Hi,

Suppose we have a database or a filesystem over a disk.

Suppose we are keeping a record of the number of reads and writes per
a periodic interval (say 1 min) coming onto a disk. based upon this
history is it possible to predict what will be next few number of
reads/writes per interval?

If yes how much is that forecasting reliable?

Thanks
ASh

[ Post a follow-up to this message ]

Ash wrote:
> Hi,
>
> Suppose we have a database or a filesystem over a disk.
>
> Suppose we are keeping a record of the number of reads and writes per
> a periodic interval (say 1 min) coming onto a disk. based upon this
> history is it possible to predict what will be next few number of
> reads/writes per interval?

Yes, its possible, but there are a lot of potential variables that could
make the prediction little more than guesswork. If the environment is
relatively constant, (i.e. same application, same usage pattern etc) then
prediction will be statisically useful, the more things change though (say
you have two applications on the system that get used to differing degrees)
then the less useful the prediction becomes.

>
> If yes how much is that forecasting reliable?


Define "reliable" how do you want to use the data?


--
Nik Simpson

[ Post a follow-up to this message ]

Thanks Nik for your reply.

See what I mean is that suppose we have got oracle database over a set
of disks. Due to lot of load we find that one of the disks is proving
to be bottleneck. Now I want to predict whether this disk shall remain
as bottleneck for long or not. If yes then I shall use a much faster
disk in place of that. That is what I mean by reliable. Suppose we
predict that certain type of disk load  remains decreasing the
response time in future, then only we shall take some action of
replacement. Otherwise if it is a transient bottleneck it wont make
any sense to   do any action. right?



"Nik Simpson" <n_simpson@bellsouth.net> wrote in message news:<XcPlc.21$UG2.3@bignews2.bells
outh.net>...
> Ash wrote: 
>
> Yes, its possible, but there are a lot of potential variables that could
> make the prediction little more than guesswork. If the environment is
> relatively constant, (i.e. same application, same usage pattern etc) then
> prediction will be statisically useful, the more things change though (say
> you have two applications on the system that get used to differing degrees
)
> then the less useful the prediction becomes.
> 
>
>
> Define "reliable" how do you want to use the data?

[ Post a follow-up to this message ]

"Ash" <amujoo@yahoo.com> wrote in message
news:60aab6b4.0405042222.7dc8b7db@posting.google.com...
> Thanks Nik for your reply.
>
> See what I mean is that suppose we have got oracle database over a set
> of disks. Due to lot of load we find that one of the disks is proving
> to be bottleneck. Now I want to predict whether this disk shall remain
> as bottleneck for long or not. If yes then I shall use a much faster
> disk in place of that. That is what I mean by reliable. Suppose we
> predict that certain type of disk load  remains decreasing the
> response time in future, then only we shall take some action of
> replacement. Otherwise if it is a transient bottleneck it wont make
> any sense to   do any action. right?

Just observe the load over some significant period of time (at least days).
If only one or two of your disks seem to be a bottleneck over that time
(indicating that the load is not well-balanced over the storage), and the
bottleneck is an actual problem (either constant, or intermittent but
severe), then either redistribute the load more evenly by reconfiguring the
database layout or replace the problem disks with faster ones.

- bill

[ Post a follow-up to this message ]