Tuesday Apr 23, 2013

Solaris 11 outperforms RHEL 6 on 2 socket Intel servers

As a long time Sun employee, I've often heard the term "Slow-laris" applied to Oracle's premier Unix operating system.  Most frequently this was in comparison to the Linux OS running on small two socket servers.  I will admit that in the Solaris 8 and 9 timeframe engineering decisions were made to benefit scalability to 64 sockets that sometimes penalized smaller servers.  In addition, because of Solaris long history and derivation from ATT and BSD Unix code, there was undoubtedly a bit of code labeled, "if it ain't broke, don't fix it."  With the advent of Solaris 10 and Dynamic Tracing, (DTrace) we actually hunted down and killed a number of those legacy code segments using a new philosophy labeled internally, "If Solaris is slower than Linux on the same hardware, it's a bug."

As a result, Solaris 11 provides higher performance than Red Hat Enterprise Linux 6.3 on basically identical 2 socket hardware as measured by the SPECjbb benchmark.  According to SPEC:

The SPECjbb2013 benchmark has been developed from the ground up to measure performance based on the latest Java application features. It is relevant to all audiences who are interested in Java server performance, including JVM vendors, hardware developers, Java application developers, researchers and members of the academic community.

Java is one of the predominant enterprise programming environments for mission critical applications and many of Oracle's products are written in Java.

This chart from the SPECjbb site shows the performance of our X3-2 Intel based server with 16 cores and 128 GB of RAM running Solaris 11.1.  The X3-2 tested features the Intel E5-2690 CPU @ 2.9 Ghz.

X3-2 Chart

By comparison, an HP ML350P with the identical Intel chip and clock speed running RHEL 6.3 produces this chart.  Clearly, Solaris 11 produce a smoother response curve with higher numbers for both MaxjOPS and Critical jOPS.  In addition, the X3-2 system requires only 1 rack unit vs. 4 rack units for the HP model reducing data center requirements. 

HP Chart

 To summarize, Solaris is faster than RHEL 6 on small servers and more scalable and responsive on large servers including our SPARC T5 servers.

At the same time, it provides virtualization, security and availability features unavailable on RHEL including:

  • Solaris zones
  • Network virtualization
  • ZFS file system
  • Dynamic Tracing
  • Predictive self-healing
  • Service Management Facility
  • Trusted Extensions 
  • Image packaging system

See more at:

  • Jeff Victor's blog
  • Oracle's Performance Blog
  • SPEC and the benchmark name SPECjbb are registered trademarks of Standard Performance Evaluation Corporation (SPEC). Results as of 4/22/2013, see http://www.spec.org for more information.
  • SPARC T5-2 75,658 SPECjbb2013-MultiJVM max-jOPS, 23,334 SPECjbb2013-MultiJVM critical-jOPS. Sun Server X2-4 65,211 SPECjbb2013-MultiJVM max-jOPS, 22,057 SPECjbb2013-MultiJVM critical-jOPS. Sun Server X3-2 41,954 SPECjbb2013-MultiJVM max-jOPS, 13,305 SPECjbb2013-MultiJVM critical-jOPS. SPARC T4-2 34,804 SPECjbb2013-MultiJVM max-jOPS, 10,101 SPECjbb2013-MultiJVM critical-jOPS. HP ProLiant DL560p Gen8 66,007 SPECjbb2013-MultiJVM max-jOPS, 16,577 SPECjbb2013-MultiJVM critical-jOPS. HP ProLiant ML350p Gen8 40,047 SPECjbb2013-MultiJVM max-jOPS, 12,308 SPECjbb2013-MultiJVM critical-jOPS. Supermicro X8DTN+ 20,977 SPECjbb2013-MultiJVM max-jOPS, 6,188 SPECjbb2013-MultiJVM critical-jOPS. HP ProLiant ML310e Gen8 12,315 SPECjbb2013-MultiJVM max-jOPS, 2,908 SPECjbb2013-MultiJVM critical-jOPS. Intel R1304BT 6,198 SPECjbb2013-MultiJVM max-jOPS, 1,722 SPECjbb2013-MultiJVM critical-jOPS.


Solaris 11 provides smooth, scalable performance on SPECjbb 2013

Oracle released SPEC Benchmark results for the T5-2 and X2-4 processor using the SPECjbb 2013 benchmark. Who would be interested in SPECjbb performance? According to SPEC:

The SPECjbb2013 benchmark has been developed from the ground up to measure performance based on the latest Java application features. It is relevant to all audiences who are interested in Java server performance, including JVM vendors, hardware developers, Java application developers, researchers and members of the academic community. 

Jeff Victor has posted an excellent comparison of the T5 SPECjbb performance to our competitors on a per core basis.  To me, the charts tell the biggest part of the story,  Oracle's Solaris 11 on both SPARC and X86 shows smooth scaling with excellent response times over a wide range of transaction counts.

First, let's look at the results for the SPARC T5-2 server with 2 CPU sockets and 32 cores.  The vertical access marks "response time" so a lower number is better.  The horizontal axis is the number of Java operations being performed.  The blue dots indicate the median response time at each level of operations being processed.  Notice how Solaris 11 and the SPARC hardware provide smooth, predictable performance up through 60,000 jOPS.

(Note: You may not be able to see the full chart width on this page.  Right-click and open image in new tab to see the full chart.) 

T5-2 Chart

 Now let's look at Oracle's X2-4 Intel based system also running Solaris 11.  The X2-4 has 4 CPU chips with 40 total cores.  Here Solaris 11 also provides smooth scaling of performance.

X2-4 chart

For comparison, I've also selected HP's most powerful Intel based server the DL980 with 8 CPUs and 80 cores.  This system, however is running Red Hat Enterprise Linux 6.3.  On this chart you will see that RHEL 6 takes a dive in median response time shortly after 27,000 jOPS. Response time drops from 10 milliseconds to 100 milliseconds at around 27,000 jOPS.  Oracle's T5-2 stays below 100 milliseconds all the way to about 62,000 jOPS. Also note how the minimum response times fall apart at around 20,000 jOPS where the T5-2 stays consistent through 57,000 jOPS.

While admittedly, the 80 core DL980 reaches a higher total MaxjOPS throughput number than the 32 core T5-2, the Solaris 11 based system provides smoother scalability in a 2 socket system that requires only three rack units of space.  If that's not enough horsepower, we also offer a T5-4 and T5-8 system.  Need more?  Our M5-32 data center server scales to 32 sockets, 192 cores and 1536 threads. The M5-32 also supports up to 32 TB of RAM. All support our no cost Logical Domains virtualization capability.

HP DL980 Chart

Summary:

 If you want a proven, enterprise class, scalable OS for SPARC (from Oracle or Fujitsu) or X86 based platforms (from Oracle or many third party vendors), choose Solaris 11.  Predictability in response time is important to your enterprise customers.

All Oracle servers under Premier Support for systems include:

  • 7 x 24 on-site hardware support
  • Solaris (SPARC or X86), Oracle Linux (x86 only) and Oracle VM support (SPARC or X86)
  • Integrated Lights out Management
  • Oracle Enterprise Manager Ops Center support 

For more information on recent SPARC T5 world records, see https://blogs.oracle.com/BestPerf/.

  • SPEC and the benchmark name SPECjbb are registered trademarks of Standard Performance Evaluation Corporation (SPEC). Results as of 4/22/2013, see http://www.spec.org for more information.
  • SPARC T5-2 75,658 SPECjbb2013-MultiJVM max-jOPS, 23,334 SPECjbb2013-MultiJVM critical-jOPS. Sun Server X2-4 65,211 SPECjbb2013-MultiJVM max-jOPS, 22,057 SPECjbb2013-MultiJVM critical-jOPS. Sun Server X3-2 41,954 SPECjbb2013-MultiJVM max-jOPS, 13,305 SPECjbb2013-MultiJVM critical-jOPS. SPARC T4-2 34,804 SPECjbb2013-MultiJVM max-jOPS, 10,101 SPECjbb2013-MultiJVM critical-jOPS. HP ProLiant DL560p Gen8 66,007 SPECjbb2013-MultiJVM max-jOPS, 16,577 SPECjbb2013-MultiJVM critical-jOPS. HP ProLiant ML350p Gen8 40,047 SPECjbb2013-MultiJVM max-jOPS, 12,308 SPECjbb2013-MultiJVM critical-jOPS. Supermicro X8DTN+ 20,977 SPECjbb2013-MultiJVM max-jOPS, 6,188 SPECjbb2013-MultiJVM critical-jOPS. HP ProLiant ML310e Gen8 12,315 SPECjbb2013-MultiJVM max-jOPS, 2,908 SPECjbb2013-MultiJVM critical-jOPS. Intel R1304BT 6,198 SPECjbb2013-MultiJVM max-jOPS, 1,722 SPECjbb2013-MultiJVM critical-jOPS.

About

Jim Laurent is an Oracle Sales consultant based in Reston, Virginia. He supports US DoD customers as part of the North American Public Sector hardware organization. With over 17 years experience at Sun and Oracle, he specializes in Solaris and server technologies. Prior to Oracle, Jim worked 11 years for Gould Computer Systems (later known as Encore).

Search

Archives
« April 2013 »
SunMonTueWedThuFriSat
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
16
17
18
19
20
21
22
25
26
27
28
29
30
    
       
Today