Oracle FAQ | Your Portal to the Oracle Knowledge Grid |
![]() |
![]() |
Home -> Community -> Mailing Lists -> Oracle-L -> Re: Optimal hard disk layout for Oracle
Caution: NT on Intel bias present - but I'm trying to outgrow it.
context: 4 CPUs max. and NO RAID 5.
Imagine for a moment that you have a fixed budget to devote toward an Oracle
Server.
This amount includes your Oracle licensing, OS, hardware, etc.
also assume that your site doesn't have a site license, but is stuck paying by
the *Power Unit*.
Enterprise Edition $100/MHz (list), think twice before salivating over 1 GHz
Xeons.
Standard Edition $15/MHz (list) - 4 x 700 MHz x $15/MHz = $42,000.
Figure on $500 per I/O channel for Dell, $1000 per I/O channel for Compaq. Figure on $500 per 18 GB drive for Dell, $1000 18 GB drive for Compaq.Yes, its likely that you won't even need 25 % of the space on these drives. Storage is relatively cheap, Power Units are relatively expensive.
Strategy 1. Hold the Wintel Line - MHz is king, single I/O channel.
throw as much Mhz and RAM at is as possible.
Single Fast Ethernet Card, Network tape drive. No recovery accelerators
installed.
Vendors love this, as they make a premium on 4 and 8 way CPU boxes.
Companies like it, as need for proper tuning can be ignored - paper DBA can
administer system.
Single RAID 5 array, based only on size of storage needed. Write back cache, no
UPS.
Recovery time not an issue, as this company hasn't had a recent crash.
One can only tell the system is I/O bound if you know enough to look.
Since Oracle is billing by the MHz, this is good business.
"We don't need no stinkin' OFA."
Never tested backup/recovery strategy, using logical exports, fortunately.
Performing nightly backup of an open database, the Cluetrain hasn't stopped here
yet.
Strategy 2. Read the OFA doc - desires minimal seek time - higher concurrency.
deploying box that can only store 10 internal drives, e.g. Dell PowerEdge 4400
Box has 3 I/O channels - removable bay cage (2 drives), 8 drive cage split in
half, 2 PCI bus channels.
decided on 2 RAID 1 vols, rest JBOD.
Has OS/Oracle binaries/system on 1 RAID 1 vol
online redo logs on another RAID 1 vol (also destination for logical exports)
leaves 6 drives for OFA - without RAID.
Has non-paper DBA on staff, doesn't fear performing recovery when one of the
non-RAID drives dies.
External LVD tape drive (Ecrix VXA-1 33/66 is my current favorite)
Has actually tested recovery strategy, DBA reads this list. 8^)
Strategy 3. RAID on all vols - relying on Ultra 160/m 10,000 RPM drives for
performance
configured for maximum write and read rate, concurrency not as much an issue.
Dual Fast Ethernet NICs configured for Load balance, fail-over.
again, Dell PowerEdge 4400, 10 internal drives, 3 I/O channels - 64 MB of cache
per channel
vol drives RAID
0 2 1 OS, Oracle Binaries, System, online redo 1 4 0+1 IDX, TEMP 2 4 0+1 USR, RBSno hot spares, but has 2 cold spare drives sitting in a box, locked up. no DBA on site - box remotely monitored by 3rd party vendor.
Strategy 4 - discovered external cabinets (2). all channels RAID + hot spared.
Moved up to 4 CPU box. Gigabit Ethernet. 3 PCI bus channels, 64 bit, 66 MHz.
imagine that you might implement parallel server someday.
Design the boxes such that you only have instance specific files in the internal
cabinet
6 internal drives: (3 internal I/O channels used - 3 Quad channel RAID
controllers)
I/O vol drives RAID
internal
0 0 3 1 + hot spare OS, Oracle, arch logs1 1 1 3 1 + hot spare online redo 2 2 2 1 arch logs 2 external 3 3 2 1 System, exports 4 2 1 RBS_Small, RBS_Big 1 + hot spare 4 5 5 0+1 + hot spare Temp1,Temp2 5 6 5 0+1 + hot spare Indx1, User2 6 7 5 0+1 + hot spare User1, Indx2 28 drives - including 6 hot spares. Yum.
Has n+1 power supplies, spare RAID controller, NIC on site.
Still, it doesn't have a dedicated volume/I/O channel for storing a disk copy of
backups.
There was a white paper that had that configuration, complete with 4 tape drives
Received on Sat Sep 02 2000 - 01:06:41 CDT
![]() |
![]() |