RAID Gurus - Help!! Urgent!!

I have to tune an 1TB oracle 9i database on Linux that has 1 - 400GB hard disk for OS, system, temp, redo, undo and controlfiles and 3 - 400GB RAID 0 with one SCSI controller for all the other datafiles (tables, indexes etc.,).

I need some info on how the raid controller works when reading and writing to 3 disks in a RAID 0 configuration and how to best to distribute various oracle files in this configuration. For example, if I distribute control/undo/redo/temp between 1 Non-RAID OS disk and 3-RAID 0 disks, does that help or create more contention when writing data to datafiles?

Thanks.

sorry, but you cant do much with 3 disks

what are you trying to tune here exactly - what are the wait events in your database

RAID 0 ?? You don't have any redundancy for any of your data on your system? You really need to reconsider that configuration.

Quote:

---

Originally Posted by ebrian

RAID 0 ?? You don't have any redundancy for any of your data on your system? You really need to reconsider that configuration.

---

What to re-consider with 3 disks..?

If I were the OP, I would reconsider reconfiguring all the drives to RAID 1+0. Having the ability to recover my data after a media crash would be more important than the "potential" for better performance. Admittedly, it depends on the ultimate use of the system.

Doesn't matter how marginal the performance difference may be, the performance will be ZERO if a drive fails without some form of redundancy.

Nevertheless, a 1TB database on 4-drives is just asking for trouble. Drives are cheap !!! That's one thing that seems impossible to convince management of. In the long run, it would be cheaper to buy the hardware than it would to incur the labor costs and the inevitable data loss associated with restoring the system from a media failure.

As the saying goes, "Never enough money to do it right the FIRST time, but ALWAYS enough money to do it the SECOND, THIRD or FOURTH time."

A bit of background on why this config used - This database collects network sniffing info, which is not considered mission cricital. Hence they went for a closed box RAID 0 configuration. The customer is reponsible for any backups. All the datafiles on bunched up on 3 disk RAID 0. OS/redo/control/temp/undo and bunched together on 1 - non RAID disk. Here are the wait event for 7 million inserts:

Here are the top 5 wait events based on the statspack report:
Event Waits Time (s) Ela Time
------------------------------------- ------------ ----------- --
log file sync 44,342 21,530 52.61
log file parallel write 75,561 10,885 26.60
db file sequential read 4,609,459 2,982 7.29
db file parallel write 15,473 2,847 6.96
control file parallel write 7,220 802 1.96
----------------------------------------------------------------------

Can someone point any info on how a single SCSI raid controller works on a disk array? Are the disk reads/writes asychronous and parallel with just one controller? Or are they queued? Does it help if we have 3 controllers for 3 disk RAID?

The closed box can not accomodate more disks unless the whole box is changed.

look at where your waita are - 75% writing to redo logs - which ar on local disks

Quote:

---

... which is not considered mission cricital.

---

Then why do you worry about performance.

For 1 TB DB, I could not imagine 1 disk controller configuration.

As long as end users are not complaining about performance, no company would buy additional HW. This is my experience.

Some tablespaces are RAID and others on Non-RAID disks.
What is backup mechanism you have?

Tamil

Quote:

---

Originally Posted by tamilselvan

What is backup mechanism you have?
Tamil

---

anyone wanna place bets on exports?

How about upgrading to 10gR2 and using the asynchronous commit feature?

http://download-east.oracle.com/docs....htm#ADFNS1018

Thanks guys.

Tamil,

How does SCSI RAID controller controls reads and writes on a RAID 0 disk array?

Assuming the stripe size of 64K and an insert is trying to write to a datafile that is striped on all three disks and a query is trying to read from tables and indexes that are striped on all three disks, how does the RAID controller handles the requests? Does it permit asynchronous read and write on all the disks in parallel? Where can I get info on this?

Hmm....... asynchronous commit feature? Good for performance.. but....

Quote:

---

The obvious danger of asynchronous commits is that your DBMS is no longer ACID complient. Specifically, in using an asynchronous commit, you have traded durability for speed. An asynchronous commit returns before the data has actually been written to disk, so if the database crashes before some buffered redo is written to the logs, or a file I/O problem prevents the log writer from writing buffered redo to a non-multiplexed log, then the commit is lost.

---

http://orafaq.com/node/93

Also see the last part of the metalink doc: 336219.1.

Buy a useable storage array that meets your requirements for Availability, Performances and cost.

Just 4 disks is not enough to get even moderate availability/perfomance. See your own stats with High log sync wait event!