BM Seer Unofficial thoughts from an anonymous Sun employee

Performance per core is very misleading & more on SPECjAppServer World Record

Friday Jul 20, 2007

SPARC is leading the world: World Record UltraSPARC T1 (Niagara) and UltraSPARC IV+ on SPECjAppServer. I had all of the details in yesterday's posting. But I quickly skimmed over a couple of important details that I'll talk a bit more about in this posting.

To keep all of the application servers fully supported one requires a fast database server and efficient database software. For this benchmark Sun used IBM DB2 9.1 database. In the past Sun has set many world records with this database software on a variety of benchmarks.

Another thing of note is that Sun only used 24 dual-core UltraSPARC IV+ chips in comparison to 32 dual-core Itanium2 chips. That means a lot less SPARC chips than Itanium chips and Sun gives more performance.

Also note I only use chips and cores to denote the configuration. Performance per core is very misleading in every comparison between vendors.

In the application tier sun used ten Sun Blade T6300 Server Modules (10 US T1 chips total) to get 8% better performance than HP's 6 rx6600 servers (24 Itanium2 chips total). Sun fits 10 blades in 10RU versus HP's 42RU (6x Four dual-core Itanic2 in 7RU), Sun uses a lot less hardware and a FOURTH less space!

See how totally misleading if we compared cores thinking cores were equivalent. Sun has 80 cores in the appl tier in this config versus HP's 48 cores in the application tier. Price/core is very different in EVERY vendor. Performance per core a smoke and mirrors marketing trick that should not be used by any honorable vendor.

World Record UltraSPARC T1 SPECjAppServer2004 benchmark. The Sun Blade 6000 systems extremely flexible you can have SPARC, AMD Opteron, and Intel Xeon processor-based server modules, as well as the Solaris, Linux, and Windows operating systems. Wow!

Ten Sun Blade T6300 Server Modules delivered a World Record result of 8253.21 SPECjAppServer2004 JOPS@Standard, this result use (1x UltraSPARC T1 1.4GHz) and connected to a Sun Fire E6900 Database System (24x UltraSPARC IV+ 1.95 GHz).

For more info and yesterday's posting see http://blogs.sun.com/bmseer/entry/world_record_ultrasparc_t1_specjappserver2004

Disclosure Statement:

SPEC, SPECjAppServer reg tm of Standard Performance Evaluation Corporation. Results from www.spec.org as of 07/18/2007. SPECjAppServer2004 10 Sun Fire T6300 (80 cores, 10 chips) and 1 Sun Fire E6900 (48 cores, 24 chips) 8253.21 SPECjAppServer2004 JOPS@Standard. SPECjAppServer2004 6 HP rx6600 (48 cores, 24 chips) and HP 9000 Superdome (64 cores, 32 chips) 7629.45 SPECjAppServer2004 JOPS@Standard. SPECjAppServer2004 10 Sun Fire X8420 (80 cores, 40 chips) and 1 Sun Fire E6900 (48 cores, 24 chips) 7174.56 SPECjAppServer2004 JOPS@Standard. SPECjAppServer2004 11 HP rx3600 (44 cores, 22 chips) and HP 9000 Superdome (64 cores, 32 chips) 6812.79 SPECjAppServer2004 JOPS@Standard. SPECjAppServer2004 20 IBM xSeries BladeCenter HS20 (40 cores, 40 chips) and IBM eServer p5 570 (16 cores, 8 chips) 4368.02 SPECjAppServer2004 JOPS@Standard.

Certified Results 8253.21 SPECjAppServer2004 JOPS@Standard
Reference Date: Jul 18, 2007
Systems: 10 x Sun Blade T6300 Server Modules
1 x Sun Fire E6900, 96GB
4 x Sun StorageTek SE3510 FC Array
Processor/GHz of Server: 10 x UltraSPARC T1 1.4 GHz
24 x UltraSPARC IV+ 1.95 GHz
Operating System: Solaris 10 8/07, Solaris Containers
Software: BEA WebLogic 10.0 Advantage Edition
IBM DB2 9.1 Enterprise Edition
JVM: J2SE 5.0 update 12

Like this post? del.icio.us | furl | slashdot | technorati | digg
Comments:

Post a Comment:
Comments are closed for this entry.