Re: Second HBA to east io on E250?
From: Keg (rhugga_at_yahoo.com)
Date: 09/29/04
- Next message: Cathy Arima: "Article: Cross-Referencing HTML the Tiger Way"
- Previous message: jimleong: "Re: restore Unix system"
- In reply to: Ste Kearney: "Re: Second HBA to east io on E250?"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Date: 29 Sep 2004 12:30:41 -0700
Ste Kearney <stek1961@macSPAM.com> wrote in message news:<BD79EBE8.732E%stek1961@macSPAM.com>...
> On 24/9/04 3:15 pm, in article BD79EA77.732A%stek1961@macSPAM.com, "Ste
> Kearney" <stek1961@macSPAM.com> wrote:
>
> > We have one of these venerable old machines hosting an Oracle application
> > and a message queue on it's internal disks.
> >
> > It's suffering from delayed writes and the Oracle boys want to move the
> > message queue off the disks onto a bog simple flexpack with a 36gb disk on
> > the end of Sun PCI SCSI Controller (X6540A), leaving Oracle to churn away on
> > the internal disks....
> >
> > My thought is to move the MQ onto another machine entirely, thinking even a
> > specced up E250 (I don't know how specced up this one is yet) might struggle
> > with anything made by Oracle plus a message queue.
> >
> > Just looking for opinions based on this scant information! I'm still waiting
> > for copies of the iostat and vmstat logs etc.....
> >
> > Of course, this system is REALLY important but obviously not enough to
> > warrant an newer box!
>
> More info, Oracle boys want to add:
>
> 1 x SG-XDSK020C-72G Sun Flexipack
>
> 2 x 36GB HDD
>
> ! x X6540A Sun dual PCI card Ultra scsi controller
>
> 1 x X3832A 2M VHDCI to 68pin SCSI cable
>
> Steve
Well segregating the I/O onto more drive spindles will give you better
performance (unless your bus is saturated at which point you will need
a larger box or move some components onto another system as you
suggested)
I would first look at the basics such as what Raid level are the
current data volumes? If you using only the internal disks I would
guess you are using Raid 5 which you really want to avoid when using
Oracle) Also, is your Oracle I/O workload segregated to different
disks now? ie: You don't want index and data tablespaces on the same
physical disk(s). You want to also place undo and temp tablespaces on
seperate physical disk(s) as well. Those are common areas where
performance is greatly hindered. Also write your redo logs to another
physical disk if possible. (Think about everything that happens for a
single oracle transaction and you could be performing 5+ disk
reads/writes per transaciton: 1) data is written, redo log written,
archive logs are also being written/archived, temp space is being
used, undo space is being used. All this can be exhausintg if you
haven't segregated I/O onto multiple drive spindles) Also, stay away
from these large disk drives you see now days. We use only 18gb 15k
rpm disks for our Oracle databases. Yes you save space/power using
larger disks, but ths IOPS are very poor on these large disks.
Hope this helps,
cc
- Next message: Cathy Arima: "Article: Cross-Referencing HTML the Tiger Way"
- Previous message: jimleong: "Re: restore Unix system"
- In reply to: Ste Kearney: "Re: Second HBA to east io on E250?"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Relevant Pages
|
|
