paragasu wrote:

    There are three I/O schedulers in the 2.6.24 distribution.  The
    deadline, anticipatory, and completely fair.  You will see these as
    options when you run make gconfig (or whichever config you run).
    Completely Fair (CFQ) is now the default.  This was mildly controversial
    when they made the switch.  It is a bit complex.  The deadline scheduler
    is simple and compact, and is alleged to be good for heavy database
    workloads.

    This may be a bit like making a mountain out of a molehill.  I just
    mentioned this because while you are configuring you custom kernel, you
    will want to give a little thought to such matters.

    see /usr/src/linux/Documentation/ to start with.

thanks Mark Allums,
Honestly, i spent less time looking on the kernel side. i will look at the documentation
for the I/O scheduler options. great information you have ;)


There is a lot more than just the I/O scheduler. It just popped into my head because I am doing the configuration right now, myself, and because of the debate/controversy about it. Also, you will want to build the device drivers you need as modules, and configure the system to unload modules when they aren't needed.

Trim everything mercilessly. Examples: You don't need battery support if the computer is not a laptop. If you don't need anything but basic cryptographic support, leave it out, and do crypt in user space. If the thing only has one NIC, or one type of NIC (i.e., all the same), leave out support for all the others. (In general, if you don't have the hardware, you don't need the code.) Leave out as much debug code as you can.

And so on, etc.

The 2.6 kernel is a little bloated IMO. I don't think that three I/O schedulers need to be compiled in, but you can do that, and pick which one to use at boot time. (For example.) (There's a config file.)

I would suggest going to the 2.4 kernel, but too much hardware and software depends on 2.6 now to revert.



--
Mark Allums


--
To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]

Reply via email to