Thanks to the diligence of my coworker Scott Maziarz we have been able to run some practical performance tests on the new Sun Fire T2000 we got on a try and buy program from Sun.
About our environment
While we’re not truly equipped for traditional load testing here at Plymouth State University I decided early on that our daily datamart scripts would be a good test of this system’s capabilities. These scripts are a combination of vendor supplied and homegrown code which create datamarts for reporting. The scripts run separately take between a minute and several hours to complete. They rely heavily on joins and functions so, while there is a fair amount of disk I/O, the speed of the processors and memory should play a large role.
Oracle 9iR2 was used for the testing on both systems. The database configuration was not altered, so despite the point that the T2000 has twice as much memory as our V440s (16GB versus 8GB) Oracle will be using the same amount of memory on both systems. We also unfortunately did not have the opportunity to do any performance tuning on the T2000. As such I consider this a comparison with a straight out of the box T2000. With some careful system tuning the T2000 would probably perform even better.
The two systems
Sun Fire T2000
- 8 core 1.0GHz UltraSparc T1Processor
- 16GB memory
- 2 * 73GB 10K RPM SAS hard disk drives
Sun Fire V440
- 4 * 1.0GHz UltraSparc IIIi
- 8GB memory
- 4 * 73GB 10K RPM Ultra320 SCSI hard disk drives
While these two systems are different in many ways they are comparable in price (the V440s are actually quite a bit more expensive, but this one is around two years old now.) All testing was done on internal disks making the results dependant on the entire system performance.
Oracle was not reconfigured to take advantage of the extra memory in the T2000.
The tests
For the testing we chose a set of 19 datamart creation scripts which we run daily in our production environment. To assure that there is no additional load on the V440 we ran the tests on our reporting instance which is on a relatively idle system.
The first test is to run the 19 datamart scripts staggered. This is how we run them in production. Four scripts are started every half hour to spread out the load on the server. There’s a lot of overlap, but the staggering is enough to keep the system responding normally.
The graph above represents the runtime of each job individually. A shorter line represents a quicker runtime and we can see that the T2000 consistently outperformed the V440 often running n a quarter the time!
The total runtime for all 19 jobs was 2915 minutes on the V440 and 847 on the T2000. On average it took only 29% as long for the jobs to run on the T2000!
For the second test we ran all 19 jobs at once.
We still see a significant improvement in performance, but not as high as when these were run staggered. Here the T2000 completed the tasks in 59% of the time of the V440. I attribute this to contention for disk on the T2000.
Thanks to Scott Maziarz for running the bulk of these tests and compiling the results for me.
Conclusions
The Sun Fire T2000 has certainly proven its worth. Some may be put off by the relatively low processor speed (the model tested was a mere 1GHz) but it is clearly not an impediment. The 8 core CPU seems to be up to the challenge and I’m sure with additional tuning I’m sure they’d scream.
With increased performance and higher efficiency than the comparably priced V440, the T2000 will definitely be in our future. The lower energy consumption and lower heat output would be a welcome change in our already taxed data center and the compact 2U size should be an easy fit in any rack.
Check out more details at Sun’s website and if you’re still not convinced apply for your own free 60-day trial!
It looks like the first graph is wrong–it looks just like the second graph.
You’re right Brian. I got the images messed up in the WordPress upload window. I’ve corrected the graphs. The numbers are correct.
Thank you Brian!
Jon
John H. said:
Good points John! To address these questions I have added the “The two systems” section.
Hi, We have installed Oracle 10gR1 on Solaris10@T2000. Found that thhe read and write intensive process takes considerably too much time despite the super fast HD. Any clues ?? Also in the console window we see 16CPU vs 4Cores.?? Will appreciate any tips on Oracle setup under t2000.
Can you please share the customization values in /etc/system for both systems that you tested?
We actually did not apply any custom /etc/system values. Unfortunately we ran out of time in our trial and had to work on a relatively untuned system.
M Irfan,
The system will use 4 threads per core but top and other system utilities will treat each thread like a CPU.
As for the performance problems, make sure you have applied all the appropriate patches to 10gR1 or go directly to 10gR2. I have heard lots of horror stories about optimizer problems in 10gR1.
Beyond that, be careful of how you have the disk configured. If you’re doing a software mirror or (worse yet) software RAID you will probably kill all disk performance.
Hope this helps. Sorry I can’t be more specific, but you know, every install is different.
Did you make any changes to the Oracle setup (parameters)? We are testing a 9i installation using 11.5.10 apps on both SparcIII systems and a T2000. Users are complaining of slow performance on the T2000 instance. We’ve gone through tuning documents for the T2000 but I wonder about tuning the Oracle database/apps to utilize the threads technology.
Thankx.
Kurt,
We didn’t tune the instance at all. We wanted it to be configured the same on both systems. As for taking advantage of the threads, as long as the database will run at least as many processes as you have threads (4 x number of cores) the operating system should take care of the rest.
I’d bug your Sun reps if you can’t get the system to perform. They offered to help me at every turn when I had the demo.
Sorry I can’t be of more help.
We are looking at using T1000 or T2000’s for the apps tier and maybe the database tier for 11.5.10. Oracle Applications.
I believe Kurt has been having performance issues using 11.5.10 apps with a T2000, did you resolve the issue in the end Kurt? Did the T2000 beat the Sparc iii based servers and what were the server specs?
Any information gratefully received.
Our lab has several new V440 servers, purchased from our sun reseller MCA, http://www.mcac.com and they have also loaned us a T2000 to benchmark. We plan to run several test scripts, that we have written, that will demonstrate the performance & comparison of these two models.
I will provide our test results & set-up shortly.
I have recently set up Oracle on T2000. I am gettign slow performance complaint. Database import (130 GB) takes about 45 hrs on this box. On E3500 server, the same takes about 30 hrs. Could any one help me on tuning this . Thanks a bunch !!
Baburajan,
Since import/export is very read/write intensive I’d say the difference could be the disk. Are you seing lots of i/o wait in top?
We are looking at using T1000 or T2000’s for the apps tier and maybe the database tier for 11.5.10. Oracle Applications.
I believe Kurt has been having performance issues using 11.5.10 apps with a T2000, did you resolve the issue in the end Kurt? Did the T2000 beat the Sparc iii based servers and what were the server specs?
Any information gratefully received.
T2000 is a throughput based machine. If application has too much serial like a single threaded batch job, then performance will not match traditional architectures and may run longer.Such as export&import&alter table move etc. Would be a little slower on t2000….
Hi,
Anybody out there running a sizable ClearCase Multisite
environment on Sun Fire T2000 Servers? I’d like to know the comparision in performance against, say, the v880 or v440.
Please let me know.. thanks in advance !
We are using the Banner system from SunGard Higher Education with an Oracle database. Has anyone had experience with the Banner Oracle database on a Sun T2000 server? Does it perform satisfactorily? The pricing of the T2000 is attractive, but performance is more important.
Scott,
First, in the interest of full disclosure, I am currently working for SunGard Higher Education, although I was not when I originally wrote this article.
I do not know of any institutions who are using T2000s for Banner databases but my opinion is that it should work fine for small to medium schools. Note that this is just my opinion, and not that of SGHE (I’m not typically involved in this type of capacity planning.)
If you were to go with a T2000 you will obviously want to use some type of external disk array for your data. The T2000 drives are limited in capacity (they are the size of laptop drives) but are sufficient as system disks.
Sorry I can’t give you more of a confirmation on this. You may consider putting in a support ticket or polling other schools through one of the client mailing lists.
Best of luck.
Hi Jon,
Have you also tested Oracle 10g on the T1000? I am trying hard on Google to find user experiences. Sun on their website recommends T1000 for web applications and not as a database server but I do not see significant changes in the specs between the T1000 and the entry level T2000 servers. This post is quite old. What’s the status now wrt Oracle performance on these servers? Any new experiences? Unfortunately in India I cannot avail the Try/Buy offer from Sun.
Aniruddha,
Unfortunately I haven’t worked on the T2000 in a while and didn’t have the chance to evaluate 10g on it. You’re right that Sun pushes them as an application server, but the multiple cores and plenty of RAM make them solid for Oracle databases too. I’d love to see how they’d perform in a RAC configuration but unfortunately don’t have the resources or time.
Aniruddha,
try starting from here
http://www.sun.com/bigadmin/content/submitted/real_world_tests.jsp
although a bit outdated.
Regds