the intro material. Don't be
intimidated by all the detail, just RTFM before you start!
1. Use dd(1) to completely erase the drives. This is useful to make
sure that any existing labels/wedges/etc. get removed, and also
makes it easier to initialize the raidset parity. On Western
Digital SATA-II 500GB drive, it took less than two hours (each)
for this step. (The operation starts off at ~110GB/sec, but will
get slower as the heads get closer to the center of the drive
where there are fewer sectors per cylinder.)
dd if=/dev/zero bs=32k of=/dev/rwd2a
dd if=/dev/zero bs=32k of=/dev/rwd3a
2. (This step might not be necessary but I wasn't taking any chances!)
Use fdisk(8) to set up DOS/Windows partition data. Make sure you
use the "-A 2048" option for alignment.
fdisk -uai -A 2048 wd2a
fdisk -uai -A 2048 wd3a
(If ytou are using an older version of fdisk and get the error
fdisk: Bad argument to the -A flag.
you can specify "-A 2048/2048" instead.)
Set the partition 0 type to 169 (NetBSD), and let it occupy the
entire drive. If you're going to boot from this raidset, you
should also make this the active partition, and update the boot
blocks when prompted.
3. Use disklabel(8) to provide a NetBSD disklabel on each member of
the raidset. Create an 'e' partition that describes the whole
disk - so it is essentially a copy of the 'd' partition. Here's
my partition info:
5 partitions:
# size offset fstype [fsize bsize cpg/sgs]
c: 976771120 2048 unused 0 0 # (Cyl. 2*- 969020)
d: 976773168 0 unused 0 0 # (Cyl. 0 - 969020)
e: 976771120 2048 RAID # (Cyl. 2*- 969020)
4: Now create a raid.conf file to describe the raidset. Mine is a
simple RAID-1 (mirror):
START array
#numrow numcol numspare
1 2 0
# Identify physical disks
START disks
/dev/wd2e
/dev/wd3e
# Layout is simple - 64 sectors per stripe
START layout
#Sect/StripeUnit StripeUnit/ParityUnit StripeUnit/ReconUnit RaidLevel
128 1 1 1
# No spares
#START spare
# Command queueing
START queue
fifo 100
5. Initialize the raidset using raidctl(8). Until you do this, there
are no raid "component labels" on the member drives, so you need to
use the "-C" option; if the raidset has been previously initialized,
you should use "-c" instead (this is what /etc/rc.d/raidframe uses
at system startup).
raidctl -C /etc/raid1.conf raid1
6. Next, initialize the component labels with raidctl.
raidctl -I 201105061405 raid1
(For the serial number, I just used the current date/time; any 64-bit
integer will work.)
7. Now use raidctl to initialize the parity of the raidset. On my 500GB
drives, this took about 1.5 hours.
raidctl -i raid1
8. Now you can treat the raidset as a normal disk. First you need to
write your NetBSD disklabel (this label is separate from the label
on the physical drive). Make certain that all the offsets are
multiples of the drive's native block size; otherwise you will have
unaligned transfers and performance will be very bad! (This is
also why, in step X above, we use "-A2048" for fdisk rather than
letting it default to 63!")
Here is what my drive looks like:
7 partitions:
# size offset fstype [fsize bsize cpg/sgs]
c: 976770944 0 unused 0 0 # (Cyl. 0 - 953877*)
d: 976770944 0 unused 0 0 # (Cyl. 0 - 953877*)
e: 536870912 0 4.2BSD 2048 16384 0 # (Cyl. 0 - 524287)
f: 104857600 536870912 4.2BSD 2048 16384 0 # (Cyl. 524288 - 626687)
g: 335042432 641728512 4.2BSD 2048 16384 0 # (Cyl. 626688 - 953877*)
9. Now, just as you would for a real disk, use newfs to initialize the
filesystems. Then mount them and start using your new raidset!
newfs /dev/raid1e
mount /dev/raid1e /mnt
10.If your kernel includes the RAID_AUTOCONFIG option, you can use
raidctl to let your system automatically reconfigure the raidset at
every boot. (If you don't do this, system startup will still take
care of this, as long as your configuration file is properly named
"raid[0-9].conf" or "raid[1-9][0-9].conf" and is located in /etc
directory.)
raidctl -A yes raid1
11.If you're going to boot from the raidset, you probably want to have
an 'a' partition. Here's the partition from my other, bootable,
raidset:
6 partitions:
# size offset fstype [fsize bsize cpg/sgs]
a: 41943040 0 4.2BSD 2048 16384 0 # (Cyl. 0 - 40959)
b: 62914560 41943040 swap # (Cyl. 40960 - 102399)
c: 488395008 0 unused 0 0 # (Cyl. 0 - 476948*)
d: 488395008 0 unused 0 0 # (Cyl. 0 - 476948*)
e: 125829120 104857600 4.2BSD 2048 16384 0 # (Cyl. 102400 - 225279)
f: 257708288 230686720 4.2BSD 2048 16384 0 # (Cyl. 225280 - 476948*)
12.You will also need to run installboot(8) to make the raidset bootable.
Do this for each member drive:
mount /dev/raid0a /mnt
installboot /dev/wd0a /usr/mdec/bootxx_ffsv1 /mnt/boot
installboot /dev/wd1a /usr/mdec/bootxx_ffsv1 /mnt/boot
13.Finally, you need to make sure your kernel has the RAID_AUTOCONFIG
option, and enable the raidset as the root device:
raidctl -A root raid0
-------------------------------------------------------------------------
| Paul Goyette | PGP Key fingerprint: | E-mail addresses: |
| Customer Service | FA29 0E3B 35AF E8AE 6651 | paul at whooppee.com |
| Network Engineer | 0786 F758 55DE 53BA 7731 | pgoyette at juniper.net |
| Kernel Developer | | pgoyette at netbsd.org |
-------------------------------------------------------------------------
Home |
Main Index |
Thread Index |
Old Index