Current-Users archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

Re: raidframe performance question

        hello.  As far as I know, the boot blocks don't know how to boot off
of a raid5 set directly at all.  Having wanted to do what you did many
years ago, I did the small non-raid partition at the front of each disk
trick so the boot blocks can get the kernel loaded and then the kernel can
load the raid5 set as root, just as you  expect.  This gives me the ability
to boot the system no matter which disks I might lose, but still have the
redundancy of having all filesystems on a raid5 set.
        The only trick is that when I upgrade the system, I have to remember
to put new kernels on the little non-raid partitions, because the kernel in
/ isn't the kernel actually loaded.
        So the steps look like:

1.  Set up the disklabel on each disk, leaving a small 1GB partition at the
beginning for a copy of an OS installation and reserving the rest for the

2.  Configure the raid5 set.

3.  Partition the raid5 set as appropriate, /, /var, /usr, etc.

4.  Load OS onto the raid5 set

5.  Load OS onto one of the non-raid partitions.

6.  Dump/restore that non-raid partition onto the other non-raid partitions
so as to insure identical copies of the installation on each disk.

7.  Installboot on each disk, asking the boot blocks to boot off the 'a'
partition of each disk.

8.  Run raidctl -A root raid0
or wich ever raid set you want to be root.

        Now, when you boot the system, you'll load the kernel from one of the
non-raid partitions, but when the system is up, it looks like it's fully
self contained on the raid5 set.
My df looks like:

%df -k
Filesystem  1K-blocks      Used     Avail Capacity  Mounted on
/dev/raid0a  19543657   2624395  15942080    14%    /
/dev/raid0e  48860623  22788746  23628846    49%    /var
/dev/raid0f 600067956  62037020 508027540    10%    /usr/home
/dev/raid0g 200020836  27003860 163015936    14%    /usr/local
/dev/raid0h  97722686  18395355  74441197    19%
/dev/raid0i 112518729  10325793  96567000     9%    /var/yoshipics

On Jun 15, 11:23am, Dave Burgess wrote:
} Subject: Re: raidframe performance question
}   On 5/13/2011 8:40 AM, Paul Goyette wrote:
} > Nope.  -A simply specifies the alignment of the DOS partitions (and 
} > the offset of the first partition).  If your drive might have "native" 
} > sectors larger than 512 bytes, you need to use this, otherwise all 
} > your I/Os will get split across drive-addressable sectors and 
} > performance will be very bad (tm) as the drive reads one physical 
} > sector, updates and re-writes the latter portion, then reads the next 
} > sector and updates the early portion.
} >
} > Note that most versions of NetBSD's fdisk(8) have a parsing bug, and 
} > will not accept "-A 2048" (where the offset is supposed to default to 
} > the alignment).  Instead you have to specify "-A 2048/2048".  I fixed 
} > this error as soon as I found it!
} >
} Paul,
} I'm trying something a little more advanced and am having some problems.
} Instead of trying to do what you did (which does seem to work fine) I'm 
} trying to build a new system on a new RAID-5 array without any 
} "non-RAID" partitions.  I'm also doing this from the install CD (and to 
} make it even more challenging, the AMD64 5.1 boot CD).
} Before the install, I drop into the shell and do all of the setup for 
} the raid array.  I get the array all built and configured, and then drop 
} back into the install program and run through the "normal" steps, but 
} install on the RAID array instead of one of the disks.
} Everything works OK until I try to boot.  If I set up a small wd0 root 
} partition I can get the boot sectors to boot the RAID-5 array, but if I 
} build the RAID array, use the whole set of disks for the array, and try 
} to boot, I get all sorts of boot problems.
} If I set up a minimal install on each of the wd[0-3] drives (since 
} there's no SURE way to predict which drive the BIOS is going to try 
} booting from) and set up the installboot for each, I can get the array 
} to boot with no problems.
} Once the array boots up, everything is wonderful, using the 2048 
} alignment and making sure all of the "logical drives" (raid0a, etc.) 
} aligning on 1M boundaries makes this WICKED fast.
} If anyone has any idea how to get the RAID-5 array to boot after the 
} "raidctl -A root" and the appropriate disklabel and installboot 
} directives, I'd love to hear them.
} Dave
>-- End of excerpt from Dave Burgess

Home | Main Index | Thread Index | Old Index