Consolidation of 30 x86 Servers onto One SPARC T3-2
By Brian Whitney on Sep 30, 2010
One of Oracle's SPARC T3-2 servers was able to consolidate the database workloads off of thirty older x86 servers in a secure virtualized environment.
The thirty x86 servers required 6.7 times more power than the consolidated workload on the SPARC T3-2 server.
The x86 configuration used 10 times the rack space than the consolidated workload did on the SPARC T3-2 server.
In addition to power & space considerations, there are also administrative cost savings resulting from having to manage just one server, as opposed to thirty servers.
Gartner says, "They need to realize that removing a single x86 server from a data center will result in savings of more than $400 a year in energy costs alone".
The total transaction throughput for the SPARC T3 server (132,000) was almost the same as the aggregate throughput achieved by the thirty x86 servers (138,000), where each x86 running at 10% utilization.
The average transaction response time on the SPARC T3-2 server (24 ms) was just a little higher than the average transaction response time on the Intel servers (19.5 ms).
|Sun Fire X4250
2x 3.0GHz Xeon
1x 1.65GHz SPARC T3
\* power consumption includes storage and periperal devices
total throughput for 30 Intel systems = 30 \* 4600 = 138,000
total watts for 30 Intel systems = 30 \* 320 = 9600
Results and Configuration Summary
x86 Server Configuration:
16 GB memory
6 x internal 146 GB 15K SAS disks
RedHat Linux 5.3
Oracle Database 11g Release 2
SPARC T3 Server Configuration:
256 GB memory
2 X 10K 300 GB internal SAS disks
1 x Sun Storage F5100 Flash Array storage
1 x Sun Fires X4270 server as COMSTAR target
Oracle Solaris 10 9/10
Oracle Database 11g Release 2
This demonstration was designed to show the benefits of virtualization when upgrading from older X86 systems to one of Oracle's T-series servers. A 30:1 consolidation was shown moving from thirty X86 Linux servers to a single T-Series server running Oracle Solaris in a secure virtualized environment. After the consolidation, there was still 20% headroom in the SPARC T3-2 server for additional growth in the workload.
The 200 scale iGen OLTP workload was used to test the consolidation. The x86 system was loaded with iGen clients up to a level of 10% cpu utilization. This load level for x86 systems is typically found in many data centers.Thirty Oracle Solaris zones (containers) were created on the SPARC T3-2 server, with each zone configured identically as the Oracle configuration on the x86 server. The throughput on each zone was ramped up to the same level as on the Intel base server.
The overall CPU utilization on the SPARC T3-2 server, together with the average iGen transaction response times were then measured along with the power consumption.
Key Points and Best Practices
Each Oracle Solaris container was assigned to a processor set consisting of eight virtual CPUs. This use of processor sets was critical to obtaining the reported performance number. Without processor set, the performance was reduced to about one-half the reported performance number.
Once the first container was completely configured (with Oracle 11g and iGen installed), the remaining containers were created by a simple cloning procedure, which took no more than a few minutes for each container.
Setting up a standalone x86 server with Linux, Oracle and iGen is a far
more time consuming task than setting up additional containers once the
first container has been created.
- Oracle Solaris Virtualization
- Optimizing Single Instance Oracle Databases on Oracle's T-series Servers (pdf)
- Dynamic SGA Tuning of Oracle Database on Oracle Solaris with DISM (pdf)
- Oracle Database on Sun SPARC Enterprise T-series Systems (pdf)
- Deploying Oracle Database on the Oracle Solaris Platform – An Introduction (pdf)
- Virtualization Options for Oracle Database Deployments on Sun SPARC Enterprise T-series Systems (pdf)
Copyright 2010, Oracle and/or its affiliates. All rights reserved. Oracle and Java are registered trademarks of Oracle and/or its affiliates. Other names may be trademarks of their respective owners. Results as of 9/20/2010.