Sun SPARC Enterprise M9000 Server Delivers World Record Non-Clustered TPC-H @3000GB Performance
By Brian Whitney on Oct 11, 2010
Oracle's Sun SPARC Enterprise M9000 server delivered a single-system TPC-H 3000GB world record performance. The Sun SPARC Enterprise M9000 server, running Oracle Database 11g Release 2 on the Oracle Solaris operating system proves the power of Oracle's integrated solution.
Oracle beats IBM Power with better performance and price/performance (3 Year TCO). This shows that Oracle's focus on integrated system design provides more customer value than IBM's focus on "per core performance"!
The Sun SPARC Enterprise M9000 server is 27% faster than the IBM Power 595.
The Sun SPARC Enterprise M9000 server is 22% faster than the HP ProLiant DL980 G7.
The Sun SPARC Enterprise M9000 server is 26% lower than the IBM Power 595 for price/performance.
The Sun SPARC Enterprise M9000 server is 2.7 times faster than the IBM Power 595 for data loading.
The Sun SPARC Enterprise M9000 server is 2.3 times faster than the HP ProLiant DL980 for data loading.
The Sun SPARC Enterprise M9000 server is 2.6 times faster than the IBM p595 for Refresh Function.
The Sun SPARC Enterprise M9000 server is 3 times faster than the HP ProLiant DL980 for Refresh Function.
Oracle used Storage Redundancy Level 3 as defined by the TPC-H 2.12.0 specification, which is the highest level. IBM is the only other vendor to secure the storage to this level.
One should focus on the performance of the complete hardware and software stack since server implementation details such as the number of cores or the number of threads will obscure the important metrics of delivered system performance and system price/performance.
The Sun SPARC Enterprise M9000 server configured with SPARC VII processors, Sun Storage 6180 arrays, and running Oracle Solaris 10 operating system combined with Oracle Database 11g Release 2 achieved World Record TPC-H performance of 198,907.5 QphH@3000GB for non-clustered systems.
The Sun SPARC Enterprise M9000 server is over three times faster than the HP Itanium2 Superdome.
The Sun Storage 6180 array configuration (a total of 16 6180 arrays) in this benchmark delivered IO performance of over 21 GB/sec Sequential Read performance as measured by the vdbench tool.
This TPC-H result demonstrates that the Sun SPARC Enterprise M9000 server can handle the increasingly large databases required of DSS systems. The server delivered more than 18 GB/sec of real IO throughput as measured by the Oracle Database 11g Release 2 software.
Both Oracle and IBM had the same level of hardware discounting as allowed by TPC rules to provide a effective comparison of price/performance.
IBM has not shown any delivered I/O performance results for the high-end IBM POWER7 systems. In addition, they have not delivered any commercial benchmarks (TPC-C, TPC-H, etc.) which have heavy I/O demands.
TPC-H @3000GB, Non-Clustered Systems
|Sun SPARC Enterprise M9000
2.88GHz SPARC64 VII
|HP ProLiant DL980 G7
2.27GHz Intel Xeon X7560
|IBM Power 595
|Unisys ES7000 7600R
2.6GHz Intel Xeon
|HP Integrity Superdome
1.6GHz Intel Itanium
QphH = the Composite Metric (bigger is better)
$/QphH = the Price/Performance metric (smaller is better)
QppH = the Power Numerical Quantity
QthH = the Throughput Numerical Quantity
Complete benchmark results found at the TPC benchmark website http://www.tpc.org.
Configuration Summary and Results
512 GB memory
4 x internal SAS (4 x 300 GB)
Database: Oracle Database 11g Release 2 Enterprise Edition
|Database Size:||3000 GB (Scale Factor 3000)|
|TPC-H Composite:||198,907.5 QphH@3000GB|
|Total 3 year Cost:||$3,037,900|
|Database Load Time:||3:40:11|
Benchmark DescriptionThe TPC-H benchmark is a performance benchmark established by the Transaction Processing Council (TPC) to demonstrate Data Warehousing/Decision Support Systems (DSS). TPC-H measurements are produced for customers to evaluate the performance of various DSS systems. These queries and updates are executed against a standard database under controlled conditions. Performance projections and comparisons between different TPC-H Database sizes (100GB, 300GB, 1000GB, 3000GB and 10000GB) are not allowed by the TPC.
TPC-H is a data warehousing-oriented, non-industry-specific benchmark that consists of a large number of complex queries typical of decision support applications. It also includes some insert and delete activity that is intended to simulate loading and purging data from a warehouse. TPC-H measures the combined performance of a particular database manager on a specific computer system.
The main performance metric reported by TPC-H is called the TPC-H Composite Query-per-Hour Performance Metric (QphH@SF, where SF is the number of GB of raw data, referred to as the scale factor). QphH@SF is intended to summarize the ability of the system to process queries in both single and multi user modes. The benchmark requires reporting of price/performance, which is the ratio of QphH to total HW/SW cost plus 3 years maintenance.
Key Points and Best Practices
- The Sun Storage 6180 array showed good scalability and these sixteen 6180 arrays showed over 21 GB/sec Sequential Read performance as measured by the vdbench tool.
- Oracle Solaris 10 10/09 required little system tuning.
- The optimal 6180 configuration for the benchmark was to set up 1
disk per volume instead of multiple disks per volume and
let Oracle Solaris
Volume Manager (SVM) mirror. Presenting as many volumes as possible to
Oracle database gave the highest scan rate.
- The storage was managed by SVM with 1MB stripe size to match
with Oracle's database
IO size. The default 16K stripe size is just too small
for this DSS benchmark.
- All the Oracle files, except TEMP tablespace, were mirrored
under SVM. Eight 6180 arrays (128 disks) were mirrored to another 8
6180 arrays using 128-way stripe. IO performance was good and balanced
across all the disks with a round robin order. Read performance was the
same with mirror or without mirror. With the SVM mirror the benchmark
passed the ACID (Atomicity, Consistency, Isolation and Durablity)
- Oracle tables were 128-way partitioned and parallel degree for
each table was set to 128 because the system had 128 cores.
This setting worked the best for performance.
- CPU usage during the Power run was not so high. This is because parallel degree was set to 128 for the tables and indexes so it utilized 128 vcpus for the most of the queries but the system had 256 vcpus.
- vdbench tool
- Transaction Processing Performance Council (TPC) Home Page
- Ideas International Benchmark Page
Sun SPARC Enterprise M9000 198,907.5 QphH@3000GB, $15.27/QphH@3000GB, avail 12/09/10, IBM Power 595 QphH@3000GB, 156,537.3 QphH@3000GB, $20.60/QphH@3000GB, avail 11/24/09, HP Integrity Superdome 60,359.3 QphH@3000GB, $32.60/QphH@3000GB avail 06/18/07, TPC-H, QphH, $/QphH tm of Transaction Processing Performance Council (TPC). More info www.tpc.org.