Current-Users archive

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

Re: command cache and io priority

On Sat, 16 Jan 2010 17:17:52 -0500
Noah McNallie <> wrote:

> Recently after configuring process scheduling priority and packet 
> scheduling priority, i tested nice with simply tar -zxvf of gnusrc, i 
> was amazed at what i assumed was a system that could take heavy load and 
> still respond.
> - operating systems that i know of seem to have a command cache, even 
> under heavy disk io load. if one where to say commit 'uname -a' it would 
> take perhaps 2 seconds to respond, though the next time it would respond 
> fairly instantaneously. NetBSD does not seem to be doing that, the 
> second go was just as slow; tested numerously
> - despite this, i feel there would have to be an io scheduler  in any 
> adequate operating system. linux i found has 'ionice' as part of their 
> cfq system so i'm assuming netbsd has this feature also, please correct 
> me if i'm wrong.

It's possible that for your particular setup you might need to play
with the BUFQ_READPRIO or BUFQ_PRIOCSCAN options, which have an impact
on disk I/O.

I remember about a project to write a more general I/O scheduler, but
I'm not sure what the status of that project is...

I'm not sure what's up with the repeated uname -a command taking a
while to execute though, as the buffer cache should normally make the
second and other invokations instantaneous.  Nothing particular shown
in dmesg?

Home | Main Index | Thread Index | Old Index