[Disksim-users] Request coalescence

[YipKei Kwok]

Hi Greg and the Disksim community,

Thanks for the hint. I analyzed my experiment results right away.

However, the cache hits cannot justify the number of missing requests.
Here, I define the number of missing requests to be the difference
between number of requests specified in the trace and the number of
"Total Requests handled" by the Overall I/O System, IOdriver and Disk.

Here is some data. (Note: The result is obtained with a fresh Disksim
installation with validation test passed)

-- #(missing requests) = 92359 - 86339 = 6020
-- System logorg #0 disk Buffer hit ratio:          2879   0.033345
-- Disk Buffer hit ratio:          2879    0.033345

Note: 2879 is the hit count.

There is quite a distance between the number of missing requests
(6020) and the hit count (2879).

I am looking forward to hear from you. Thank you!

-Yipkei

P.S. I have some other questions about Disksim and I will send it in
another email.

On 5/13/07, Greg Ganger <ganger at ece.cmu.edu> wrote:
>
> Cache hits would cause the number of requests at the disk to be less
> than the number in the trace.  That's something to look at.
>
> Greg
>
>
> On Sun, 13 May 2007, YipKei Kwok wrote:
>
> > Hi,
> >
> > I am running Disksim w./ a trace workload.
> >
> > Problem:
> > The amount of "Total Requests handled" by the disk is less than the
> > number of requests specified in the trace.
> >
> > I suspect that is because of request coalescence/overlapping/combining.
> >
> > disksim_ioqueue: Overlapping request scheme
> > -----------------------------------------------------------
> > - We have already set this parameter 0 so that overlapping requests
> > are treated as independent.
> >
> > disksim_disk: Combine seq writes
> > -------------------------------------------
> > - The default value of our HP C3323 hard drive is 1. We experienced
> > _exactly_ the same problem even when we set it to 0.
> >
> > I/O Driver Statistics
> > -------------------------
> >> From the statistics, both "IOdriver Overlaps combined" and "IOdriver
> > Read overlaps combined" are both 0. So, no requests are combined on
> > the I/O driver level.
> >
> >
> > Is there any other level of the I/O hierarchy could cause request
> > coalescence? Or, are there some other reasons, other than request
> > coalescence, which cause the problem that I am facing?
> >
> > Thank you very much!
> >
> > -Yipkei
> >
> > P.S. Here is our experimental setup.
> > I/O trace-driven
> > raid 0 - 1 HP C3323A hard drive.
> > (Originally, there are 4 disks in the RAID. For simplicity, now we
> > only use 1. So, strictly speaking, it's not a RAID.
> > Number of processors = 1
> >
> > --
> > Web site:
> > http://360.yahoo.com/yipkeikwok
> > http://www.mcs.csueastbay.edu/~ykwok/
> > email addresses:
> > yipkeikwok at gmail dot com
> > ykwok2 at utep dot edu
> > ICQ UIN:    2309842
> > Google Talk ID:    yipkeikwok at gmail dot com
> > Yahoo! Messenger ID:    yipkeikwok
> > MSN Messenger ID:    yipkeikwok at hotmail dot com
> > _______________________________________________
> > Disksim-users mailing list
> > Disksim-users at ece.cmu.edu
> > https://sos.ece.cmu.edu/mailman/listinfo/disksim-users
> >
>

-- 
Web site:
http://360.yahoo.com/yipkeikwok
http://www.mcs.csueastbay.edu/~ykwok/
email addresses:
yipkeikwok at gmail dot com
ykwok2 at utep dot edu
ICQ UIN:    2309842
Google Talk ID:    yipkeikwok at gmail dot com
Yahoo! Messenger ID:    yipkeikwok
MSN Messenger ID:    yipkeikwok at hotmail dot com