Oracle FAQ Your Portal to the Oracle Knowledge Grid
HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US
 

Home -> Community -> Mailing Lists -> Oracle-L -> Re: OT: high WIO on Linux

Re: OT: high WIO on Linux

From: Christo Kutrovsky <kutrovsky.oracle_at_gmail.com>
Date: Mon, 13 Mar 2006 16:35:52 -0500
Message-ID: <52a152eb0603131335h43e5af25ld35193fba4cee985@mail.gmail.com>


In my opinion, the biggest problem with "%WIO" (wait io) is the confusion it creates. Same thing with "%util" from iostat.

Many people are either following the rule of thumb that WIO is idle time, others are thinking that it's the cpu needed to do IO.

The right way to look at it, in my opinion, is the "what if" case. Something like "had this process not waited for IO, it would've used that much CPU"

As far as "%util" , sometimes refered to "%disk time" (Windows) it's a bit special due to queue lenght and access pattern.

I like very much to explain disk array behaviour with an example of a library where your only access to books is via employees. You would stand at the desk and ask for a book, and the employee would go and get it for you. In the RAID array case, then you would have multiple employees at the desk, each assigned to a section of the library. If you ask for a book at a time, having more then 1 employee does not matter, and you could be "utilising" close to 100%. You could handle much more capacity, had you asked for multiple books at the *same* time as multiple employees could go and fetch them from different sections.

What I am trying to say is that you may have 80% disk usage, but in fact be able to handle 4x more IO.

--
Christo Kutrovsky
Senior Database/System Administrator
The Pythian Group

On 3/10/06, Kevin Closson <kevinc_at_polyserve.com> wrote:

> >>>
> >>>First of all, Linux sucks. Let me clarify things a bit: the
> >>>whole concept of %WIO makes me cringe. CPU works in user
> >>>mode, kernel mode, is on the interrupt stack or is idle.
> >>
> I agree with Mladen on on this aspect of Linux (and many others).
> The %WIO thing is goofy. I should point out, however, that the
> interrupt handlers are kernel mode cycles. So it is actually,
> Kernel,User, Idle loop. This is why a broken (chatty) hardware
> device runs CPUs to saturation in kernel (sys) mode.
>
> From a database perspective, Linux is about where good Unix
> implementations where circa 1995. There were so many Oracle-special
> OS features that the OSDs could exploit for scalability that
> Linux can't dream of offering because the focus of Linux is
> not RDBMS and the hardware is too general purpose--albeit EXTREMELY
> fast in the case of AMD.
>
> Like I say, Linux stands for (L)inux (I)s (N)ot (U)ni(X). But that
> doesn't make it hopeless either. You just have to watch your
> expectations.
>
> --
> http://www.freelists.org/webpage/oracle-l
>
>
>
-- Christo Kutrovsky Senior Database/System Administrator The Pythian Group -- http://www.freelists.org/webpage/oracle-l
Received on Mon Mar 13 2006 - 15:35:52 CST

Original text of this message

HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US