Oracle FAQ Your Portal to the Oracle Knowledge Grid
HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US
 

Home -> Community -> Mailing Lists -> Oracle-L -> RE: Which RAID will be better solution?

RE: Which RAID will be better solution?

From: Rosenthal, Jonas S. <JRosenth_at_oxhp.com>
Date: Sat, 24 Jun 2000 11:51:39 -0400
Message-Id: <10538.110362@fatcity.com>


It depends on how true the oltp really is with regard to the i/o for each transaction and the performance capability of your das.

You really should look at a sampling of users and see what kinds of transactions they are performing. The question becomes "what is the definition of their transaction" based on system tools, v$sesstat,v$sysstat, and v$sessio and tkprof (or any other monitoring tools like oem, etc...). And what is their think time or are they banging away constantly. Is each user going to be constantly active? Lots of questions. What is the growth potential for the user load? What is the capability of the DAS for response time vs. heavy load response time? Heavy load is better if the response time has little or no degradation vs. DAS which has great response time but slows with stress. Same can be said of the main system.

In general, For a hard load, going raw with raid0+1 is great depending on the manufacturer, if you have the bucks. Raid 5 if fine for light loads. I've seen raid 3 used but I have no personal experience with it. Redo's should generally be raw. But these are general statements. You need to experiment and get more info about your users stats before you commit to anyone's wishes.

Jonas

-----Original Message-----
From: Paul Drake [mailto:paled_at_home.com] Sent: Saturday, June 24, 2000 4:33 AM
To: Multiple recipients of list ORACLE-L Subject: Re: Which RAID will be better solution?

Ravi Babu wrote:
>
> Hi Friends,
> I want to do RAID configuration for a database which is high OLTP
> environment. 50-60 users are connected at a time. Performance is very
> important. How many disk arrays will be better?

MORE! 8) I saw a decent article on Dell's Website in the Power Solutions Magazine.
(yes, Hi - I deploy/maintain Oracle Servers on WinNT at remote Client Sites.
 Someday, I may be an Oracle on Unix bigot also ... but for now, NT)

http://www.dell.com/us/en/biz/topics/power_ps3q99-raid.htm

Both the Oracle 8i DBA Handbook and the Oracle 24 x 7 book (Oracle Press) have good discussions concerning optimal layouts. I believe that their dream system had 21 drives on independent controller channels -

	14 drives is where they started waking up ... 
	and 7 drives is where they started to get reasonable.

RAID 5 can bite you. It bit me 2 weeks ago. I'd avoid it except for read-only tablespaces. That leaves RAID 1 and RAID 10.
Sure, I run RAID 0 on my workstation, but I take incremental backups every night, full on weekends.

Performance will seem most important, until you try to recover the database.
Design it with fault-tolerance and recovery in mind, first.

Is this Standard Edition or Enterprise Edition? Version? Platform?

  1. Hot sparing
    • are you planning on just hot-swappable, or will you be hot-sparing?
    • if you are hot-sparing, each channel needs its own spare
  2. Backups to disk - will you be backing up to disk before copying off to tape?
  3. Backup strategy
    • will you be performing a full logical export before the cold backup each week?
    • will you be performing cold backups over the weekend?
    • will you be performing hot backups during the week?
  4. will read only tablespaces be used? Big reduction in backup space requirements.
  5. do you plan on duplexing the online redo logs?

A simple Intel based system is as follows:

Pair of 866 PIII CPUs
lots-O-RAM - say - 2 GB. when you pay by the power unit, 2 (triple or quad)-channel RAID controllers (need 4 external channels) Internal cabinet split backplane.
2 External storage cabinets with split backplanes. cabling
18 GB Ultra 160/m drives (good price point these days)

volume	RAID	drives	storage	(GB)	contains
0	1	2	17.4		OS, Binaries, SYSTEM, swap
1	1	2	17.4		online redo, archlogs
2	1	2	17.4		indexes, dumps
3	1	2	17.4		temp
4	1	2	17.4		rbs
5	10	6	52.2		user_data

If you're configuring with Parallel server in mind, I'd put the OS/binaries and log files in the internal cabinet. The second instance can mount the index and data files in the external cabinets.

64 bit, 66 MHz, multiple PCI bus mainboards are the way to go.

Remember, disk I/O is only part of it - front side bus means alot too - PC 133.
Quad-port Fast Ethernet or Gigabit Ethernet is also a factor. You wouldn't spend all that on disk I/O only to have 1 10/100 NIC, would you?

If you want to speed up RBS or Logs, set them up as RAID 10. Will you be using a write-back cache setting (battery on the controller card + UPS)?
Some people recommend the use of solid state drives for the online redo logs for speed.
If you do - duplex them.
18 GB drives were chosen just for minimizing the number of spares needed.
'Extra' room will serve well as staging areas for hot/cold backups to disk.

Most likely, you won't be using swap at all.

If you run the datafile_IO scripts - you'll have a better idea which tablespaces have the most read/write I/O. If you have to start condensing datafiles onto volumes, you'll be able to make an informed decision as to which to put together.

So there's a start.

Paul

-- 
Author: Paul Drake
  INET: paled_at_home.com

Fat City Network Services    -- (858) 538-5051  FAX: (858) 538-5051
San Diego, California        -- Public Internet access / Mailing Lists
--------------------------------------------------------------------
To REMOVE yourself from this mailing list, send an E-Mail message
to: ListGuru_at_fatcity.com (note EXACT spelling of 'ListGuru') and in
the message BODY, include a line containing: UNSUB ORACLE-L
Received on Sat Jun 24 2000 - 10:51:39 CDT

Original text of this message

HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US