Re: Storage array advice anyone?
- From: Martic Zoran <zoran_martic@xxxxxxxxx>
- To: mzito@xxxxxxxxxxx, JHostetter@xxxxxxxxxxxxxxxxxxxx
- Date: Fri, 17 Dec 2004 23:49:07 -0800 (PST)
Matt,
That is very good point.
But there is another view from Larry's idea that you
want to put as much applications as possible into the
centralized system (grid, RAC, whatever, ...).
In this case it is not only management, but also
performance issues as we all know mixing two or more
apps inside Oracle with apps having totally oposite
characteristics (SLA's, ..) are giving you a big
headache.
I mean the idea is great, cheap HW, cheaper SW,
possible cheap administration (even 10g RAC for me is
much more complex then 9i, having in mind the variety
of possibilities you can implement now RAC,
certification matrix is now very nice, you can use
this or that, with variety of cluster SW, ..).
For me this is very similar to the virtualization that
tend to save $/GB.
Our, or at least mine, main concern most of the time
is the I/O throughput in operations/second or MB/s
whatever higher.
That is what any application wants. Now mixing the
databases and other apps files under the
virtualization idea, spreading everything across
hypervolumes, LUN's, disks is not giving you the
possibility to say OK, these read/writes should have
average I/O in 90% cases.
I am not saying that the virtualization cannot be used
properly beraing in mind some apps needs big I/O
throughput during the night (OLAP) and some during the
day (OLTP) then you can mixed them under the
virtualization. Also the virtualization will make
sense in the same way as for Oracle if you have some
kind of smart logic implemented in the way of resource
managers driven by your rules.
This is my main question:
Does the storage HW recognized I/O requests coming
from the different apps and can nicely made numbers
needed to not ruin global SLA numbers?
Does any storage HW currently has somekind of rules
driven resource manager that is used in the practice,
where rules can be made from the apps level?
At the end, everything comes from the business logic
(user requests) then I assume if you cannot control
each piece of the SW/HW downthere you cannot get the
proper response.
I suppose this requires that apps or Oracle can send
app id or name to the storage device then storage HW
can understand who is sending and what to do with the
request.
Sorry if I am not aware of what is going on in the
storage industry.
This is probably my weakest area.
Thanks in advance.
Regards,
Zoran
--- Matthew Zito <mzito@xxxxxxxxxxx> wrote:
>
> There's a difference, though - the mainframe was
> massive hardware with
> massive redundancy implemented in hardware,
> resulting in a massive
> cost, but very low maintenance
>
> Larry's vision is one big database, one big
> application instance,
> running on hundreds of small commodity servers with
> redundancy
> implemented in software, resulting in a very low
> cost but very high
> maintenance.
>
> Into the gap steps companies like mine (and our
> competitors) that try
> to bridge the gap - aid in the management of large
> numbers of commodity
> servers.
>
> Thanks,
> Matt
>
> --
> Matthew Zito
> GridApp Systems
> Email: mzito@xxxxxxxxxxx
> Cell: 646-220-3551
> Phone: 212-358-8211 x 359
> http://www.gridapp.com
__________________________________________________
Do You Yahoo!?
Tired of spam? Yahoo! Mail has the best spam protection around
http://mail.yahoo.com
--
//www.freelists.org/webpage/oracle-l
Other related posts:
