Subject: Re: RAIDFrame and RAID-5
To: Thomas Hertz <thomas@hz.se>
From: Greg A. Woods <woods@weird.com>
List: current-users
Date: 09/08/2003 15:25:30 
[ On Sunday, September 7, 2003 at 22:31:47 (+0200), Thomas Hertz wrote: ]
> Subject: RAIDFrame and RAID-5
>
> I've been running a raid-5 set with three disks more or less
> successfully for several months now. It took a little tweaking to get it
> reasonably stable (with alot of help from the tweaks by Paul Ripke in
> kern/20191). Last week I got myself another two disks, and went through
> the process of backup up all my data to create a brand new shining
> five-disk raid-5 array.
>  
> Apparently this was a mistake. It seems as if the bigger the diskspace,
> the more unstable is RAIDFrame. So, does anyone have any piece of advice
> to make this usable? Should I increase uvmexp.reserve_kernel even more?
> What are the consequences of doing so? (it's set to 48 now, was thinking
> of increasing it to 128).

I've got two RAIDframe RAID-5 arrays on my development server at the
moment:  a 5-disk array for /home and a 4-disk test array.  I have had
some problems which seemed to involve softdep during shutdown and
crashes, particularly with an old 1.5W kernel, but without softdep
they've been been quite solid.

Note I've not made any of the changes suggested in PR#20191.

I've not yet had enough experience with it running under 1.6.2_STABLE to
know if it's safe yet again to turn on softdep, but I'm going to assume
not for the time being.

My system has 320MB of RAM.  Sysctl says:

	vm.nkmempages = 20455
	vm.anonmin = 10
	vm.execmin = 5
	vm.filemin = 10
	vm.maxslp = 20
	vm.uspace = 8192
	vm.anonmax = 80
	vm.execmax = 30
	vm.filemax = 50

and "vmstat -s" says, amongst many other things:

        1 reserve pagedaemon pages
        5 reserve kernel pages

I don't know that I've pushed it all that hard but I have done full
builds with all but source in my $HOME, including simultaneously running
the build both locally and on another relatively fast machine using NFS
and with switched 100baseTX/FDX between the two.

I hope to convert the 4-disk array into a 7-disk array with a hot spare
soon so I'll see if that makes any difference.


> <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.0 Transitional//EN">

Please do not ever send HTML, rich text, or otherwise stylized e-mail,
to me or to any public mailing list.  Not all mail readers will
recognize such formats, and their added volume is generally a total
waste of bandwidth, storage, and processing power for everyone.  HTML in
particular is a potential security threat, so much so that some
firewalls and some mailing lists filter it entirely -- especially since
CERT and Microsoft have jointly anounced a very major flaw in the HTML
rendering engine used in all Microsoft products (in versions still
widely in use, and which isn't even properly fixed in the most recent
releases).

For more information see, for instance, the following articles:

	http://www.georgedillon.com/web/html_email_is_evil.shtml

	http://www.georgedillon.com/web/html_email_is_evil_still.shtml

	http://www.greydragon.org/library/email_list_etiquette.html

Please send all your messages as plain text only.


-- 
						Greg A. Woods

+1 416 218-0098                  VE3TCP            RoboHack <woods@robohack.ca>
Planix, Inc. <woods@planix.com>          Secrets of the Weird <woods@weird.com>