Subject: Re: Linux on a Netfinity </geekalert>
Sat Sep 05 15:02:49 1998



> I hope this post isn't too far off topic. If anyone can direct me to a
> better place to send this type of request, please let me know.

it's dead on target, as far as i can see.. the care & feeding of
servers is an important part of the web consulting game.   the type of
server you use (OS/hardware/connectivity) places a whole slew of
constraints on the tools which will be available to you in building a
site, and hence the site itself.



> The Redhat 5.1 release does not recognize the builtin IBM ServerRAID
> II controller. We tried a few of the Adaptec choices to no luck ...
> any suggestions?

we ran into the same problem here a few months ago, trying to find a
RAID driver for a set of compaq Proliant servers.   eventually, we
gave up, scrapped the RAID idea, and just mounted the drives with a
plain-vanilla Adaptec SCSI controller.   apparently, the RAID and SCSI
Ultra Wide controllers out there are still new enough that there
aren't any stable drivers for them, yet.

the good news is that you don't really lose much by mounting the
drives normally.   yes, adding more drives if you need to expand your
storage is a bit less convenient, but the difference is negligible..
five minutes versus two, once every couple years or so.


the system we've worked out here (and which i'm rather proud of,
actually), is to mount the drives as bulk storage under the /mnt
directory, and store your web data there:

    /mnt/
        drive_A/
            p_01_150M/
                domain_one.com/
                domain_two.com/
                domain_three.com/
            partition_02/
            partition_03/
            partition_04/
        drive_B/
            p_01_2400M/
                big_domain.com/
            partition_02/
            partition_03/
            partition_04/


then you use symlinks in the htdocs/ directory, to point the httpd to
the right files for each domain:

    /httpd/htdocs/
        domain_one.com      -> /mnt/drive_A/p_01_150M/domain_one.com/
        domain_two.com      -> /mnt/drive_A/p_01_150M/domain_two.com/
        domain_three.com    -> /mnt/drive_A/p_01_150M/domain_three.com/
        big_domain.com      -> /mnt/drive_B/p_01_2400M/big_domain.com/


the links give you a layer of abstraction.   they allow you to separate
your file storage from the web file tree.   that's a good thing,
because the two systems have different goals.   a good storage system
is flexible and expandable.. you can move things from place to place
easily, and add more storage without having to play silly games.   a
good web tree, on the other hand, is very sparse and clean, giving you
URLs which are simple, and easy to remember.

if you try to make one system do both jobs, you end up doing one thing
well and one badly, or spreading the badness out between both sides.  
neither option is very appealing.   by using links to separate the two
systems, you can do both well, then map one onto the other.