9iR2: Windows Server 2003 vs Sun Solaris 9

[richyip]

Hi guys,

I did a performance testing between Windows Server 2003 running on my Dell desktop computer and Sun Solaris 9 on SunFire 12K. Note the differences in Oracle version. I insert records from a loop inside stored procedure.

Test Criteria:
- 200MB datafile
- Create a new user
- No other users connected

Windows Server 2003
-------------------
Oracle 9i 9.2.0.3.0
With OLAP and Oracle Data Mining options
P4-2.4GHz with 256MB RAM
Block size 8k
1 member each in 3 groups of redo logs

Insert 100,000 rows - 8.0s
Insert 1,000,000 rows - 74.0s


Sun Solaris 9(2Gbps connection to SAN)
-------------
Oracle 9i 9.2.0.1.0 64-bit
With Partitioning and Oracle Data Mining options
Sun Sparc 64-bit 4 X 900MHz with 4GB RAM
Block size 4k
3 members each in 3 groups of redo logs

Insert 100,000 rows - 12.2s
Insert 1,000,000 rows - 135s

Questions
---------
1. Why a desktop PC running Windows Server 2003 can beat SunFire 12K?
2. What are the performance bottleneck in SunFire 12K?
3. Is it the latest 9.2.0.3.0 is faster?

Pls give your comments. Tks.

[abhaysk]

I can see N reasons for this...

* P4 2.4GHz vs Sun Sparc 64-bit 4 X 900MHz -- I see P4 will side track 4 900MHz processors, if NOT all 4 processors are used concurrently.

* Network Speed if datafiles are remote.

* Server's handling of I/O rate.

* I see desktop PC has just 256MB RAM vs Sun's 4GB RAM...i belive you has not taken advantage of this factor.

* Block size 8K vs 4K...will not be affected much..but to some extent yes.

And many more...

Abhay.

[richyip]

Let me clarify:
- I have checked and 4 CPU were utilized in the insert.
- Datafiles are in the SAN storage so will be slower.
- The Domain I tested was not used, the other Domain is in production but low system load.
- all the SGA and PGA are very low in P4 but SGA and PGA are tuned in SunFire 12K.

Is 9.2.0.3 has better performance? I'm not sure.

[richyip]

However, I expect the SunFire 12K should be at least slightly faster as its running 64-bit OS and Oracle and tune to multi-users and multi-processing.

[abhaysk]

Originally posted by richyip
Let me clarify:
- I have checked and 4 CPU were utilized in the insert.

How did you check this, i mean how efficiently is it being used.
To My knowledge P4 processors are always more efficient then any other processors.

Originally posted by richyip
- Datafiles are in the SAN storage so will be slower.

You are not clear on this part.
Do you mean, when loading into SUN box the datafiles are remote...& for WIN 2000 PC datafile on host itself?

If this is true...you gotta the ans.

Originally posted by richyip
- The Domain I tested was not used, the other Domain is in production but low system load.

Again Not Clear ....

Originally posted by richyip
Is 9.2.0.3 has better performance? I'm not sure.

There will be hardly few changes in ver 9.2.0.1 and 9.2.0.3..
So not much influence..

Abhay.

[SANJAY_G]

I would say a benchmarking of no use if made on two disimilar hardware configuration, on two dissimilar storage configuration, on two dissimilar database patch version and two different database configuration.
If you are interested in benchmarking results, check this..
www.tpc.org

[julian]

Originally posted by richyip
Hi guys,

I did a performance testing between Windows Server 2003 running on my Dell desktop computer and Sun Solaris 9 on SunFire 12K. Note the differences in Oracle version. I insert records from a loop inside stored procedure.

Test Criteria:
- 200MB datafile
- Create a new user
- No other users connected

Windows Server 2003
-------------------
Oracle 9i 9.2.0.3.0
With OLAP and Oracle Data Mining options
P4-2.4GHz with 256MB RAM
Block size 8k
1 member each in 3 groups of redo logs

Insert 100,000 rows - 8.0s
Insert 1,000,000 rows - 74.0s


Sun Solaris 9(2Gbps connection to SAN)
-------------
Oracle 9i 9.2.0.1.0 64-bit
With Partitioning and Oracle Data Mining options
Sun Sparc 64-bit 4 X 900MHz with 4GB RAM
Block size 4k
3 members each in 3 groups of redo logs

Insert 100,000 rows - 12.2s
Insert 1,000,000 rows - 135s

Questions
---------
1. Why a desktop PC running Windows Server 2003 can beat SunFire 12K?
2. What are the performance bottleneck in SunFire 12K?
3. Is it the latest 9.2.0.3.0 is faster?

Pls give your comments. Tks.

Who knows how you have configured the OSs and Oracle ....

[richyip]

Both Oracle versions are installed mostly by default settings. And I would say the Sun 12K is slightly tuned than the Windows one. Yeah I know its different configurations on both machines, but I really don't expect such an expensive machine will lose out to a cheap Intel box. I'm still unsure of the performance bottleneck.

[marist89]

Originally posted by richyip
Let me clarify:
- I have checked and 4 CPU were utilized in the insert.

doubt it. It's unlikely that a single PL/SQL session would use all four processors.

- Datafiles are in the SAN storage so will be slower.

Than a single IDE disk? No way.

- The Domain I tested was not used, the other Domain is in production but low system load.
- all the SGA and PGA are very low in P4 but SGA and PGA are tuned in SunFire 12K.

Not relavent.

Is 9.2.0.3 has better performance? I'm not sure.

Sure, but not that much.

There's so many variables. Howevever, it's perfectly conceivable that a single user process runs faster on a 2.4 Ghz Intel processor than a 900Mhz Sparc. You're just comparing pure compute power at that point.

However, start running multiple users and your single CPU box will be overwhelmed quickly. If you're trying to benchmark database performance, benchmark it with actual user loads.

[richyip]

Yes definitely I agree with you. When handling multi-users, the Sun 12K will blow the Intel box away.

FYI, from the 'mpstat' I can see each of 4 processors idling was at 70-80% when running a single stored procedure. The load is transfered from each processor to another processors.