Re: High performance random-access disk operations (Raidframe?)

[David Brownlee <abs%netbsd.org@localhost>]

On 16 May 2010 02:17, Chris Ross <cross+netbsd%distal.com@localhost> wrote:
>
> On May 15, 2010, at 17:53, Matthew Mondor wrote:
>>
>> I've indeed seen "RAID 10" used for such circumstances (that is, a RAID
>> 1 for redundancy, which is stripped using RAID 0 for performance). ÂI
>> would assume the raidframe performance to be decent with a good
>> configuration, unless the machine is quite slow...
>
> ÂWell, there are a lot of issues here. Â(1) RAID 1+0 requires 4 disks Â
> (right?). ÂNot necessarily a problem, but worth noting. Â(2) I don't actually 
> care about the data, so the cost of the "RAID 1" part is a waste. Â(3) As it 
> turns out, the filesystems I was originally looking at combining *aren't* t 
> he same size, which is unsupported by RAIDframe anyway.
>
> ÂSo, I'm working with ccd(4) at the moment. ÂWe'll see how that works.
>

At least for raid1 raidframe will just ignore any additional space on
the larger partition, which may be an option.

You mention (in another email) that you're using this for squid so
presumably you'll have the squid config sized to use all the ram.
Otherwise if you didn't care about the data between reboots more
esoteric ideas could include using them as swap and setting up a large
tmpfs filesystem (I have a box with a 12GB /tmp on tmpfs). Not that
thats necessarily a *good* idea in your case :)

I assume you're mounting the partition with log? based on your
concerns about data loss you could also try async (and ensure any boot
fsck is run with -y)